I remember there were many chipsets which were either rather warm or rather temperature sensitive in mid 2000s thereabouts. Cheap boards used fans because a little noisy fan is so cheap. More expensive boards threw a larger metal area on there on the same chips.
They want you to buy a new board after the 2 year warranty or whatever when the small fan fails. I cant find another reason of bringing back the small shitty fans...
I think the 2 main reasons against small finned heatsinks are (from AIBs perspective): 1. Where do we put our giant logo? 2. How do we install RGB on it? So Its technically impossible to cool a chipset like this ;)
exactly, they want to keep the fashion accessories so badly, they would rather slap a potential point of failure (a fan) on a lump of metal, rather than put actual cooling fins on the thing.
Gigabyte already proved that the cooling fan is unnecessary, and that the logo can simply be moved upward, actually improving visibility as it is now above the PCIE slots, although a part of me still kind of wishes that they brought over the copious amounts of RGB they had on the X470 Gaming 7 WiFi. It is just such a shame they couldn't bring the large heatsinks down to their lower end X570 boards, so I guess shame on all board manufacturers, Gigabyte included.
I remember hearing some guy in the youtube tech sphere talking about how the motherboard manufacturers were saying NVME RAID with PCI-E 4 SSDs was putting much more stress on the chipset which is why they all went for active cooling on them. Might be worth testing that out and seeing if your generic heatsink would be enough to cool that too.
3 Theories: 1. The 4.0 link to the CPU uses a lot of power 2. There is a FW/SW bug that prevents power saving on unused ports 3. x570 is a repurposed IO die, and some of the unused silicon is not power gated off due to a FW or silicon issue
IMO it's probably because of the chipset needing to push 4.0 signals to the CPU. Perhaps the power reduces if you put a Ryzen 2000 CPU in X570 since it will not support 4.0 signaling between the chipset and the CPU? Or maybe it's just that the X570 chipset is not entirely a purpose built chipset and more like re-using an already existing design from the CPU I/O die?
That's my understanding too. As far as I can tell the X570 "chipset" is just a 14nm version IOC of the IOC in the Zen 2 chip. It uses the 2 infinity fabric links as PCIe 4.0 multiplexers because each infinity fabric link is basically just a PCIe x32 link at the hardware level. So all the 24 PCIe 4.0 lanes are routed to the chipset before going to the NVMe drives and PCIe lanes. That's probably the reason for the high idle consumption of the chipset with a Zen 2 CPU installed. An easy test would be to measure again with a Zen or Zen+ CPU instead an see what the power consumption would be in that case.
@Wasted waster They're still routed through the chipset. All 24 lanes. And can be multiplexed into a maximum of 40 lanes (typically less). There is a genuine need to release PCIe 4.0 this generation. PCIe 4.0 NVME drives was just around the corner, AMD helped getting an early version NMVe controller out the door early but by CES 2020 most other controller makers would announce their version as well. AMD knew this and prepared to have the option in their current products instead of potentially having to wait 6-8 months to put the feature into their platform, and letting Intel get a possible win. The entire Zen 2 launch is a showcase in how to completely destroy your competition in every field. That's reason enough to include PCIe 4.0. And like some reviews have mentioned the PCIe 4.0 being on the 5000-series cards, while not important for regular gaming, is extremely important for encoding and decoding for example streaming where a 5700XT beats the 2080ti, not the 2080, not the 1080ti, no the 2080ti.
@Der8auer - Don't know if you saw it, but Ian Cuttress had mentioned that AMD used the I/O die as the chipset, which may be built on 14nm and they used the I/O die built on 12nm GF for the die on the CPU supposedly for better memory support/overclock. They originally planned on a full 8 lanes of infinity fabric between the CPU and the chipset, later cutting it down to 4 lanes to lower the power consumption from the rumored 15W to 11W. So, you cut off the memory channels on the chipset I/O variant, along with other features, but you have to power IF, which seems to be power hungry, allegedly. So part could be using IF to connect chipset to the socket (part 1 of the theory). This would allege that there is 4W solely needed to drive 4 lanes of IF though. Part 2 of the theory is that, because it is an I/O die as a chipset, it was not made with the power saving features a chipset designed as a chipset from the ground up would have. So higher idle power consumption. Because of this, I'd love if you could dive into the power consumption of the I/O die on the CPU to compare the values for what devices are connected and power draw, while then having to figure out how much of that is going to memory, inefficiencies of 14nm versus 12nm GF process, etc. So understanding the I/O die on the CPU may help to further understand the I/O die as a chipset, including power consumption. Just a thought.
One thing to keep in mind, there are entire CPU built on 14nm that draw less power at idle and just barely more at load *with* the memory controller and IGP built in. AMD didn't enable any power management to speak of. It could be that caused issues with Infinity Fabric, latency was too high when ramping up to normal clocks, etc...
I believe you are correct that the IO and chipsets were manufactured on 12 and 14 nm but I think you have them backwards as most sources are quoting the CPUs as being 7nm/14nm not 7nm/12nm
X570 is also 12nm. It's the EXACT same chip that's on the CPU's. There is no 14nm "small" (mainstream) I/O die. That silicon simply does not exist. Your "theory" was mostly right though, as seen with TRX40 (which uses the same chipset silicon). Z490 (AMD's version) was originally on the roadmaps as an AM4 platform with the fully enabled chipset, just like on Threadripper, but it was canned for cost & product segmentation reasons (X570 was already at enthusiast pricing, and people who needed more I/O than it provided would be better served by Threadripper anyways).
My theory: They knew X570 boards drew way more power so, they just lied and told everyone it was Gen4 tech that was causing it. Not thinking that someone would be curious enough to look into where the extra power consumption is coming from. Thank you for there videos, the community appreciates it!
Worst case scenario : When northbridge chipset very hot system become hang, freeze then shutdown itself. isn't about power consumption but temperature in closed PC case
It's the 2x wider PCIe 4.0 link to the CPU (which is a constant, and it's actually an 8x lane bus on die as seen with TRX40 [which uses the same chip], but with only 1/2 wired up [but the entire bus still needs power to work]) that's sucking up the power, so they technically weren't wrong at all to say that PCIe 4.0 was the cause; just not in the way most people had thought/assumed. Also, the M.2 slot der8aur is using here is the main one.... aka the one routed directly to the CPU, not to the chipset. If he'd have used a different M.2 slot, I imagine the power consumption figures would have been higher.
IDK. It's been confusing for me. Two NVME drives running directly off cpu lanes or cpu lanes to the PCIe sockets don't even go through the PCH. Raid on sata III goes through the PCH. Mobo's with that M.2 socket just a few inches away from the cpu will maintain signal integrity in gen4 x4. I can't understand why X570 with limited cpu lanes runs so many of them to the PCH anyway. How many USB, SATA III and USB gen2 sockets do they think people want. Just like an X370/470 boards we're left with 16X or 8X 8X for PCIe sockets. I just found on Corsair's web that a gen4 nvme will run on a gen3 socket. I think this NDA stuff hurt AMD. At launch it was spend a ton of money on stuff that's a secret. On Asus site I can't tell if a new bios released the 5th of july will boot Ryzen 3000 chips or not. All the cpu compatability charts still stop at 2000 series cpus. Please begin every sentence with the term "I think?" fFletchs, best wishes.
Very good video, I really enjoy these in depth analysis. You should try pci-e SSDs in Raid, that should load up the chipset even more. The fact that this chipset uses more power than some of the ultrabook cpus is quite amusing, add to that the probably not very good reliability of chipset fans and maybe X470 could be a better choice than X570 if you only loose the pcie4.0 capability.
Currently no doubt X470 is a better platform if you are going for energy efficiency, also you basically need to mod the board with a passive cooler in order to get a silent PC. All in all this is a bad job on the board partners, at the prices these are sold for they could have made a better passive cooler, probably would cost not cost a lot more.
