You can tell the mans a classy guy, just look at the background lighting the props and stuff. Then you think = before he even says the first words of the script, your eyes are telling you that a lot of effort went into this. I been watching for ages, he had this level of polish even at 800 subscribers lol I think it's amazing he doesn't have millions of subs. This has been the best tech channel for ages.
he says he's a creator, but what exactly does he create besides this channel? VFX？ Art? Illustrations? Game assets? seeing his "sophisticated" diagrams, i doubt he's an artist.
Bruv, I gotta pause to tell you. This is a LOT of work to go through, to explain PCIe. I watched it whilst thinking, this is the way to make a dull subject come to life. Instead of bits you have sports cars, instead of nanowires you have highways. This channel is seriously under-used, you should have millions of people at least! Best tech channel BY far and has been for a long time.
The pros and cons of wanting wanting to have a professional creator PC. You can get by using a cpu that only has 20 or 24. But once you populate all of the m.2 drives and all of the pcie cards everything is running at half the intended speed. If you want to have a professional creator setup spend the money on the threadripper or equivalent from Intel. Lastly, having the latest and greatest generation doesn’t mean the software or equipment will run smoothly. Or even run at all. Stability is king. Good luck 🍀
And often some of the PCIe slots or SATA ports will be disabled if you install M.2 because the lanes are shared between them. If you even have the extra slots/ports now that most mainstream motherboards only have 3 or 4 PCIe slots and 4 SATA ports, that's enough for gaming PCs but extremely limited for workstations.
Running two dozen NVME drives at half speed would be great. Most boards only really attach 3~5 suitable NVME drives and nothing more, at any speed. No U.2 at all, and M.2 is just way too large of a connector for more than a boot+primary software.
Why everything would run at half speed? They are connected to their own individual lanes. They will work at 100% speed. You are spreading misinformation.
a better "highway" analogy would have been if each generation goes faster. gen1-10mph, gen2-20mph, gen3-40mph, gen4-80mph, gen5-160mph. so you can either have more lanes or have the "cars" go faster. So 2 lanes of gen1= 2 cars going 10mph while gen2 can do the same as gen1 with one lane cause theyre going twice as fast
I was successful as an engineer and supervisor in the corporate world with a company that manufactured high-tech electronic components. Problems often arose in production because technology was not communicated to the production workers in a language they did not understand. You remind me of me forty years ago. We both have the ability to function as a bridge connecting important technical concepts to those who need to become more familiar with those concepts. Your skills will serve you well. I've been completely bored with computer youtube sites that do not take the time to explain the thousands of acronyms so a person can receive the value the presenter is trying to communicate. Another wonderful aspect of your videos is you do not try to entertain your audience with dry humor or quirky graphics. Making every minute of viewing your work more valuable.
Not near as much as I appreciate yours. Had I found your channel several months ago I could have saved a couple thousand dollars. Now I am totally rebuilding the first machine into a machine that will serve me better. I wish now I had gone with intel rather than amd, but it is too late now. @@theTechNotice
That’s one of the primary reasons I use old used xeons for my NAS systems rather than cheaper and more power efficient low-end consumer-grade hardware. I wanted those lanes for network cards, host bus adapters for SAS or SATA drives, or even NVME drives.
Yeah but pcie 5 is double pci e 4 and thats double pcie 3 so even tho limited amount of 5 its still 4 times more than 3 (pcie 3=40 / pcie5 =10 (x 2 pcie3=20, x2 pcie3 =40)
@@dexterjsullen Generational improvements to PCIe bandwidth do mean better speeds with fewer lanes, but there are few ways to fully leverage those faster lanes. Gen 5 and even Gen 4 bulk storage, storage controllers, and networking are quite expensive (think high hundreds if not thousands of dollars), and carrier cards that can split those high speed lanes into more usable configurations are also very expensive. Most cheap cast-off enterprise gear is Gen 3 anyways, so systems that support at least Gen 3 are the most budget-conscious choice.
this is probably the best explanation of PCIE i've seen in youtube, i know fully understand what pc builders mean when you are wasting PCIE lanes and the transfer speeds, thank you
You give the best information hands down!!!! I built my first editing machine watching your videos 3 years ago and it has never failed! Thank you!! Your information is changing lives!!!!
This is why my latest build is an Intel Xeon W7-2495X with 64 PCIe5 Lanes plus the ASUS Pro WS W790 ACE chipset lanes. The W7-2495X 24-Core at 4.8GHz all-core also benches at 150% faster than the i9-13900K in multi-thread tests.
I got the same motherboard myself but settled on the 12 core 2455X. :) Depending on price and performance increase I might upgrade to a higher core count model when the next generation emerald rapids come out. If they come out.
@@stefannilsson2406- Other than issues of compatibility with some models of M.2 drives, my workstation has been running great. The 512GB of memory has been awesome. I do Unreal Engine 5 work plus massive terrains with TerreSculptor.
I watched "MANY" Videos and read a lot about PCIE Lanes and to be honest... This is the most simple and complete video on the internet Unfortunately, we can't have both Core performance and PCIE lanes at the same time I guess that's life :)
Aww man I really wish you would start covering MIDI audio interfaces and equipment! I started making music but the guys doing the tutorials are not easy to understand. I think you could do a better job lol even though it's totally new to you. Also glad to see your channel doing so well! Congratulations and God bless :)
I think a more accurate way to describe it is that the number of lanes equate to the lanes of a highway, while the generation is the speed limit of that highway. Each generation is double the speed of the last, but has to run at the lowest speed of the device or the endpoint (whichever is slowest).
I'm so happy seeing your channel get bigger and the goodies you can show us. You're my 'number one' source for 'Creator PC Info'! I've been realizing I really need more PCIE lanes so I can rig an M.2 PCIE card with 4 M.2 for fast backup retrieval along with my existing SATA card. And I don't like having to go into BIOS to bifurcate.
What do you mean you don't like having to go into BIOS to use bifurcation? Bifurcation has been used in computing for the longest time. While it is normally a function associated with Enterprise/Servers computing, where you can actual make adjustments in how you divide your lanes on a particular PCIe slot, it has always been hidden in the background on consumer devices. I am unsure about your negative connotation on Bifurcation. Using it has no speed degradation, especially on consumer Gen 3/4 drives.
@@sliceofmymind mostly because I am concerned that I might be reducing my PCIE lanes other components use. I have the GPU, three M.2 drives and a SATA controller card that controls 10 SATA drives. I'd just like to have a ton of slots and know whatever I plug in is going to run as fast as it can.
There was a reason long ago I made a move from the intel 3770k 4 core 8 threads with 16 pcie lanes and dual channel memory to a 4930k 6 core, 12 threads with 40 pcie lanes and 4 channel memory.. not sure why Intel went away from these type chips.. I guess they just wanted to short us from having real connectivity.. I can still run 3 way sli with my machine.. LOL but until Intel stops being cheap with the pcie lanes, I am not buying.. and my old board for the 4930k does not have any of the m.2 slots on them.. it just has the older 6 gbs sata connects
This. This right here is why I moved to a Treadripper build. Got my 5965WX last week and the only surprise I have so far is how smooth windows 11 actually is. But yeah, that expansion slot freedome is a godsend. Grabbing a 2nd GPU soon and a capture card to fillout the rest, but damn!
100 agree. Best bang for your buck as a CC is the lowest core count TR PRO. The 128 lanes is a game changer. the new Xeon are priced way out of most budgets due to the RDIMM ram.
I personally bought the Xeon platform instead of TR. I don't need 128 lanes, but I wanted more than 20/24. So I got the w5 2455X with 64 gen 5 lanes. It is also quite a lot faster than the TR's in single core applications because of overclocking support. And you can overclock these a lot. Intel really went conservative on the clock speeds and left a lot of headroom on the table. The CPU and mobo was cheaper than entry level TR, but the memory as you stated was more expensive, so all in all I'm probably around the same ballpark as a "low end" TR system. Though, the DDR5 RDIMMs are pretty cool. The Xeons only support buffered/unbuffered ECC RDIMMs, but for the first time in history as far as I know, we now have RDIMMs with X.M.P. I have 6000MHz CL32 ECC sticks. :)
I have a lower end Z790 motherboard that has a whole bunch of PCIe slots (only 1 x16) as well as a handful of USB 3.2/3.1 and six SATA. My GPU and boot SSD are attached to the CPU lanes; the CPU has a 16-lane port for the GPU and a 4-lane port for a SSD. My chipset bus has the following on it: * Chipset USB 3.2 controller * Chipset SATA controller (with 6 usable ports) * Other chipset odds and ends * Motherboard 802.11AX/BT 5.2 card (Bluetooth attached via the above USB) * Motherboard 2.5GBit Ethernet * Motherboard audio * Secondary NVMe drive * IDE/PATA controller (connected by an on-card PCIe to PCI bridge) * COM/LPT port combo card * Additional USB 3.0 card * Sound card * An older graphics card
A rather extended comment regarding the ASUS M.2 Hyper-type PCIe cards that support multiple M.2 NVMe SSDs: The PCIe cards that can house multiple M.2 NVME SSDs, usually 2 or 4, can be largely classified into 2 types: an M.2 to PCIe RAID card and an M.2 to PCIe adapter. The main feature that distinguishes these two types of PCIe cards is the existence of something called a "PCIe switch chip". These chips are responsible for receiving PCIe lanes from the motherboard and splitting into multiple devices; in this case, M.2 NVMe drives. M.2 to PCIe RAID cards have these PCIe switch chips, and are typically marketed as being "no PCIe bifurcation needed" (more on that later). Proper M.2 to PCIe RAID cards are usually hassle-free as they can offer hardware-level RAIDs and are plug-and-play while being cross-compatible with various platforms (including Mac desktops with PCIe slots!), but it comes at a hefty price tag, with even the cheapest 4-drive enclosures reaching $200. Cheaper cards typically support only up to PCIe 3.0 speeds (the version is dependent on what the onboard PCIe switch chip supports) and prices skyrocket even further if you seek PCIe 4.0 speeds. M.2 to PCIe adapters are a more budget-friendly option that only has multiple M.2 slots and does not feature an onboard PCIe switch chip. Now, this type of card is not plug-and-play, and requires a BIOS level configuration called "PCIe bifurcation". This feature allows for you to configure the x16 slot (usually the first PCIe slot for most consumer motherboards) to be split into multiple separate lanes, such as x8x8, x8x4x4 and x4x4x4x4. While most modern consumer motherboards, except for some extremely basic ones, support PCIe bifurcation, the freedom to which you can configure depends on the platform and the CPU. For Intel 1200 and 1700 platforms, PCIe bifurcation is limited to x8x8, meaning that the motherboard can only detect up to 2 NVMe drives using an M.2 to PCIe adapter. For AMD AM4 and AM5 platforms, you are given a more generous option of x8x8, x8x4x4 and x4x4x4x4, so AMD platforms can detect all 4 NVMe drives using x4x4x4x4. However, if your CPU is a G-series APU, x4x4x4x4 bifurcation is not available and the most you can get is x8x4x4, so AM4 APU desktops can detect at most 3 NVMe drives using this method. Now, the biggest issue with M.2 to PCIe adapters is that the availability of PCIe bifurcation heavily depends on whether your motherboard supports it in the first place, and there is no easy way of getting this information if the motherboard manufacturer does not include it in the specifications. The only manufacturer I could find that keeps track of all bifurcation options is ASUS, who is probably keen on making their M.2 Hyper PCIe adapter as forwards compatible as possible, and you can find such information on either the specs or support pages. For the rest of the motherboards, you need some other deviant methods involving the motherboard's BIOS. (I will leave this up to the reader to google it since it is way beyond the scope of both this comment and my knowledge.) As it stands, trying to expand your M.2 drives using PCie slots is quite an adventure where either your wallet will suffer or your brain will. However, if you are up for the challenge, it can be a long-term solution to storing your data while retaining the swift speeds that PCIe interfaces can offer.
@@3k3k3 Depends on the platform. If you'rw talking about the top-of-the-line WRX80-SAGE that supports 39o5WX or 59o5WX Threadripper PRO CPUs, it turns out that you can put *seven* ASUS Hyper M.2s all at x4x4x4x4 bifurcation, giving you a total of *28 M.2 NVMe drives!* Depending on which motherboard you use, ASUS sometimes uses the term to the tune of "M.2 Hyper mode", but since the base technology of it is PCIe bifurcation, it is most likely that any switchless M.2 to PCIe adapters are compatible with bifurcated PCIe slots. I say this because after an ungodly amount of googling, I found that there are alternatives to ASUS M.2 Hyper. If you are to sport PCIe 3.0 NVMe SSDs, chances are you can get away with using a card that comes with decent heatsink enclosure, and if you have some really efficient SSDs like the Hynix P31, you may even get away with heatsink-less barebones adapters too. If you are to sport PCIe 4.0 having an active cooling solution is likely a good idea, but ASUS's solution, while being one of the most well-known and accessible of the bunch, it also comes with a pretty significant downside of being 270mm in length, which can be a lot if you are gunning for a more compact enclosure. It turns out that a Sabrent 4-drive adapter, which supports 4 PCIe 4.0 NVMe drives and a cooling fan, is about 180.34mm (7.1 in) in length, so if a size limitation is your issue, this could be a good alternative. Taking a peek at some lesser known markets like AliExpress can take you to even more budget-friendly options, or wilder ones like having 4 M.2 slots on a low profile enclosure. The rabbit hole here is way deeper than you expect, so if you are into M.2 expansions, well... good luck on your search efforts! 😆
@@CsQ_RandomRepository interesting, the reason i am asking is that i was playing around (testing) different setups with a Gigabyte GC-4XM2G4, and this card features 2 dip switches and the manual says that these have to be set according to number of cards in the system. And the max is 3 The ASUS cards doesn't feature any switches.. Anyway, thank you for the answer, it's a shame that there aren't that many cards out there that supports 8 or more drives but with a PCI bridge chip, that would be interesting :)
@@3k3k3 8 M.2 to PCIe converters are almost exclusively RAID cards with PCIe switches because otherwise, a theoretical 8-way PCIe bifurcation (which to my knowledge no motherboard supports) would cause poor drive speeds since they are given x2 bandwith instead of intended x4 lanes. So yeah, 8 M.2 to pcie RAID cards... you are now flying close to 4 digit price tags and companies that offer serious server/workststion solutions like HighPoint, OWC and Sonnet. This is probably the most extreme of M.2 expansion solutions money can buy.
personally i am getting there. 3 u.2 drives, good gpu, dedicated 4x capture and recording cards, 10gig etc... i'm done with desktop setups. thanks for the great vid!