Circuit designer here one thing that could explain the power consumption is that to build circuits like serial links at higher speeds often requires added complexity or entire architectural changes. Designers try to make power scale back at lower speeds so that power consumption there is still competitive but there is always some overhead and if you've jumped a significant architectural boundary (ie. adding amplifier stages or samplers) it will be difficult or impossible to achieve the power that a more application-specific circuit can achieve at lower speeds.
They have to make sure the chipset keep under some temperature threshold in a hot closed case (open air testing is irrelevant) with the chipset cooling obstructed by expansion cards. In a situation with high ambient temperature and negligible airflow due to something like 2 large graphics cards above (or worse, some small hot airflow because of these), 9 W is quite a lot.
9 watts is not a lot. Go back and look at some beautiful X58 or so boards. The X58 chipset had, what, a TDP of 24.1W? Back then we could cool those just fine *passively*, and this at a time when running quad-SLI was a still a thing.
@@osgrov yeah but that thing was not obstructed by the GPUs at all, so had airflow from the overal airflow in the case. Completely different situation so incomparable
Possibly an unfinished design. They may have been rushed. Just saying, it's possible that the engineers pulled a rabbit out of their hat to get it working at all in the time allocated, not poor design just not had the time to finish. That kind of crap happens when you pick the release date before the product is finished.
In the early 2000's we were having severe electrical issues causing a loss of power in a car engine. Anyway, the mechanic's figured out the coil pack was putting a charge in the engine coolant system. The answer was to test the coolant for an excessive electrical charge . ( I don't remember the voltage the manufacture came up with to be excessive.) The solution was to change the coolant if it exceeded the limit. If you can stick tester lead in your coolant to find out if it's charged. But, if it is, i don't have a clue on how fix it. It's the only place i can think where the missing electricity would go if you've already removed all the extra fans from the board.
@@mafcarvalho Unfortunately there was no multi nvme testing done but you can see the added sata devices don't increase its power consumption much. Essentially it looks like an old cpu - burning 100% power all the time with no idle states and the only thing that changed its power consumption was the state of the workload it was performing.
Remember the X570 chipsets are made and designed by Global Foundries and AMD not ASmedia like the previous ones, so i guess they suck at optimizing power consumption
...ooor, it was designed by AMD as a universal I/O die because it's the 14nm version of what's found in the Ryzen 3000 series CPUs(in 12nm form). And, the whole point of doing this is that AMD is contractually obligated to support Global Foundries with a minimum order of silicone each year, and, since AMD has moved Ryzen onto 7nm and GloFo doesn't have an operational 7nm line right now, leading AMD to move CPU production to TSMC, they needed to offer GloFo something that they can manufacture which AMD needs to be able to make a profit on to make it worth it; and, seeing as how motherboards need a chipset to control the I/O functions of a motherboard, "might as well..."
GloFo had absolutely no involvement with AMD's I/O die designs outside from being the place where they are made. AMD did the designs for both I/O chips COMPLETELY in-house.
Not really. He just showed on this video that adding PCIe 4.0 devices barely make the power usage change at all. Energizing 4x PCIe 4.0 lanes did almost nothing at all. If PCIe 4 was inherently power hungry this would not be the case. It's just an excuse, PCIe 1.0, 2.0 and 3.0 all use about the same power, same as 802.11a/b/g/ and 802.11n/ac/ax. There is a minor change to the physical layer and a major change to the signalling/modulation.
It's far from the first time chipsets have had fans. I'm glad that isn't the norm any more, but I've got plenty of old boards with chipset fans. On the other hand, the heat sinks depend on the CPU and case fans to keep the chipset cool but if you're running water cooling then the chipset needs a fan or it's own cooling.
In theory b550 will be a good option. Any PCI-e slots for GPU and the primary m.2 slot should still be gen 4 (direct to CPU) and any PCI-e slots attached to the PCH will remain at gen 3. This means your primary boot drive can still be running at 5GBps and your Radeon GPU can have it's placebo PCI
G'day (Schönen Tag) Roman, Thanks for doing this video so we can understand the real power usage figures, I have great respect for people who do testing to find the real answer, as opposed to others who are getting their facts from thinking up an answer or just Googling Rumour Videos, Almost 2 weeks of testing to get these results & then Filming twice for German & English versions, I don't speak German but I like putting on the German version to listen to when typing my comments as I think you deserve the extra effort from me as a viewer considering the extra effort you put in for us doing an English version. P.s. I find it interesting It takes 18 minutes to say everything in English but 25 minutes to say it in German Danke noch einmal (Ja, ich habe betrogen, indem ich es gegoogelt habe) 😁
No idea but if your using the reference CPU cooler and a GPU dumping heat into a mid size case that tiny heat sink at 74c could get a lot hotter. Then again its not like people are going to be playing intensive games and running Crystal mark bench in the background either.
It looks to me like the power draw drop going to 7nm is reversed with gen4 and more cores. The VRM on 7nm cpus is an increase not a "same oh same oh" found on X470 boards, and the PCH chip set has turned into a boiler. I think I need an open loop water cooler and a monoblock that covers the entire board. If your going air cooled you better get a case with 2 200MM fans on the front. Make that 2 230mm fans.
Completely agree with you... 2019 and I also don't understand why brands keep pushing "gaming" shenanigans. I want a simple black PCB, with no gaming logos or inscriptions, just a minimalistic industrial design theme, apple style if you like to call it that way, with good thermals.
Tip for future micro soldering, use flux. It works wonders and you can never use too much of it. Also look into an air rework station for removing smds. Much safer than direct heat from an iron.
Fantastic video, I love stuff like this. One thing you failed to mention is that these new PCIe4.0 drives run at the same speed on X470 boards. Or at least, they do on the Crosshair Hero VII. I realise it wasn't the purpose of the video, but it further compounds the question mark over X570.. in terms of extra power, extra noise, dodgy fans that aren't required and the only advantage is something that you can get on X470. I expect that in time newer SSD's might be able to take advantage of PCIe4.0 but at this stage.. they haven't. Same thing with video cards.
Back in the days of P965, P35 and P45 chipsets, we had copper finned heatsinks on the VRM, northbridge and southbridge with heatpipes linking all those heatsinks together. I wish to see them come back though. It looks cool
Tin, how on earth does it "sip" so much power? I can't even fathom that "filling a cable with a bunch of moving electrons in case you need to send a signal" would cost the TDP of an entire laptop.
its not a purpose built chipset. In order to save development cost (or asmedia just wasnt ready) amd reused the IO die as a motherboard chipset. Lots of functions end up not being used (dram controller, IF links) but probably still use some power.
Awesome job. That heatsink is exactly what I was planning on doing, but sadly the only mATX board didn't have alc1220. P45 chipset was 18W and that was passively cooled, so had no doubt this could too. X570 heatsinks are just piece of metal with hardly any fins and horrible fan slapped on top, it's like they didn't even try.
My guess would be the x570 isn't ideal because it wasn't designed to be a PCH it's was an Zen 2 io die. AMD reworked the io die and turned it into a PCH. It probably isn't as efficient is making a dedicated chipset controller.
You know, if you're going to the trouble of soldering, you can just use the ammeter functionality of your probe. Probes basically have current clamps inside them, and the speed of measurement will be about the same if you're using the voltmeter functionality to measure the current from an external current clamp.
Thanks for investigating the X570 chipset power consumption Roman. I've thought that to... Why not just have a larger passive heatsink with more surface area? I dread using a small dinky little fan like that. If the fans were made by Noctua or Be Quiet! then that would be something. But yes, a fan is not necessary if the passive cooling is done right.
Thank you for a great and insightful video. Wow, the amount of underlying knowledge and skill before being able to do what you did in there. Lovely to see a nerd on a component level (with awesome soldering skills).
This is from a reddit thread: "The primary reasons are packaging and signal integrity. It's better for the modern chipset to be closer to the PCI-e slots because PCI-e signals are more susceptible to interference than old PCI or AGP signals. The result is that the chipset ends up in an area where a video card might extend. In order to prevent interference with video cards, the chipset heatsink needs to be low profile - hence the fans." Could this be the reason for the fans?
just remember it was still hitting 70C which is hot enough to crash Phenom/FX cpus if you tried to run that hot. if it's gonna run that hot i don't mind the fan.
the coolers on the boards are the main deterrent for me getting a 570 board. i have half a mind to wait for the 550 variants just to get something that does neither add noise nor something that can mechanically break like fans tend to do.
As for power consumption of this chipset - Ryzen v1000 is an embedded cpu that has 32 pcie lanes. If Ryzen 3000 has a v1000 replacement in the series then the io die has 32 pcie lanes out . We can see from the existence of the 12 and 16 core that it supports two infinity fabric links internally. Ryzen 1000 and 2000 series has a multi-purpose piece of silicon that can be a infinity fabric link or a pcie nexus or sata. I expect this is the same in Ryzen 3000. A pcie nexus is 16 lanes of pcie. So this io die has four of these multi-purpose io links. Which means this chip has the equivalent of 64 pcie lanes in it. If power gating is not being used correctly then it has 64 lanes all light up at idle. This would explain why adding devices does so little over idle power consumption.
2 things just to be sure : Is that ..."overwattage " safe for the mobo ... I mean is two times regular chipset consumption ... Could that stuff reduce the lifespawn of the mobo? Do you still recommend buy a x570 mobo instead a x470 with good phases (like taichi) to pair with the 12-16 core processor?
Did you test the gen 3 drive in crystal disk mark when attached to X570? If so was the performence any better than when attached to X470, Just wondering if the extra power consumtion could be related to better performence owing to higher throughput on X570, even with older devices.
maybe they just want them to look more premium and high end. they do look cool, even though it's adding a potential failure point. noise is probably negligible unless you're going for a completely silent build, even then I'm not sure you'd hear it through a case. given the size of the heatsinks on x570 you could just disable the fan, or replace the heatsink solution if you have a problem with it.
For most mainstream consumers, I think you're right. Still, since I don't have a 470 board right now anyway, I'm willing to pony up an extra $100 or so just to experience raid zero nvme's & 10,000MB/s general storage in 2019. And yes, I'll find a reason :-)
Hi. Love your in depth technical analysis. Also, thank you for calling out on the non requirement of fans for the chipset. I had a feeling those were unnecessary months ago. Thank you for confirming that. I have a very simple question. Why are motherboard manufacturers releasing all their boards with such insane power delivery? I get it that a MSI Godlike or an ROG Crosshair or Aorus Extreme has to live up to its name and price and have at least 12 to 100(pardon my sarcasm) phase designs. But why do the normal boards, sub 200$ boards also need to have these? It has been shown that a 12 core 3900X draws less power than a 2700X under full stress load. It hardly draws more that 150W under extreme circumsatances. An 8 phase proper power delivery would be sufficient. Thus, it is perfectly possible that a good 200$ board could run it. But bacause of this image that AMD and the board manufacturers and the general public has created, all X570 boards are expensive because of high end power delivery systems. Also if possible, please make a video showing that it is perfectly fine to run the 12 core 3900X from a good X370 or B450 board, let alone an X470 as I believe those are perfectly capable of running the new chips upto the 12 core cos of the less power draw. I know the board manufacturers are making all the boards future proof for the 16 core 3950X, but it would be good for AMD if the media also portrayed that and not make it look like Ryzen 3000 needs a large amount of phases and high end boards. Really, someone like you saying that an X370 or proper B450 like the Tomahawk is more than enough for the 3900X or the 3700X respectively would make a lot of this confusion clear. Sorry for the long post. Thanks. :)
Using: 2x PCIE3.0 chipset bound NVME in RAID0+ 1x PCIE3.0 CPU bound NVME unraided+ 2x SATA 7200RPM HDDs in RAID0+ 1x SATA 7200RPM HDD unraided. My Aorus master reaches 75~82C easily with the chipset fan set to performance in a high airflow environment. The videocard is a 3 slot 2080 Super that goes right above the chipset. You should really test those extreme cenarios for x570 temperature and consumption, really overload it with RAID0 and fill all NVME slots plus use some SATA RAID0 together as well. That would be a great video.
History coming full circle? Chip set fans in 2019 o.O Can't wait till they re introduce Delta fans to the rest of our rigs, Might be time to buy shares in hearing aid companies. /s
There are a few things you can do to saturate the PCIe 4 lanes better, such as running NVMe RAID0 off a PCIe slot, or running some compute loads through the GPU with a PCIe 4 GPU. You barely scratched the surface of what PCIe 4 can do by testing just a single NVMe drive. I would surmise that if you tried to do more of a content creators style build for example, with say two 5700XT's (for video or 3D rendering) and 4 NVMe drives off the thrid PCIe slot, or even just three in the slots on the mobo, you would get that power consumption up enough to need a fan.
Current consumption might fluctuate very fast, which the multimeter won't even register. I also doubt the clamp is suitable for this kind of measurement. Maybe hook the clamp up to an oscilloscope instead of a dmm.
I think the reason why it has higher power is due to running a few pcie lanes via the pch. More than normal to bump the total lanes more than the competition.
I've heard somewhere that i/o-die and x570 chipset die are the same and when I compared size (on photos) it looked to be about the same size. So if they reuse i/o-die as chipset it makes complete sense that it requires more power because it's multi purpose.
That's the conclusion I came to. The fact that this isn't designed to be a PCH controller it was designed to be an IO die for a CPU, the power requirements might be higher.
Eh, a few things here. You should just use the multimeters current mode when measuring these as you are already soldering wires on finally the inductor lifting was not the best idea, your simply adding to the switching inductors inductances and the current on them will be all over the place (at the switching frequency) and it probably should have been tapped off elsewhere or measured with an oscilloscope (but likely would not help either as i doubt the bandwidth of that clamp is above the switching frequency) it might be better to try and find that rail elsewhere, those polyfuses next to it look like a good bet
At least four review sites (including you) have noted that high idle voltage now, not even counting the chipset. For the chipset, I wouldn't have actually expected it to eat much power unless there were a ton of PCIe-4 devices plugged in. Sure, it will use 4 lanes of PCIe-4 between the CPU and the chipset, but that's just 4 lanes. What is going on, indeed! The chipset just shouldn't be producing enough heat to require that little fan to turn on, under any load, short of having a lot of PCIe-4 peripherals. So perhaps there is a voltage control issue. I can't think of a single reason for that X570 chipset to burn so much power either, other than being over-volted. Makes no sense to me either. I'm not entirely sure those resistors are 0-ohm bridges. Are you sure they aren't current sense resistors? i.e. very low but non-zero ohms? The voltage could easily read the same on your 3.5 digit multi-meter. However, if the SMD resistor has a single digit '0' sillkscreened onto it, then it *is* a zero-ohms bridge. -Matt
A reason for the fan on the chipset may be because of time constraints the chipset was the main bottleneck for x570's release, so I can imagine a scenario where the motherboard manufacturers suddenly had to deal with a 15w rated chipset. Being unable to produce and most importantly validate a passive heatsink in time. As for testing the power consumption: you can still test usb as well to see if t hu at may increase it
Isn't the inter-connect between the CPU and chipset running PCI-Express 4.0? So regardless of what you connect you will still have increased power consumption due to the inter-connect between the CPU and the chipset? As AMD Ryzen 2000-series does not feature support for PCI-Express 4.0, what happens if you replace the Ryzen 3000-serie CPU with a Ryzen 2000-series CPU? Do you notice and difference in power consumption? Regardless of all of this, it seems like the cooling is built to be overkill on purpose. Barely anyone will ever reach and wattage that will actually require active cooling. Hopefully you can disable the fan in the UEFI firmware, or at least have it not run unless you hit high temps.
You are spot on!!..I test on my entry level MSI x570 GAming Plus, with 0 rpm or 1100rpm the temp are the same, 70º, and this after 1,3H streaming and gaming the same time!!...So i think some fishing about the reason why they put active cooling on chipset...