This is why I'm still running an X299 system. It's not expensive like a Xeon, but has way more PCIE lanes. I'm running an i9-7960X so it has 44 PCIe lanes. I do photogrammetry that uses CUDA cores to accelerate the calculations to make 3D models from photos. So I run a 3070 Ti on my primary slot, and two GTX-690 cards on other slots. The GTX-690 is a dual GTX-680. So I have 5 GPU cores in my system and the two 690's have a total of the same number of CUDA cores as my 3070 Ti. Currently the first 690 is on the second X16 slot and the third is on the X4 slot. That's ok because I don't really need it to be running X16 on the second and third card. I also have a Tesla K80 datacenter GPU that I'm hoping to add as well. The problem is drivers. Windows doesn't like using multiple GPU drivers. To get the 3070 Ti and the 690s to all work I had to go back to nVidia driver version 471.41. That was the last driver that supported the 30 series AND the older 690's. Once I upgrade to a 40 series I probably won't be able to use the 690's anymore. I MIGHT be able to use the Tesla K80 though since it's a completely different kind of card. Not sure. I would probably be ok just using the chipset to feed the second and third cards on a newer system, but it's kind of nice knowing I have all that bandwidth available. The only reason I didn't put the third card on the last X16 slot (which is actually only an X8 electrically) is simply because the riser cables are too fat and I ran out of room.
The Z790 Aero G can only do x16 to directly to CPU on PCIe 5, 4x on PCIe 4 to the CPU (M2A_CPU m.2 slot). There are actually two m.2 slots that hit the CPU, one is the PCIe 4 one, other other is a PCIe 5 one. But if you use the PCIe 5 one (M2C_CPU), then the first PCIe slot operates at 8x only, wasting even more lanes :/
I always questioning sometimes how absolutely weird Intel PCIE layout can be. Just look at those mobo that has two PCIE x16 that only apart by one slot. Like what the heck they even supposedly be doing
For the description of the generations and lanes and the analogy of cars and highways. For each generation I would have used the example that the speed limit is doubled each time as the lanes themselves are quite literally physical.
Oh one thing you forgot to mention is that a PCIe Gen 4 GPU will switch down the Gen speed when it is not used in power saving mode. This frees out the lanes for M,2 SSDs to run fast if they had the x4 lane speed configured. Also the CHipset has a PCIe bridge function that can multiplex more physical lanes but they dynamically share the available lane bandwidth. Latency included as well due to the extra hop to the CPU controllers. Rule of thumb in p[ower consumption is the higher the Gen type, the more power they consume!.
Had to come and reply to this as I was scrolling comments, what? What platforms are you operating on when a 5.0 PCIe lane is switching down to another lane? It isn't a software or hardware switch like "power saving" mode in your basic Windows settings. PCIe slots do not switch down bandwidth generations. The generation is built into the CPU itself paired with the chipset. They're backwards compatible, yes, but... it doesn't "downgrade" itself to 5.0 to standard PCIe 2.0/3.0/4.... Lol.
As it is getting into how many pcie lanes you have you also have to look at where is the pcie signaling clock as plugging in pcie 3 into pcie 5 then pcie 4 into pcie 5 then pcie 5 into pcie 5 if all 3 slots are on the same signaling clock then everything will run at pcie 3 speeds. It mostly is fixed now by having each slot on a different signaling clock but on same boards bifurcation will allow you to have more slots though less lanes per slot they will also all run at the same pcie generation. The way the ASUS Pro WS W790-E got around it is the CPU has enough lanes that everything has their own pcie 5.0 lanes from the CPU and thus no bifurcation. Now it is a 1300USD board and the CPU for that to be so is on the Xeon W-3400 series so starting at 1700USD and goes up from there. It isn't a cheap system but if you need everything to have direct to CPU pcie lanes and enough lanes for everything to not have to share it is going to cost you. He got how the rear I/O is connected wrong. The rear I/O the video out is directly connected to the CPU, most if not all USB ports are connected to a USB controller then the CPU , the rear I/O NIC to a network controller then to the CPU. The audio is connected to the audio chip then to the CPU. We have been moving away from connecting to the south bridge as it is slower.
LGA1700 CPUs (12, 13, 14 gen) only have 16 lanes of pcie 5.0. That's why on MBs that support an m.2 pcie 5.0 slot if you populate that it will automatically switch to x8 for the GPU. So directly from CPU there are 16 lanes gen 5 and 4 lanes of gen 4 (plus additional x8 gen 4 lanes for connection with the chipset).
I really think it's time to explain pcie lanes as networking connections, bandwidth, latency, and routing. What DMA is and how some hardware can bypass it altogether with device to device data transfers.
Another great video might be how to power a 4090 with the proper PSU. I've been hearing of a lot of people ruining their cards by using an under powered PSU
One thing to be noticed, the second-to-the-last PCIE slot on Asus Sage are PCIE 5.0 x8, which the remaining 8 lanes are used for 2 m.2 slots on the right side of board.
Thank you so clear explanation Can you next explain motherboards and how to choose a motherboard what expansions integrated or slots to look out for plz
Also memory channels are important. Threadrippers and Xeons use 8 of them. It would be nice to understand if r-dimm memory are better for these motherboards and check speed vs unbuffered dimms. DDR4 Vs DDR5 would be a nice comparison.
R-Dimm or registered ECC memory does error correcting (checks the integrity of the data) where U-Dimm doesn't - basically if you really care about your data you run a system with R-Dimm ECC memory. There is no performance benefit with R-Dimm, most of them run a bit slower than your typical consumer U-Dimms
Versed in PCIe lanes, I clicked, as your perspective differs from the pure gamer. In the beginning, there was an ommission--the 13900k has 20 gen 5.0 lanes and 4 gen 4.0 lanes. This allows 16 5.0 lanes, bifurcation aside, and 4 lanes for an NVMe or some other peripheral directly connected to the CPU, as various adapters are available to turn it into a standard PCIe 4x 4.0 slot. I don't fool with AMD--too many headaches. AMD and Intel do not report lanes equivalently. Saying the AMD with 24 compared with 20 for the Intel 12th and 13th generations is incorrect. Intel reports accessible lanes. With Z590 with Rocket Lake CPUs, Z690, and Z790 have 20 user accessible lanes. There are other lanes present, 8x, IIRC, to the PCH for other built-in peripherals such as LAN, wireless LAN, USB, Thunderbolt, SATA, or other NVMe. AMD reports total lanes, accessible or not.
i've actually been looking at one of these CPUs for a future workstation at home. I'm not a creator or 3D renderer but I might have to render some things for work. Most of what I'll do will consist of AI and in general HPC which will take a lot of processing power and VRAM So i'm thinking of getting a Lenovo ThinkStation P620 with a 32 cores 5975wx CPU, 128 GB of RAM that i'll probably expand later and a dual A6000 GPU combo. Most likely gonna dual boot too to have Windows for anything video related and Linux for anything coding related
Why I am edging toward threadripper. Not because cores but lanes. Lower number of cores to 12, make it the F version, then much lower tdp too. Great for desktop that needs lanes not cores.
How do you use cables to prevent the GPU's from covering up the lanes? How do you power two 4090's? It would be cool to see a video where you build a workstation with two GPU's. Thanks for the great info in these types of videos
I'm rebuilding my NAS with an Epyc CPU/mobo for this exact reason. Could have spend less on consumer stuff but I need 10gb LAN, HBA and a 4x NVME SSD card. With all those thrown in I still have PCIE slots left for a GPU if I need it.
You know they make consumer boards that already have 10GBe included. Sometime people think they need overkill, when it's rarely the case. As something you'd be using strictly as NAS, with a single 10Gbe you can't even use the full bandwidth of a single gen 3 NVME drive. You only gain a large pool of flash with a 4x card, that you'd never get the full bandwidth potential out of. 10Gbe translates to 1.25GBps, which is the absolute limit of getting data in and out of it. My 4 7200RPM RAIDz1 drives already max at 800Mbps, just with the onboard SATA. Doubling that to 8 with an HBA ultimately gets limited by the NIC. That HBA could run off of the 4x chipset slot. The typical use for a GPU in a NAS is media encoding, and that doesn't even demand the GPU be allotted all 16 lanes. The $500 Asus X670e Pro Art has 10Gbe (and TB4), 3 Nvme slots, ECC support, and CPUs with integrated graphics. Sure the iGPU would only be useful for the setup until media servers are actually willing to address its media encoder, but a 6 core 12 thread CPU is only $240 on the platform. More than enough for a dedicated NAS. I've been running truenas in a VM with 4 threads and 8GB of RAM for 2 years, in a PC otherwise used for more virtualization and gaming. You should be doing way more with it to need Epyc.
well done video! for illustration purposes, I'd rather recommend using trains instead of cars in way like this: motherboard is a town, cpu is main train station, chipset is secondary train station, each lane is a railway line, PCIe generation is amount of wagons a train can deliver ;)
Why is the Chip-Set type and features only emphasized with PC Motherboards? When I look at Laptops, I might see if they support Thunderbolt 4, and such, but they never mention the Chip-Set, like they do with Motherboards.
This video is very under-rated! I am now looking at my MSI MEG X570S ACE MAX like it's a child lol now I feel the need to move up to a ThreadRipper/Workstation because I am bottlenecked with my 4 m.2 NvME Drives x 2 M.2 running in RAID. Its time for me to upgrade anyways as I am on Gen 3/4 components.
I see now that the difference between Server Xeon processor and Workstation Xeon processors are the number of pcie lanes (server platinum xeon 80 lanes, xeon w-2400 64 lanes for workstation, xeon x-3400 112 lanes for workstation). But with a motherboard that is designed for workstation purpose but accepts the Xeon Server (which can be found cheaply than workstation Xeons) we get better performance at lower cost.
Hey Sarge, it seems that even with the best consumer level stuff if you plan to do any serious 4k+ 6k+ workflows, rendering etc. One should probably look at saving$ & going straight for a workstation type setup ? I know that sends the budget into the stratosphere. With YOUR HELP (thank you)...I have built a nice I91300k w/Asus 790mb Intel ArcA770 16 gb card...but do get hitching, bottlenecks, rubberbanding from time to time working with 5.3k off the go pros etc. It is a huge improvement working on the rig compared to the 8 year old laptop I had..
Good presentation! Ryzen 9 having 24 lanes only and dual channel config is straight criminal! Those who game are fine with their R5-R7 but R9 is meant to be a creator CPU and those specs are laughable.. Most comments: want more lanes? move to Threadripper or EPYC..
Is the reason that the Thread Ripper and Xeon processors have more pins because of all the PCIe and memory lanes. Meaning that we won't get anymore PCIe lanes until CPU manufacturers change the CPU socket and at the moment they won't do that.
I was hoping to get a question answered by this video, but I’m afraid it didn’t. I have a 5900x on a phantom gaming x570 MoBo and I’m running a 3070 and 2 NVMe drives, am I short on lanes? If so, what should I be experiencing, such as slowing down somewhere? This is really my first PC and I don’t quite understand the fine aspects of a PC. I know my CPU has 20 lanes, but then he confused me by saying MoBos can alleviate bottlenecks some how with their chipsets and I got confused.
It’s not clear to me whether or not the chipset acts as a traffic-cop when dealing with all the extra peripherals attached to it. Can someone please clarify?
Well that was disappointing. Very oversimplified and at times kind or wrong. Intel 13gen cpu and x670 cpu have 28 lanes worth of bandwidth which are differently allocated. I am really missing information about bifurcation, how much bandwidth a gpu, a capture card, network or ssd storage really need. Which is much less that most people think. Most gpus are fine with 8 gen 4 lanes, 10gbe need technically 1 gen3 lane and so on
Let's say a gpu uses 16 Gen 5's. Can you maybe the just use 32 Gen 4's to get the same amount of "highway" since Gen 4 is half the size? Or does it not work like that?
how motherboard know which Gen is slot? example B450 is Gen3, but if we put 5000 series cpu, will GPU rtx4090 work in Gen3 standard or Gen4?? Because CPU support gen4??? lanes go directly to CPU. How that work??
how can i tell whast is using my lanes? for example ..after the GPU what is going on the remaining CPU lanes and what is using the chipset lanes.. can devices be assigned CPU lanes specifically? Assuming CPU lanes provide better data tranfers.. lets say i buy a faster m.2 and a slower one.. can i assign the faster m.2 to the remaining CPU lanes and leave the slower one to be used my the chipset? Thank you!!
question about M2 drives~ I'm building an AMD 7700x pc with ASRock B650 MB, it has a gen 5 pcie slot, also a gen 4 & gen 3 slots, but only the gen 5 has a heat sink... I have a 1 tb m2.0 gen 3, and getting a new one gen 4, which slots should i use for both?
Here's a question if I put in an expansion card for more storage and force my 4090 to run on 8 gen 5 lanes will it still work just as well? Since it normally runs on 16 gen 4 lanes.