In my experience from the past (both amd nor Intel) when northbridge chipset very hot system become hang or freeze then shutdown byself. In the worst scenario case temp, airflow or ambient too hot small fan it can help you. first I didn't understand what was wrong with my old system ! struggle few months until I finally realized northbride was very hot and decided to reduce northbrige voltage that's work but reducing overclocking rate too. Edit; My temperature CPU under full load 70~77℃ isn't too hot. LOL 😂
What are safe temps for the sb on the chipsets? Trying to figure out what fan curve to make and if i'll instead need to get a heatsink like you did. Also what are the dimensions of the heatsink and how did you attach it? I see some heatsinks can be attached just by sticking.
The cIOD on Matisse works on PCIe 4.0 on all boards including X570/400/300 Series. That enables PCIe4 NVMe support on 400/300 Series boards. The chipset uses the same design. Maybe the chipset runs on PCIe 4 all the time.
Great video. I have a small test in mind that you didn’t do. You used PCIe Gen 4, yes, but only one device and only 4 lanes. I would try the gen 4 nvme + a Radeon 5700(or XT), which is a total of 20 Gen 4 PCIe Lanes. Maybe with all the lanes occupied it makes a difference?
Many thanks for testing this... and Yep. This is very trange... have you been asking mother board vendors hard questions after testing this? Pci 4.0 need better wires, but that should use even less electricity because thicker wires should be better conducts (in reality there is no difference unles you put heck a lot current in there) Maybe They just use higher voltage, like when overclocking memory etc. And that is why effiency is thrown out of window in here. Aka a Little bit more stability, by using really high voltage and thus using a lot of energy for minimal benefit? That is my ques...
Yeah. Companies Are very good at making things last as Little as possible, while over the warrantry time. Maybe companies Are not happy because amd motherboards can be used so long time and just upgrade cpu and gpu when needed. Now They kill the motherboards as soon as possible ;) my foliohat is too tight!
it is the line connection between the CPU and Chipset. 4.0 has some power for the 3950x and above to handle and the Connection made above 70 degree in AMD inside tests. So they desided to be sure the new MB can handle upon Ryzen 4th gen, which will be much more power stressing the new MB's. It's for the "we guarantee you 3 years on same mobo" Thing as it's still thought to be up/down ward compability.
When you add the pcie graphics card to the mix you should had stressed tested with some benchmark that really stressed the pcie bandwidth I think Sisoftware Sandra had a module test for that. Furmark doesn't bottleneck half a pcie1x. Or 2 or 3 nvme drives, just stress then witn file copy from 1 to another. Just some ideas that might work. Good luck.
Great video, just the info I was looking for. Wanting to replace the monstrosity of a 40mm fan PCH cooler on my x570 board with a passive cooler and wasn't sure it could handle the job. Looks like it should work just fine especially being right in front of the case fan. I'm using a compact video card so I have all the room i need for it. Funny thing I have a couple Zalman passive northbridge coolers from like 2005 I almost threw away thinking I'd never have use for them.
so for average person this means, if the x570 mb chipset fan fails it does not mean that the chipset will fry ? it still has to have some sort of heatsink for the fan to cool right ?
Wonder if they aren't having production issues making these new x570 chipsets, and are essentially having to overvolt them to get the clocks they need. I wonder if there's anything hinting at this? Is the main 1.075V supply now a higher voltage than the old x470? As for it idling high, it might just be that they have to run the whole thing at full speed even if there aren't pcie 4 devices present. Hmm. Not great tho. One option for measuring current could be to use very low value shunt resistors (like 0.01 ohm, available in smd) and measuring the voltage across them on the board, maybe even with a scope. Maybe a little easier than large loops of wire.
Looking at the idle difference between x470 chipset and x570 and nvme gen3-4 difference points at x570 chipset just being crap, much worst designed than x470 ones. Great video analysis!!!
What's the total resistance of those wires ? More resistant less power right ? or the PSU compensates ? Also what's the point of this video? Why the chipset PC is important ? Like those RGB LEDs pull more power than the chipset. EDIT: okey i wrote before i watched that last 3 minutes. I love the GB P67 UD7 mobo, but i guess that heatsink design can't be applied to the new mobos becase of the nvme slot right under the CPU socket.
Someone must have had a warehouse full of 40mm fans they needed to unload.
I remember there were many chipsets which were either rather warm or rather temperature sensitive in mid 2000s thereabouts.
Cheap boards used fans because a little noisy fan is so cheap.
More expensive boards threw a larger metal area on there on the same chips.
It's the 40mm fan mafia. They'll give you an offer you cannot refuse. Or you'll find thermal paste in your CPU socket.
@@SianaGearz yes, i imagine it's something to do with price - that it's cheaper to use fans.
They want you to buy a new board after the 2 year warranty or whatever when the small fan fails. I cant find another reason of bringing back the small shitty fans...
Delboy and Rodney finally flipped these fans.. lovely jobly...
I think the 2 main reasons against small finned heatsinks are (from AIBs perspective):
1. Where do we put our giant logo?
2. How do we install RGB on it?
So Its technically impossible to cool a chipset like this ;)
exactly, they want to keep the fashion accessories so badly, they would rather slap a potential point of failure (a fan) on a lump of metal, rather than put actual cooling fins on the thing.
Gigabyte already proved that the cooling fan is unnecessary, and that the logo can simply be moved upward, actually improving visibility as it is now above the PCIE slots, although a part of me still kind of wishes that they brought over the copious amounts of RGB they had on the X470 Gaming 7 WiFi. It is just such a shame they couldn't bring the large heatsinks down to their lower end X570 boards, so I guess shame on all board manufacturers, Gigabyte included.
@@thetrashman5252 Too bad they didn't make their ITX motherboard on X570 passively cooled when it came to the chipset.
Dont u know rgb gives u 20c lower temps
@@bryansuh1985 Really, my CLC 280 has just a single RGB LED and I can confirm that it drops idle temps from 250C to 25C.
I remember hearing some guy in the youtube tech sphere talking about how the motherboard manufacturers were saying NVME RAID with PCI-E 4 SSDs was putting much more stress on the chipset which is why they all went for active cooling on them. Might be worth testing that out and seeing if your generic heatsink would be enough to cool that too.
yes saw some one say that with one of the PCI cards that takes 4 NVME drives to raid them i think
It draws 2.5 times more power than X470 even in idle.
Sooner or later MOBO has to be a active buy like CPU, GPU, RAM and PSU... You need x Mobo to be compatible,etc, and have feature performance.
3 Theories:
1. The 4.0 link to the CPU uses a lot of power
2. There is a FW/SW bug that prevents power saving on unused ports
3. x570 is a repurposed IO die, and some of the unused silicon is not power gated off due to a FW or silicon issue
I think # 1 is spot on, although #3 could be compounding the issue as well.
From what I've read, 3 is the case
@@scott2100 At a guess, I would say 3.
Option 3 is bang on imo
#1 should be testable. I know on the Asus boards you can drop the CPU to X570 link down to 3.0 as a config option.
40mm fan selling mafia wants to know your location.
Hope these are not huawei they shut down our Mobo cooling at the same time as our 5g. Then the dragon strikes. 🤪
IMO it's probably because of the chipset needing to push 4.0 signals to the CPU. Perhaps the power reduces if you put a Ryzen 2000 CPU in X570 since it will not support 4.0 signaling between the chipset and the CPU? Or maybe it's just that the X570 chipset is not entirely a purpose built chipset and more like re-using an already existing design from the CPU I/O die?
i think its cause its a used IO die. probably unused stuff using power. x470 was a purpose built asmedia chipset
Afaik, they are using the I/O die because Asmedia has been having issues getting theirs ready which is what will be used for B550.
Nothing to do with 4.0 signalling. Going down the wrong path here.