I have a question...if I have a 7900XT running on a PCIe 3.0 board, how much am I actually losing in performance without 4.0 available? I have a 9900K so it's probably getting time to upgrade soon, maybe later this year, but I'm just curious. I have 3 M.2's populated on the board too, which probably doesn't help lol
Would an high end AM5 APU benefit from having gen 5.0 pci-e lanes instead as AMD planned to have it 4.0, when it's putted on gen 5.0 capable motherboard ?
Now say your GPU had 16GB memory. Then 8 lanes of PCIe 4.0 can fill it in one second as it has speed of 15.754 GB/s. That is the magnitude of things. To comparison typical AMD DDR5 memory bandwidth is 70GB/s.
Do you know ANY motherboard that can do PCIe bifurcation (x4x4x4x4) for 2 of their x16 pcie slots? thinking... 16 lanes of pcie (x4x4x4x4) - first card carrying m.2's 16 lanes of pcie (x4x4x4x4) - second card carrying m.2's 4 lanes of pcie - m.2 boot device 1 lane of pcie - video card 3 lanes of pcie- other motherboard things/NIC etc === 40 lanes. Haswell XEON? 2650Lv3? or similar? so, 8x 4tb m.2 drives with 2 redundant ones.. = 6 drives worth of storage. = 24tb? Is this a reasonable thing? and also, is it a possible thing
![DON'T BUY THIS CPU! 👉 4 Mistakes to AVOID as Creators! [More Expensive & WORSE]](http://i.ytimg.com/vi/8M_zoDKw7XA/mqdefault.jpg)
![Why Your Creator PC Bottlenecks! 👉 How to find & FIX the problem? [Video, Photo & 3D]](http://i.ytimg.com/vi/s2O7P7Bmvg8/mqdefault.jpg)
I like how this is a sophisticated tech channel but has pen and paper visual metaphors to explain PC hardware.
Agreed. Pen and paper have roots for any kind of education. Whether it's tech or philosophy
You can tell the mans a classy guy, just look at the background lighting the props and stuff. Then you think = before he even says the first words of the script, your eyes are telling you that a lot of effort went into this.
I been watching for ages, he had this level of polish even at 800 subscribers lol I think it's amazing he doesn't have millions of subs. This has been the best tech channel for ages.
he says he's a creator, but what exactly does he create besides this channel? VFX？ Art? Illustrations? Game assets? seeing his "sophisticated" diagrams, i doubt he's an artist.
Bruv, I gotta pause to tell you. This is a LOT of work to go through, to explain PCIe. I watched it whilst thinking, this is the way to make a dull subject come to life. Instead of bits you have sports cars, instead of nanowires you have highways.
This channel is seriously under-used, you should have millions of people at least! Best tech channel BY far and has been for a long time.
Finally, someone else who understands the importance of more lanes
The pros and cons of wanting wanting to have a professional creator PC.
You can get by using a cpu that only has 20 or 24.
But once you populate all of the m.2 drives and all of the pcie cards everything is running at half the intended speed.
If you want to have a professional creator setup spend the money on the threadripper or equivalent from Intel.
Lastly, having the latest and greatest generation doesn’t mean the software or equipment will run smoothly. Or even run at all.
Stability is king.
Good luck 🍀
And often some of the PCIe slots or SATA ports will be disabled if you install M.2 because the lanes are shared between them. If you even have the extra slots/ports now that most mainstream motherboards only have 3 or 4 PCIe slots and 4 SATA ports, that's enough for gaming PCs but extremely limited for workstations.
@fuckingmetaldude Have you not seen the new Xeon W-3400 series CPUs? They aren't that far away from Threadripper Pro
Running two dozen NVME drives at half speed would be great. Most boards only really attach 3~5 suitable NVME drives and nothing more, at any speed. No U.2 at all, and M.2 is just way too large of a connector for more than a boot+primary software.
This X1,000,000,000!
Why everything would run at half speed? They are connected to their own individual lanes. They will work at 100% speed. You are spreading misinformation.
a better "highway" analogy would have been if each generation goes faster. gen1-10mph, gen2-20mph, gen3-40mph, gen4-80mph, gen5-160mph. so you can either have more lanes or have the "cars" go faster. So 2 lanes of gen1= 2 cars going 10mph while gen2 can do the same as gen1 with one lane cause theyre going twice as fast
I was successful as an engineer and supervisor in the corporate world with a company that manufactured high-tech electronic components. Problems often arose in production because technology was not communicated to the production workers in a language they did not understand.
You remind me of me forty years ago. We both have the ability to function as a bridge connecting important technical concepts to those who need to become more familiar with those concepts. Your skills will serve you well.
I've been completely bored with computer youtube sites that do not take the time to explain the thousands of acronyms so a person can receive the value the presenter is trying to communicate.
Another wonderful aspect of your videos is you do not try to entertain your audience with dry humor or quirky graphics. Making every minute of viewing your work more valuable.
I appreciate your kind words!
Not near as much as I appreciate yours. Had I found your channel several months ago I could have saved a couple thousand dollars. Now I am totally rebuilding the first machine into a machine that will serve me better. I wish now I had gone with intel rather than amd, but it is too late now. @@theTechNotice
That’s one of the primary reasons I use old used xeons for my NAS systems rather than cheaper and more power efficient low-end consumer-grade hardware. I wanted those lanes for network cards, host bus adapters for SAS or SATA drives, or even NVME drives.
Yes those X99 Xeons E5 26xx cpus with 40 Gen 3 PCIe lanes and Quad channel RAM in ECC. What more can you ask for the price the are right now?!!!.
Yeah the problem with efficient enterprise grade hardware like a embedded motherboard is expensive AF.
@@ImDembe If you are looking, I will be getting rid of my dual Xeon E5 2670 with MB ASUS z9pe-d8 ws. Comment here if interesting. In Sweden
Yeah but pcie 5 is double pci e 4 and thats double pcie 3 so even tho limited amount of 5 its still 4 times more than 3 (pcie 3=40 / pcie5 =10 (x 2 pcie3=20, x2 pcie3 =40)
@@dexterjsullen Generational improvements to PCIe bandwidth do mean better speeds with fewer lanes, but there are few ways to fully leverage those faster lanes. Gen 5 and even Gen 4 bulk storage, storage controllers, and networking are quite expensive (think high hundreds if not thousands of dollars), and carrier cards that can split those high speed lanes into more usable configurations are also very expensive. Most cheap cast-off enterprise gear is Gen 3 anyways, so systems that support at least Gen 3 are the most budget-conscious choice.
this is probably the best explanation of PCIE i've seen in youtube, i know fully understand what pc builders mean when you are wasting PCIE lanes and the transfer speeds, thank you
So glad someone is covering this! I think it's one of the most overlooked standards.
You give the best information hands down!!!! I built my first editing machine watching your videos 3 years ago and it has never failed! Thank you!! Your information is changing lives!!!!