That's my understanding too. As far as I can tell the X570 "chipset" is just a 14nm version IOC of the IOC in the Zen 2 chip. It uses the 2 infinity fabric links as PCIe 4.0 multiplexers because each infinity fabric link is basically just a PCIe x32 link at the hardware level. So all the 24 PCIe 4.0 lanes are routed to the chipset before going to the NVMe drives and PCIe lanes. That's probably the reason for the high idle consumption of the chipset with a Zen 2 CPU installed. An easy test would be to measure again with a Zen or Zen+ CPU instead an see what the power consumption would be in that case.
@Wasted waster They're still routed through the chipset. All 24 lanes. And can be multiplexed into a maximum of 40 lanes (typically less).
There is a genuine need to release PCIe 4.0 this generation. PCIe 4.0 NVME drives was just around the corner, AMD helped getting an early version NMVe controller out the door early but by CES 2020 most other controller makers would announce their version as well. AMD knew this and prepared to have the option in their current products instead of potentially having to wait 6-8 months to put the feature into their platform, and letting Intel get a possible win.
The entire Zen 2 launch is a showcase in how to completely destroy your competition in every field. That's reason enough to include PCIe 4.0.
And like some reviews have mentioned the PCIe 4.0 being on the 5000-series cards, while not important for regular gaming, is extremely important for encoding and decoding for example streaming where a 5700XT beats the 2080ti, not the 2080, not the 1080ti, no the 2080ti.
@Der8auer - Don't know if you saw it, but Ian Cuttress had mentioned that AMD used the I/O die as the chipset, which may be built on 14nm and they used the I/O die built on 12nm GF for the die on the CPU supposedly for better memory support/overclock.
They originally planned on a full 8 lanes of infinity fabric between the CPU and the chipset, later cutting it down to 4 lanes to lower the power consumption from the rumored 15W to 11W. So, you cut off the memory channels on the chipset I/O variant, along with other features, but you have to power IF, which seems to be power hungry, allegedly. So part could be using IF to connect chipset to the socket (part 1 of the theory). This would allege that there is 4W solely needed to drive 4 lanes of IF though.
Part 2 of the theory is that, because it is an I/O die as a chipset, it was not made with the power saving features a chipset designed as a chipset from the ground up would have. So higher idle power consumption.
Because of this, I'd love if you could dive into the power consumption of the I/O die on the CPU to compare the values for what devices are connected and power draw, while then having to figure out how much of that is going to memory, inefficiencies of 14nm versus 12nm GF process, etc.
So understanding the I/O die on the CPU may help to further understand the I/O die as a chipset, including power consumption. Just a thought.
One thing to keep in mind, there are entire CPU built on 14nm that draw less power at idle and just barely more at load *with* the memory controller and IGP built in.
AMD didn't enable any power management to speak of.
It could be that caused issues with Infinity Fabric, latency was too high when ramping up to normal clocks, etc...
I believe you are correct that the IO and chipsets were manufactured on 12 and 14 nm but I think you have them backwards as most sources are quoting the CPUs as being 7nm/14nm not 7nm/12nm
@@bappyplays No. The CPUs use a 12nm IO die. Dr Cutress "triple checked" with AMD.
X570 is also 12nm. It's the EXACT same chip that's on the CPU's. There is no 14nm "small" (mainstream) I/O die. That silicon simply does not exist. Your "theory" was mostly right though, as seen with TRX40 (which uses the same chipset silicon). Z490 (AMD's version) was originally on the roadmaps as an AM4 platform with the fully enabled chipset, just like on Threadripper, but it was canned for cost & product segmentation reasons (X570 was already at enthusiast pricing, and people who needed more I/O than it provided would be better served by Threadripper anyways).
My theory: They knew X570 boards drew way more power so, they just lied and told everyone it was Gen4 tech that was causing it. Not thinking that someone would be curious enough to look into where the extra power consumption is coming from. Thank you for there videos, the community appreciates it!
Worst case scenario :
When northbridge chipset very hot system become hang, freeze then shutdown itself.
isn't about power consumption but temperature in closed PC case
It's the 2x wider PCIe 4.0 link to the CPU (which is a constant, and it's actually an 8x lane bus on die as seen with TRX40 [which uses the same chip], but with only 1/2 wired up [but the entire bus still needs power to work]) that's sucking up the power, so they technically weren't wrong at all to say that PCIe 4.0 was the cause; just not in the way most people had thought/assumed. Also, the M.2 slot der8aur is using here is the main one.... aka the one routed directly to the CPU, not to the chipset. If he'd have used a different M.2 slot, I imagine the power consumption figures would have been higher.
Man. I am so glad I found you by accident. Thank you for all of your work in showing us this.
I heard active cooling was needed if use nvme raid
nvme raid can be made only using 1x nvme + 1x pcie card
the RAID is handled by the CPU, not the chipset. that makes no sense.
@@steamstories1279 why all the pcie lanes, then?
@@solomonshv cpu handles some of the pcie lanes.
IDK. It's been confusing for me. Two NVME drives running directly off cpu lanes or cpu lanes to the PCIe sockets don't even go through the PCH. Raid on sata III goes through the PCH. Mobo's with that M.2 socket just a few inches away from the cpu will maintain signal integrity in gen4 x4. I can't understand why X570 with limited cpu lanes runs so many of them to the PCH anyway. How many USB, SATA III and USB gen2 sockets do they think people want. Just like an X370/470 boards we're left with 16X or 8X 8X for PCIe sockets. I just found on Corsair's web that a gen4 nvme will run on a gen3 socket. I think this NDA stuff hurt AMD. At launch it was spend a ton of money on stuff that's a secret. On Asus site I can't tell if a new bios released the 5th of july will boot Ryzen 3000 chips or not. All the cpu compatability charts still stop at 2000 series cpus. Please begin every sentence with the term "I think?" fFletchs, best wishes.
Very good video, I really enjoy these in depth analysis. You should try pci-e SSDs in Raid, that should load up the chipset even more. The fact that this chipset uses more power than some of the ultrabook cpus is quite amusing, add to that the probably not very good reliability of chipset fans and maybe X470 could be a better choice than X570 if you only loose the pcie4.0 capability.
Currently no doubt X470 is a better platform if you are going for energy efficiency, also you basically need to mod the board with a passive cooler in order to get a silent PC. All in all this is a bad job on the board partners, at the prices these are sold for they could have made a better passive cooler, probably would cost not cost a lot more.
Circuit designer here one thing that could explain the power consumption is that to build circuits like serial links at higher speeds often requires added complexity or entire architectural changes. Designers try to make power scale back at lower speeds so that power consumption there is still competitive but there is always some overhead and if you've jumped a significant architectural boundary (ie. adding amplifier stages or samplers) it will be difficult or impossible to achieve the power that a more application-specific circuit can achieve at lower speeds.
They have to make sure the chipset keep under some temperature threshold in a hot closed case (open air testing is irrelevant) with the chipset cooling obstructed by expansion cards. In a situation with high ambient temperature and negligible airflow due to something like 2 large graphics cards above (or worse, some small hot airflow because of these), 9 W is quite a lot.
9 watts is not a lot.
Go back and look at some beautiful X58 or so boards. The X58 chipset had, what, a TDP of 24.1W? Back then we could cool those just fine *passively*, and this at a time when running quad-SLI was a still a thing.
@@osgrov yeah but that thing was not obstructed by the GPUs at all, so had airflow from the overal airflow in the case. Completely different situation so incomparable
Test with a ryzen 2000 CPU in x570.
Perhaps it's the infinity fabric running at pcie4 that needs the higher power.
I hope nobody buys a X570 Board to pair it with Ryzen 2000 :D
@@der8auer It might be worth testing just to isolate if that's the reason for the power consumption though!
@@der8auer For 'Boints' they will
It's not PCIe 4. It's just very poor design.
Possibly an unfinished design.
They may have been rushed.
Just saying, it's possible that the engineers pulled a rabbit out of their hat to get it working at all in the time allocated, not poor design just not had the time to finish.
That kind of crap happens when you pick the release date before the product is finished.
Thanks for all the effort put into this video. Good research and hope you can find more information on the power consumption.
Really like your videos man, cheers.
In the early 2000's we were having severe electrical issues causing a loss of power in a car engine. Anyway, the mechanic's figured out the coil pack was putting a charge in the engine coolant system. The answer was to test the coolant for an excessive electrical charge . ( I don't remember the voltage the manufacture came up with to be excessive.) The solution was to change the coolant if it exceeded the limit. If you can stick tester lead in your coolant to find out if it's charged. But, if it is, i don't have a clue on how fix it. It's the only place i can think where the missing electricity would go if you've already removed all the extra fans from the board.
If I remember well that PCH will heat up only with a RAID of PCI-E 4.0 SSD
I remember hearing that too. 🤔
@@mafcarvalho Unfortunately there was no multi nvme testing done but you can see the added sata devices don't increase its power consumption much. Essentially it looks like an old cpu - burning 100% power all the time with no idle states and the only thing that changed its power consumption was the state of the workload it was performing.
Thats correct, They said chipset temps were an issue in some scenarios with multiple nvme drives in raid.
Two NVMe's in raid 0 can easily put this chipset to its maximum..
If you go Threadripper ZEN2 you can do that with hard wired cpu lanes that don't go through the chipset.
Remember the X570 chipsets are made and designed by Global Foundries and AMD not ASmedia like the previous ones, so i guess they suck at optimizing power consumption
...ooor, it was designed by AMD as a universal I/O die because it's the 14nm version of what's found in the Ryzen 3000 series CPUs(in 12nm form). And, the whole point of doing this is that AMD is contractually obligated to support Global Foundries with a minimum order of silicone each year, and, since AMD has moved Ryzen onto 7nm and GloFo doesn't have an operational 7nm line right now, leading AMD to move CPU production to TSMC, they needed to offer GloFo something that they can manufacture which AMD needs to be able to make a profit on to make it worth it; and, seeing as how motherboards need a chipset to control the I/O functions of a motherboard, "might as well..."
GloFo had absolutely no involvement with AMD's I/O die designs outside from being the place where they are made. AMD did the designs for both I/O chips COMPLETELY in-house.
It's like the chipset does all the hard work before you use the Gen 4 device
Any chance you could get a 2nd drive and test RAID 0 vs 1 power consumption.
Looking forward to buying cute little heatsinks for my chipset! It'll be just like the good old days.
The CPU is always taking to X570 via PCIe 4.0. It is always active and there is no such thing of idle, hence the high power consumption.
The CPU is always talking to the 470 over 3.0 and the power draw isn't that high.
Not really.
He just showed on this video that adding PCIe 4.0 devices barely make the power usage change at all. Energizing 4x PCIe 4.0 lanes did almost nothing at all.
If PCIe 4 was inherently power hungry this would not be the case.
It's just an excuse, PCIe 1.0, 2.0 and 3.0 all use about the same power, same as 802.11a/b/g/ and 802.11n/ac/ax. There is a minor change to the physical layer and a major change to the signalling/modulation.
X570 running secret backdoor bitcoin miner!
Possibly in the future a lower wattage chipset could be released. I am very wary of adding a motherboard with a fan to my rig.
B550
It's far from the first time chipsets have had fans.
I'm glad that isn't the norm any more, but I've got plenty of old boards with chipset fans.
On the other hand, the heat sinks depend on the CPU and case fans to keep the chipset cool but if you're running water cooling then the chipset needs a fan or it's own cooling.
In theory b550 will be a good option. Any PCI-e slots for GPU and the primary m.2 slot should still be gen 4 (direct to CPU) and any PCI-e slots attached to the PCH will remain at gen 3. This means your primary boot drive can still be running at 5GBps and your Radeon GPU can have it's placebo PCI
I want VRM fans like the super nice EVGA DARK motherboard. I add my own, but it’s often held on by zip ties or hot glue and not as pretty, haha!
put a x470
G'day (Schönen Tag) Roman,
Thanks for doing this video so we can understand the real power usage figures,
I have great respect for people who do testing to find the real answer, as opposed to others who are getting their facts from thinking up an answer or just Googling Rumour Videos,
Almost 2 weeks of testing to get these results & then Filming twice for German & English versions, I don't speak German but I like putting on the German version to listen to when typing my comments as I think you deserve the extra effort from me as a viewer considering the extra effort you put in for us doing an English version.
P.s. I find it interesting It takes 18 minutes to say everything in English but 25 minutes to say it in German
Danke noch einmal (Ja, ich habe betrogen, indem ich es gegoogelt habe) 😁
I think this video will never get the view it really deserves. Amazing job
No idea but if your using the reference CPU cooler and a GPU dumping heat into a mid size case that tiny heat sink at 74c could get a lot hotter. Then again its not like people are going to be playing intensive games and running Crystal mark bench in the background either.
It looks to me like the power draw drop going to 7nm is reversed with gen4 and more cores. The VRM on 7nm cpus is an increase not a "same oh same oh" found on X470 boards, and the PCH chip set has turned into a boiler. I think I need an open loop water cooler and a monoblock that covers the entire board. If your going air cooled you better get a case with 2 200MM fans on the front. Make that 2 230mm fans.
Completely agree with you... 2019 and I also don't understand why brands keep pushing "gaming" shenanigans. I want a simple black PCB, with no gaming logos or inscriptions, just a minimalistic industrial design theme, apple style if you like to call it that way, with good thermals.
Wow, that's a lot of work to not have an obvious cause. Sounds like a frustrating testing experience, but thanks for diving in.
these new motherboards are starting to look like some weird scifi alien technology..its quite mind blowing how much work fab goes into a board..
Tip for future micro soldering, use flux. It works wonders and you can never use too much of it. Also look into an air rework station for removing smds. Much safer than direct heat from an iron.
I love it when you leave the mistakes in. Great video. Keep 'em coming!
Fantastic video, I love stuff like this. One thing you failed to mention is that these new PCIe4.0 drives run at the same speed on X470 boards. Or at least, they do on the Crosshair Hero VII. I realise it wasn't the purpose of the video, but it further compounds the question mark over X570.. in terms of extra power, extra noise, dodgy fans that aren't required and the only advantage is something that you can get on X470. I expect that in time newer SSD's might be able to take advantage of PCIe4.0 but at this stage.. they haven't. Same thing with video cards.
My theory - There was a golf game with the executives at the fan factory now all x570 need a fan. 2nd theory - it's Trump's fault
3rd theory - thanks, Obama...
4th theory - Watergate was a distraction, this was Nixon's long game...
Your soldering skills are amazing. Great job!
Back in the days of P965, P35 and P45 chipsets, we had copper finned heatsinks on the VRM, northbridge and southbridge with heatpipes linking all those heatsinks together. I wish to see them come back though. It looks cool
You mean like the X570 Gigabyte Aorus xtreme?
TY for all of the hard work Roman !
6:00 the big resistor is probably a ferrite bead. Hard to tell
Heh, Gen4 Phy is there, sipping power, no matter if you use it or not. That's why :)
:(
Tin, how on earth does it "sip" so much power? I can't even fathom that "filling a cable with a bunch of moving electrons in case you need to send a signal" would cost the TDP of an entire laptop.
its not a purpose built chipset. In order to save development cost (or asmedia just wasnt ready) amd reused the IO die as a motherboard chipset. Lots of functions end up not being used (dram controller, IF links) but probably still use some power.
I don't think they repurpose the IO die. they just want to save some bucks by using the older node.
If they really did repurpose an IO die then surely they would have fused off all the unneeded circuitry.
Very interessting video.
Also, it's insane that you publish your videos in german and also in english, thank you a lot !
Awesome job. That heatsink is exactly what I was planning on doing, but sadly the only mATX board didn't have alc1220. P45 chipset was 18W and that was passively cooled, so had no doubt this could too. X570 heatsinks are just piece of metal with hardly any fins and horrible fan slapped on top, it's like they didn't even try.
My guess would be the x570 isn't ideal because it wasn't designed to be a PCH it's was an Zen 2 io die. AMD reworked the io die and turned it into a PCH. It probably isn't as efficient is making a dedicated chipset controller.