The BEST Teacher / Trainer I have ever known - esp for content creators. His clarity of explaining facts makes it so easy for us.
agree - OP is most excellent Jedi Yoda - teaching us Lukes how to do things properly.
This is why my latest build is an Intel Xeon W7-2495X with 64 PCIe5 Lanes plus the ASUS Pro WS W790 ACE chipset lanes.
The W7-2495X 24-Core at 4.8GHz all-core also benches at 150% faster than the i9-13900K in multi-thread tests.
I got the same motherboard myself but settled on the 12 core 2455X. :)
Depending on price and performance increase I might upgrade to a higher core count model when the next generation emerald rapids come out. If they come out.
@@stefannilsson2406- Other than issues of compatibility with some models of M.2 drives, my workstation has been running great. The 512GB of memory has been awesome. I do Unreal Engine 5 work plus massive terrains with TerreSculptor.
I watched "MANY" Videos and read a lot about PCIE Lanes and to be honest...
This is the most simple and complete video on the internet
Unfortunately, we can't have both Core performance and PCIE lanes at the same time
I guess that's life :)
Aww man I really wish you would start covering MIDI audio interfaces and equipment! I started making music but the guys doing the tutorials are not easy to understand. I think you could do a better job lol even though it's totally new to you.
Also glad to see your channel doing so well! Congratulations and God bless :)
Love these videoes... used them to build my new 13900KS + RTX4080 Noctua build :D its awesome.
Appreciate the clear and thorough explanation!
I think a more accurate way to describe it is that the number of lanes equate to the lanes of a highway, while the generation is the speed limit of that highway. Each generation is double the speed of the last, but has to run at the lowest speed of the device or the endpoint (whichever is slowest).
I'm so happy seeing your channel get bigger and the goodies you can show us.
You're my 'number one' source for 'Creator PC Info'!
I've been realizing I really need more PCIE lanes so I can rig an M.2 PCIE card with 4 M.2 for fast backup retrieval along with my existing SATA card.
And I don't like having to go into BIOS to bifurcate.
What do you mean you don't like having to go into BIOS to use bifurcation? Bifurcation has been used in computing for the longest time. While it is normally a function associated with Enterprise/Servers computing, where you can actual make adjustments in how you divide your lanes on a particular PCIe slot, it has always been hidden in the background on consumer devices. I am unsure about your negative connotation on Bifurcation. Using it has no speed degradation, especially on consumer Gen 3/4 drives.
@@sliceofmymind mostly because I am concerned that I might be reducing my PCIE lanes other components use. I have the GPU, three M.2 drives and a SATA controller card that controls 10 SATA drives.
I'd just like to have a ton of slots and know whatever I plug in is going to run as fast as it can.
There was a reason long ago I made a move from the intel 3770k 4 core 8 threads with 16 pcie lanes and dual channel memory to a 4930k 6 core, 12 threads with 40 pcie lanes and 4 channel memory.. not sure why Intel went away from these type chips.. I guess they just wanted to short us from having real connectivity.. I can still run 3 way sli with my machine.. LOL but until Intel stops being cheap with the pcie lanes, I am not buying.. and my old board for the 4930k does not have any of the m.2 slots on them.. it just has the older 6 gbs sata connects
This. This right here is why I moved to a Treadripper build. Got my 5965WX last week and the only surprise I have so far is how smooth windows 11 actually is.
But yeah, that expansion slot freedome is a godsend. Grabbing a 2nd GPU soon and a capture card to fillout the rest, but damn!
@fuckingmetaldude buy a used one then?
100 agree. Best bang for your buck as a CC is the lowest core count TR PRO. The 128 lanes is a game changer. the new Xeon are priced way out of most budgets due to the RDIMM ram.
I personally bought the Xeon platform instead of TR. I don't need 128 lanes, but I wanted more than 20/24. So I got the w5 2455X with 64 gen 5 lanes. It is also quite a lot faster than the TR's in single core applications because of overclocking support. And you can overclock these a lot. Intel really went conservative on the clock speeds and left a lot of headroom on the table.
The CPU and mobo was cheaper than entry level TR, but the memory as you stated was more expensive, so all in all I'm probably around the same ballpark as a "low end" TR system. Though, the DDR5 RDIMMs are pretty cool. The Xeons only support buffered/unbuffered ECC RDIMMs, but for the first time in history as far as I know, we now have RDIMMs with X.M.P. I have 6000MHz CL32 ECC sticks. :)
Brilliant! @@stefannilsson2406
I have a lower end Z790 motherboard that has a whole bunch of PCIe slots (only 1 x16) as well as a handful of USB 3.2/3.1 and six SATA.
My GPU and boot SSD are attached to the CPU lanes; the CPU has a 16-lane port for the GPU and a 4-lane port for a SSD.
My chipset bus has the following on it:
* Chipset USB 3.2 controller
* Chipset SATA controller (with 6 usable ports)
* Other chipset odds and ends
* Motherboard 802.11AX/BT 5.2 card (Bluetooth attached via the above USB)
* Motherboard 2.5GBit Ethernet
* Motherboard audio
* Secondary NVMe drive
* IDE/PATA controller (connected by an on-card PCIe to PCI bridge)
* COM/LPT port combo card
* Additional USB 3.0 card
* Sound card
* An older graphics card
A rather extended comment regarding the ASUS M.2 Hyper-type PCIe cards that support multiple M.2 NVMe SSDs:
The PCIe cards that can house multiple M.2 NVME SSDs, usually 2 or 4, can be largely classified into 2 types: an M.2 to PCIe RAID card and an M.2 to PCIe adapter. The main feature that distinguishes these two types of PCIe cards is the existence of something called a "PCIe switch chip". These chips are responsible for receiving PCIe lanes from the motherboard and splitting into multiple devices; in this case, M.2 NVMe drives. M.2 to PCIe RAID cards have these PCIe switch chips, and are typically marketed as being "no PCIe bifurcation needed" (more on that later). Proper M.2 to PCIe RAID cards are usually hassle-free as they can offer hardware-level RAIDs and are plug-and-play while being cross-compatible with various platforms (including Mac desktops with PCIe slots!), but it comes at a hefty price tag, with even the cheapest 4-drive enclosures reaching $200. Cheaper cards typically support only up to PCIe 3.0 speeds (the version is dependent on what the onboard PCIe switch chip supports) and prices skyrocket even further if you seek PCIe 4.0 speeds.