You know, if you're going to the trouble of soldering, you can just use the ammeter functionality of your probe. Probes basically have current clamps inside them, and the speed of measurement will be about the same if you're using the voltmeter functionality to measure the current from an external current clamp.
Still though, great video.
Sounds like you should create a product where you replace the cooling fans with custom heatsinks :)
Precisely!
Thanks for investigating the X570 chipset power consumption Roman. I've thought that to... Why not just have a larger passive heatsink with more surface area? I dread using a small dinky little fan like that. If the fans were made by Noctua or Be Quiet! then that would be something. But yes, a fan is not necessary if the passive cooling is done right.
Thank you for a great and insightful video. Wow, the amount of underlying knowledge and skill before being able to do what you did in there. Lovely to see a nerd on a component level (with awesome soldering skills).
This is from a reddit thread:
"The primary reasons are packaging and signal integrity. It's better for the modern chipset to be closer to the PCI-e slots because PCI-e signals are more susceptible to interference than old PCI or AGP signals. The result is that the chipset ends up in an area where a video card might extend. In order to prevent interference with video cards, the chipset heatsink needs to be low profile - hence the fans."
Could this be the reason for the fans?
remarkable work! that makes me feel a bit better if I buy a x570 without active fan shit cooler on top.
just remember it was still hitting 70C which is hot enough to crash Phenom/FX cpus if you tried to run that hot. if it's gonna run that hot i don't mind the fan.
@@XX-121 it was 70c with a cheap tiny heatsink... if you have a proper one like Gigabyte has in 1 model... it will not get 70c for sure.
You should really get the hot air station :)
useless in this case
I hope motherboards manufacturers will have WOW classic moment and discover cooling solutions from 2000's
the coolers on the boards are the main deterrent for me getting a 570 board. i have half a mind to wait for the 550 variants just to get something that does neither add noise nor something that can mechanically break like fans tend to do.
As for power consumption of this chipset - Ryzen v1000 is an embedded cpu that has 32 pcie lanes. If Ryzen 3000 has a v1000 replacement in the series then the io die has 32 pcie lanes out . We can see from the existence of the 12 and 16 core that it supports two infinity fabric links internally. Ryzen 1000 and 2000 series has a multi-purpose piece of silicon that can be a infinity fabric link or a pcie nexus or sata. I expect this is the same in Ryzen 3000. A pcie nexus is 16 lanes of pcie. So this io die has four of these multi-purpose io links. Which means this chip has the equivalent of 64 pcie lanes in it. If power gating is not being used correctly then it has 64 lanes all light up at idle. This would explain why adding devices does so little over idle power consumption.
Insane dude, you're absolutely insane. Nice work, that's some accurate soldering
2 things just to be sure : Is that ..."overwattage " safe for the mobo ... I mean is two times regular chipset consumption ... Could that stuff reduce the lifespawn of the mobo?
Do you still recommend buy a x570 mobo instead a x470 with good phases (like taichi) to pair with the 12-16 core processor?
Did you test the gen 3 drive in crystal disk mark when attached to X570? If so was the performence any better than when attached to X470, Just wondering if the extra power consumtion could be related to better performence owing to higher throughput on X570, even with older devices.
Here's something you might find interesting. The PCIe4.0 drives run at the same speed on X470 boards. ruclips.net/video/45fQaCl9WlA/видео.html
maybe they just want them to look more premium and high end. they do look cool, even though it's adding a potential failure point. noise is probably negligible unless you're going for a completely silent build, even then I'm not sure you'd hear it through a case. given the size of the heatsinks on x570 you could just disable the fan, or replace the heatsink solution if you have a problem with it.
Wow. Taking an iron to your brand new $400 motherboard just to do a single test... That's commitment! (you're braver that i would be ) :-)
I've seen a lot of guys going for x470 for their 3900x builds, i think x570 ones aren't worth it.
Can we have a pro/cons of x570 discussion on here?
For most mainstream consumers, I think you're right. Still, since I don't have a 470 board right now anyway, I'm willing to pony up an extra $100 or so just to experience raid zero nvme's & 10,000MB/s general storage in 2019. And yes, I'll find a reason :-)
3900x: In Far Cry 5 you get 10 more fps on x570 compared to x470 overclock3d.net/gfx/articles/2019/07/05152055474l.jpg
Hi. Love your in depth technical analysis. Also, thank you for calling out on the non requirement of fans for the chipset. I had a feeling those were unnecessary months ago. Thank you for confirming that. I have a very simple question. Why are motherboard manufacturers releasing all their boards with such insane power delivery? I get it that a MSI Godlike or an ROG Crosshair or Aorus Extreme has to live up to its name and price and have at least 12 to 100(pardon my sarcasm) phase designs. But why do the normal boards, sub 200$ boards also need to have these? It has been shown that a 12 core 3900X draws less power than a 2700X under full stress load. It hardly draws more that 150W under extreme circumsatances. An 8 phase proper power delivery would be sufficient. Thus, it is perfectly possible that a good 200$ board could run it. But bacause of this image that AMD and the board manufacturers and the general public has created, all X570 boards are expensive because of high end power delivery systems. Also if possible, please make a video showing that it is perfectly fine to run the 12 core 3900X from a good X370 or B450 board, let alone an X470 as I believe those are perfectly capable of running the new chips upto the 12 core cos of the less power draw. I know the board manufacturers are making all the boards future proof for the 16 core 3950X, but it would be good for AMD if the media also portrayed that and not make it look like Ryzen 3000 needs a large amount of phases and high end boards. Really, someone like you saying that an X370 or proper B450 like the Tomahawk is more than enough for the 3900X or the 3700X respectively would make a lot of this confusion clear.
Sorry for the long post. Thanks. :)
Using:
2x PCIE3.0 chipset bound NVME in RAID0+
1x PCIE3.0 CPU bound NVME unraided+
2x SATA 7200RPM HDDs in RAID0+
1x SATA 7200RPM HDD unraided.
My Aorus master reaches 75~82C easily with the chipset fan set to performance in a high airflow environment.
The videocard is a 3 slot 2080 Super that goes right above the chipset.
You should really test those extreme cenarios for x570 temperature and consumption, really overload it with RAID0 and fill all NVME slots plus use some SATA RAID0 together as well. That would be a great video.
good work on the test man. thanks
History coming full circle?
Chip set fans in 2019 o.O
Can't wait till they re introduce Delta fans to the rest of our rigs,
Might be time to buy shares in hearing aid companies.
/s
Hands 👐 down man. I am pretty sure mobo manufacturers don't know what's happening either 😂
There are a few things you can do to saturate the PCIe 4 lanes better, such as running NVMe RAID0 off a PCIe slot, or running some compute loads through the GPU with a PCIe 4 GPU. You barely scratched the surface of what PCIe 4 can do by testing just a single NVMe drive. I would surmise that if you tried to do more of a content creators style build for example, with say two 5700XT's (for video or 3D rendering) and 4 NVMe drives off the thrid PCIe slot, or even just three in the slots on the mobo, you would get that power consumption up enough to need a fan.
I am sure aesthetic and the fact that active cooling enables smaller heatsink footprint were part of the reasons.
Current consumption might fluctuate very fast, which the multimeter won't even register. I also doubt the clamp is suitable for this kind of measurement.
Maybe hook the clamp up to an oscilloscope instead of a dmm.
active cooling still wouldn't be needed if the heatsink is decently large and has enough thermal capacity
I think the reason why it has higher power is due to running a few pcie lanes via the pch. More than normal to bump the total lanes more than the competition.
I've heard somewhere that i/o-die and x570 chipset die are the same and when I compared size (on photos) it looked to be about the same size. So if they reuse i/o-die as chipset it makes complete sense that it requires more power because it's multi purpose.
That's the conclusion I came to. The fact that this isn't designed to be a PCH controller it was designed to be an IO die for a CPU, the power requirements might be higher.