M.2 to PCIe adapters are a more budget-friendly option that only has multiple M.2 slots and does not feature an onboard PCIe switch chip. Now, this type of card is not plug-and-play, and requires a BIOS level configuration called "PCIe bifurcation". This feature allows for you to configure the x16 slot (usually the first PCIe slot for most consumer motherboards) to be split into multiple separate lanes, such as x8x8, x8x4x4 and x4x4x4x4. While most modern consumer motherboards, except for some extremely basic ones, support PCIe bifurcation, the freedom to which you can configure depends on the platform and the CPU. For Intel 1200 and 1700 platforms, PCIe bifurcation is limited to x8x8, meaning that the motherboard can only detect up to 2 NVMe drives using an M.2 to PCIe adapter. For AMD AM4 and AM5 platforms, you are given a more generous option of x8x8, x8x4x4 and x4x4x4x4, so AMD platforms can detect all 4 NVMe drives using x4x4x4x4. However, if your CPU is a G-series APU, x4x4x4x4 bifurcation is not available and the most you can get is x8x4x4, so AM4 APU desktops can detect at most 3 NVMe drives using this method. Now, the biggest issue with M.2 to PCIe adapters is that the availability of PCIe bifurcation heavily depends on whether your motherboard supports it in the first place, and there is no easy way of getting this information if the motherboard manufacturer does not include it in the specifications. The only manufacturer I could find that keeps track of all bifurcation options is ASUS, who is probably keen on making their M.2 Hyper PCIe adapter as forwards compatible as possible, and you can find such information on either the specs or support pages. For the rest of the motherboards, you need some other deviant methods involving the motherboard's BIOS. (I will leave this up to the reader to google it since it is way beyond the scope of both this comment and my knowledge.)
As it stands, trying to expand your M.2 drives using PCie slots is quite an adventure where either your wallet will suffer or your brain will. However, if you are up for the challenge, it can be a long-term solution to storing your data while retaining the swift speeds that PCIe interfaces can offer.
Thanks for the writeup, do you think 6 ASUS Hyper cards would work in a Threadripper Sage board?
@@3k3k3 Depends on the platform. If you'rw talking about the top-of-the-line WRX80-SAGE that supports 39o5WX or 59o5WX Threadripper PRO CPUs, it turns out that you can put *seven* ASUS Hyper M.2s all at x4x4x4x4 bifurcation, giving you a total of *28 M.2 NVMe drives!*
Depending on which motherboard you use, ASUS sometimes uses the term to the tune of "M.2 Hyper mode", but since the base technology of it is PCIe bifurcation, it is most likely that any switchless M.2 to PCIe adapters are compatible with bifurcated PCIe slots. I say this because after an ungodly amount of googling, I found that there are alternatives to ASUS M.2 Hyper. If you are to sport PCIe 3.0 NVMe SSDs, chances are you can get away with using a card that comes with decent heatsink enclosure, and if you have some really efficient SSDs like the Hynix P31, you may even get away with heatsink-less barebones adapters too. If you are to sport PCIe 4.0 having an active cooling solution is likely a good idea, but ASUS's solution, while being one of the most well-known and accessible of the bunch, it also comes with a pretty significant downside of being 270mm in length, which can be a lot if you are gunning for a more compact enclosure. It turns out that a Sabrent 4-drive adapter, which supports 4 PCIe 4.0 NVMe drives and a cooling fan, is about 180.34mm (7.1 in) in length, so if a size limitation is your issue, this could be a good alternative. Taking a peek at some lesser known markets like AliExpress can take you to even more budget-friendly options, or wilder ones like having 4 M.2 slots on a low profile enclosure. The rabbit hole here is way deeper than you expect, so if you are into M.2 expansions, well... good luck on your search efforts! 😆
@@CsQ_RandomRepository interesting, the reason i am asking is that i was playing around (testing) different setups with a Gigabyte GC-4XM2G4, and this card features 2 dip switches and the manual says that these have to be set according to number of cards in the system. And the max is 3
The ASUS cards doesn't feature any switches..
Anyway, thank you for the answer, it's a shame that there aren't that many cards out there that supports 8 or more drives but with a PCI bridge chip, that would be interesting :)
@@3k3k3 8 M.2 to PCIe converters are almost exclusively RAID cards with PCIe switches because otherwise, a theoretical 8-way PCIe bifurcation (which to my knowledge no motherboard supports) would cause poor drive speeds since they are given x2 bandwith instead of intended x4 lanes. So yeah, 8 M.2 to pcie RAID cards... you are now flying close to 4 digit price tags and companies that offer serious server/workststion solutions like HighPoint, OWC and Sonnet. This is probably the most extreme of M.2 expansion solutions money can buy.
@@CsQ_RandomRepository I guess it must be expensive to build those cards with PCIe switches, or.. someone is just making profit :)
Thank you mate, always wondered this! Great video
personally i am getting there. 3 u.2 drives, good gpu, dedicated 4x capture and recording cards, 10gig etc... i'm done with desktop setups. thanks for the great vid!
This is why I'm still running an X299 system.
It's not expensive like a Xeon, but has way more PCIE lanes. I'm running an i9-7960X so it has 44 PCIe lanes.
I do photogrammetry that uses CUDA cores to accelerate the calculations to make 3D models from photos.
So I run a 3070 Ti on my primary slot, and two GTX-690 cards on other slots.
The GTX-690 is a dual GTX-680.
So I have 5 GPU cores in my system and the two 690's have a total of the same number of CUDA cores as my 3070 Ti.
Currently the first 690 is on the second X16 slot and the third is on the X4 slot. That's ok because I don't really need it to be running X16 on the second and third card.
I also have a Tesla K80 datacenter GPU that I'm hoping to add as well.
The problem is drivers. Windows doesn't like using multiple GPU drivers.
To get the 3070 Ti and the 690s to all work I had to go back to nVidia driver version 471.41. That was the last driver that supported the 30 series AND the older 690's.
Once I upgrade to a 40 series I probably won't be able to use the 690's anymore.
I MIGHT be able to use the Tesla K80 though since it's a completely different kind of card. Not sure.
I would probably be ok just using the chipset to feed the second and third cards on a newer system, but it's kind of nice knowing I have all that bandwidth available.
The only reason I didn't put the third card on the last X16 slot (which is actually only an X8 electrically) is simply because the riser cables are too fat and I ran out of room.
Beautifully explained, wonderfully taught, and simply illustrated. Amazing job, amazing videos, keep it up brother.
The Z790 Aero G can only do x16 to directly to CPU on PCIe 5, 4x on PCIe 4 to the CPU (M2A_CPU m.2 slot). There are actually two m.2 slots that hit the CPU, one is the PCIe 4 one, other other is a PCIe 5 one. But if you use the PCIe 5 one (M2C_CPU), then the first PCIe slot operates at 8x only, wasting even more lanes :/
I always questioning sometimes how absolutely weird Intel PCIE layout can be.
Just look at those mobo that has two PCIE x16 that only apart by one slot. Like what the heck they even supposedly be doing
For the description of the generations and lanes and the analogy of cars and highways. For each generation I would have used the example that the speed limit is doubled each time as the lanes themselves are quite literally physical.
Oh one thing you forgot to mention is that a PCIe Gen 4 GPU will switch down the Gen speed when it is not used in power saving mode. This frees out the lanes for M,2 SSDs to run fast if they had the x4 lane speed configured. Also the CHipset has a PCIe bridge function that can multiplex more physical lanes but they dynamically share the available lane bandwidth. Latency included as well due to the extra hop to the CPU controllers. Rule of thumb in p[ower consumption is the higher the Gen type, the more power they consume!.