Colab with GN on this I'm curious for more info. Maybe future firmware and bios can improve power consumption especially with the idle load.
Please don't collab with Gamers Nexus, when that guy makes a mistake, he rarely admits it.
Eh, a few things here. You should just use the multimeters current mode when measuring these as you are already soldering wires on
finally the inductor lifting was not the best idea, your simply adding to the switching inductors inductances and the current on them will be all over the place (at the switching frequency) and it probably should have been tapped off elsewhere or measured with an oscilloscope (but likely would not help either as i doubt the bandwidth of that clamp is above the switching frequency)
it might be better to try and find that rail elsewhere, those polyfuses next to it look like a good bet
love ur work bro! very interesting results on the chipset
At least four review sites (including you) have noted that high idle voltage now, not even counting the chipset. For the chipset, I wouldn't have actually expected it to eat much power unless there were a ton of PCIe-4 devices plugged in. Sure, it will use 4 lanes of PCIe-4 between the CPU and the chipset, but that's just 4 lanes. What is going on, indeed! The chipset just shouldn't be producing enough heat to require that little fan to turn on, under any load, short of having a lot of PCIe-4 peripherals.
So perhaps there is a voltage control issue. I can't think of a single reason for that X570 chipset to burn so much power either, other than being over-volted. Makes no sense to me either.
I'm not entirely sure those resistors are 0-ohm bridges. Are you sure they aren't current sense resistors? i.e. very low but non-zero ohms? The voltage could easily read the same on your 3.5 digit multi-meter. However, if the SMD resistor has a single digit '0' sillkscreened onto it, then it *is* a zero-ohms bridge.
-Matt
some engineer: "wait shit did i leave that resistor in?"
A reason for the fan on the chipset may be because of time constraints the chipset was the main bottleneck for x570's release, so I can imagine a scenario where the motherboard manufacturers suddenly had to deal with a 15w rated chipset. Being unable to produce and most importantly validate a passive heatsink in time.
As for testing the power consumption: you can still test usb as well to see if t hu at may increase it
Did you disable everything in the bios ? audio, wifi, all external connectors and imputs etc ?
Isn't the inter-connect between the CPU and chipset running PCI-Express 4.0? So regardless of what you connect you will still have increased power consumption due to the inter-connect between the CPU and the chipset? As AMD Ryzen 2000-series does not feature support for PCI-Express 4.0, what happens if you replace the Ryzen 3000-serie CPU with a Ryzen 2000-series CPU? Do you notice and difference in power consumption?
Regardless of all of this, it seems like the cooling is built to be overkill on purpose. Barely anyone will ever reach and wattage that will actually require active cooling. Hopefully you can disable the fan in the UEFI firmware, or at least have it not run unless you hit high temps.
You are spot on!!..I test on my entry level MSI x570 GAming Plus, with 0 rpm or 1100rpm the temp are the same, 70º, and this after 1,3H streaming and gaming the same time!!...So i think some fishing about the reason why they put active cooling on chipset...
In my experience from the past (both amd nor Intel) when northbridge chipset very hot system become hang or freeze then shutdown byself.
In the worst scenario case temp, airflow or ambient too hot small fan it can help you.
first I didn't understand what was wrong with my old system ! struggle few months until I finally realized northbride was very hot and decided to reduce northbrige voltage that's work but reducing overclocking rate too.
Edit;
My temperature CPU under full load 70~77℃ isn't too hot. LOL 😂
What are safe temps for the sb on the chipsets? Trying to figure out what fan curve to make and if i'll instead need to get a heatsink like you did.
Also what are the dimensions of the heatsink and how did you attach it? I see some heatsinks can be attached just by sticking.
The cIOD on Matisse works on PCIe 4.0 on all boards including X570/400/300 Series. That enables PCIe4 NVMe support on 400/300 Series boards.
The chipset uses the same design.
Maybe the chipset runs on PCIe 4 all the time.
Great video. I have a small test in mind that you didn’t do. You used PCIe Gen 4, yes, but only one device and only 4 lanes. I would try the gen 4 nvme + a Radeon 5700(or XT), which is a total of 20 Gen 4 PCIe Lanes. Maybe with all the lanes occupied it makes a difference?
Just wait for the water cooled motherboards. Should wake up some nostalgia :D
I want one
Great service, thank you!
Many thanks for testing this... and Yep. This is very trange... have you been asking mother board vendors hard questions after testing this? Pci 4.0 need better wires, but that should use even less electricity because thicker wires should be better conducts (in reality there is no difference unles you put heck a lot current in there) Maybe They just use higher voltage, like when overclocking memory etc. And that is why effiency is thrown out of window in here. Aka a Little bit more stability, by using really high voltage and thus using a lot of energy for minimal benefit? That is my ques...
I have no problem with the fan BUT the real question is: How long it will take to fail??? I expect lots of RMA's
A few weeks after warranty expires
Yeah. Companies Are very good at making things last as Little as possible, while over the warrantry time. Maybe companies Are not happy because amd motherboards can be used so long time and just upgrade cpu and gpu when needed. Now They kill the motherboards as soon as possible ;) my foliohat is too tight!
Keep dem fans lubricated essentially.
it is the line connection between the CPU and Chipset. 4.0 has some power for the 3950x and above to handle and the Connection made above 70 degree in AMD inside tests. So they desided to be sure the new MB can handle upon Ryzen 4th gen, which will be much more power stressing the new MB's. It's for the "we guarantee you 3 years on same mobo" Thing as it's still thought to be up/down ward compability.
When you add the pcie graphics card to the mix you should had stressed tested with some benchmark that really stressed the pcie bandwidth I think Sisoftware Sandra had a module test for that. Furmark doesn't bottleneck half a pcie1x. Or 2 or 3 nvme drives, just stress then witn file copy from 1 to another. Just some ideas that might work. Good luck.
Luis J maybe new futuremark pci test?
Did you test with the lanes between CPU and X570 running at Gen3 instead of Gen4? It should be configurable in the BIOS.
Great video, just the info I was looking for. Wanting to replace the monstrosity of a 40mm fan PCH cooler on my x570 board with a passive cooler and wasn't sure it could handle the job. Looks like it should work just fine especially being right in front of the case fan. I'm using a compact video card so I have all the room i need for it. Funny thing I have a couple Zalman passive northbridge coolers from like 2005 I almost threw away thinking I'd never have use for them.
so for average person this means, if the x570 mb chipset fan fails it does not mean that the chipset will fry ? it still has to have some sort of heatsink for the fan to cool right ?
At least we know now that when those fans start choking on dust, the chipsets won't literally be igniting.
70c is still pretty hot...
Wonder if they aren't having production issues making these new x570 chipsets, and are essentially having to overvolt them to get the clocks they need. I wonder if there's anything hinting at this? Is the main 1.075V supply now a higher voltage than the old x470? As for it idling high, it might just be that they have to run the whole thing at full speed even if there aren't pcie 4 devices present. Hmm. Not great tho.
One option for measuring current could be to use very low value shunt resistors (like 0.01 ohm, available in smd) and measuring the voltage across them on the board, maybe even with a scope. Maybe a little easier than large loops of wire.
Looking at the idle difference between x470 chipset and x570 and nvme gen3-4 difference points at x570 chipset just being crap, much worst designed than x470 ones. Great video analysis!!!
Has anyone checked the fan control of these fans? Are they always on or only at highest loads?
What's the total resistance of those wires ? More resistant less power right ? or the PSU compensates ?
Also what's the point of this video? Why the chipset PC is important ? Like those RGB LEDs pull more power than the chipset.
EDIT: okey i wrote before i watched that last 3 minutes.
I love the GB P67 UD7 mobo, but i guess that heatsink design can't be applied to the new mobos becase of the nvme slot right under the CPU socket.
Thank you and I see comments suggesting to use a ryzen 2000 or even 1000 series, I agree
7 watts at idle is absolutely insane. That's more at idle than most SoCs under load
Amazing findings! Thank you!
Love to hear more about this. Great vid!