Had to come and reply to this as I was scrolling comments, what? What platforms are you operating on when a 5.0 PCIe lane is switching down to another lane? It isn't a software or hardware switch like "power saving" mode in your basic Windows settings. PCIe slots do not switch down bandwidth generations. The generation is built into the CPU itself paired with the chipset. They're backwards compatible, yes, but... it doesn't "downgrade" itself to 5.0 to standard PCIe 2.0/3.0/4.... Lol.
Very intuitive and informative as well
Thanks for an easy to explanation video for us tech builders - much appreciated.
As it is getting into how many pcie lanes you have you also have to look at where is the pcie signaling clock as plugging in pcie 3 into pcie 5 then pcie 4 into pcie 5 then pcie 5 into pcie 5 if all 3 slots are on the same signaling clock then everything will run at pcie 3 speeds. It mostly is fixed now by having each slot on a different signaling clock but on same boards bifurcation will allow you to have more slots though less lanes per slot they will also all run at the same pcie generation.
The way the ASUS Pro WS W790-E got around it is the CPU has enough lanes that everything has their own pcie 5.0 lanes from the CPU and thus no bifurcation. Now it is a 1300USD board and the CPU for that to be so is on the Xeon W-3400 series so starting at 1700USD and goes up from there. It isn't a cheap system but if you need everything to have direct to CPU pcie lanes and enough lanes for everything to not have to share it is going to cost you.
He got how the rear I/O is connected wrong. The rear I/O the video out is directly connected to the CPU, most if not all USB ports are connected to a USB controller then the CPU , the rear I/O NIC to a network controller then to the CPU. The audio is connected to the audio chip then to the CPU.
We have been moving away from connecting to the south bridge as it is slower.
So much Simplicity. Genius!
Love this style of explanation🏡
Great video, I'm building my first gaming pc for vr later this year.
it would be great if you talk about hardware for Unreal Engine for creators.
LGA1700 CPUs (12, 13, 14 gen) only have 16 lanes of pcie 5.0. That's why on MBs that support an m.2 pcie 5.0 slot if you populate that it will automatically switch to x8 for the GPU. So directly from CPU there are 16 lanes gen 5 and 4 lanes of gen 4 (plus additional x8 gen 4 lanes for connection with the chipset).
great explanation
I really think it's time to explain pcie lanes as networking connections, bandwidth, latency, and routing. What DMA is and how some hardware can bypass it altogether with device to device data transfers.
very good this is one of the best explanations that I’ve heard in years. Thank you very much.
Another great video might be how to power a 4090 with the proper PSU. I've been hearing of a lot of people ruining their cards by using an under powered PSU
I was wondering if you have made any best bang for buck videos using the workstation platform?
Great Explanation!!! Thanks
Nice explanation very easy to understand. Was confused about pcie for a while
One thing to be noticed, the second-to-the-last PCIE slot on Asus Sage are PCIE 5.0 x8, which the remaining 8 lanes are used for 2 m.2 slots on the right side of board.
Thank you so clear explanation
Can you next explain motherboards and how to choose a motherboard what expansions integrated or slots to look out for plz
Very useful video..... Learned a lot..... Thanks!!!
Also memory channels are important. Threadrippers and Xeons use 8 of them. It would be nice to understand if r-dimm memory are better for these motherboards and check speed vs unbuffered dimms. DDR4 Vs DDR5 would be a nice comparison.
R-Dimm or registered ECC memory does error correcting (checks the integrity of the data) where U-Dimm doesn't - basically if you really care about your data you run a system with R-Dimm ECC memory.
There is no performance benefit with R-Dimm, most of them run a bit slower than your typical consumer U-Dimms
Actually 4 channels 2slots per channel
You deserve more subs all
Your videos are clearly explained.
Thank u
Versed in PCIe lanes, I clicked, as your perspective differs from the pure gamer.
In the beginning, there was an ommission--the 13900k has 20 gen 5.0 lanes and 4 gen 4.0 lanes. This allows 16 5.0 lanes, bifurcation aside, and 4 lanes for an NVMe or some other peripheral directly connected to the CPU, as various adapters are available to turn it into a standard PCIe 4x 4.0 slot.
I don't fool with AMD--too many headaches. AMD and Intel do not report lanes equivalently. Saying the AMD with 24 compared with 20 for the Intel 12th and 13th generations is incorrect. Intel reports accessible lanes. With Z590 with Rocket Lake CPUs, Z690, and Z790 have 20 user accessible lanes. There are other lanes present, 8x, IIRC, to the PCH for other built-in peripherals such as LAN, wireless LAN, USB, Thunderbolt, SATA, or other NVMe. AMD reports total lanes, accessible or not.
i've actually been looking at one of these CPUs for a future workstation at home. I'm not a creator or 3D renderer but I might have to render some things for work. Most of what I'll do will consist of AI and in general HPC which will take a lot of processing power and VRAM
So i'm thinking of getting a Lenovo ThinkStation P620 with a 32 cores 5975wx CPU, 128 GB of RAM that i'll probably expand later and a dual A6000 GPU combo. Most likely gonna dual boot too to have Windows for anything video related and Linux for anything coding related
You explanations are magnificent!!! You should have been teacher/professor. I wish my teachers back than had your skills of explanation....
Why I am edging toward threadripper. Not because cores but lanes.
Lower number of cores to 12, make it the F version, then much lower tdp too. Great for desktop that needs lanes not cores.
How do you use cables to prevent the GPU's from covering up the lanes? How do you power two 4090's? It would be cool to see a video where you build a workstation with two GPU's. Thanks for the great info in these types of videos
some motherboards have larger spacing between pcie ports
Liquid cooling
Very clear explanation. The best video in the inet!
WOW very nice and clear explain about all staff ,thanks a lot
I'm rebuilding my NAS with an Epyc CPU/mobo for this exact reason. Could have spend less on consumer stuff but I need 10gb LAN, HBA and a 4x NVME SSD card. With all those thrown in I still have PCIE slots left for a GPU if I need it.
You know they make consumer boards that already have 10GBe included. Sometime people think they need overkill, when it's rarely the case. As something you'd be using strictly as NAS, with a single 10Gbe you can't even use the full bandwidth of a single gen 3 NVME drive. You only gain a large pool of flash with a 4x card, that you'd never get the full bandwidth potential out of. 10Gbe translates to 1.25GBps, which is the absolute limit of getting data in and out of it. My 4 7200RPM RAIDz1 drives already max at 800Mbps, just with the onboard SATA. Doubling that to 8 with an HBA ultimately gets limited by the NIC. That HBA could run off of the 4x chipset slot.
The typical use for a GPU in a NAS is media encoding, and that doesn't even demand the GPU be allotted all 16 lanes. The $500 Asus X670e Pro Art has 10Gbe (and TB4), 3 Nvme slots, ECC support, and CPUs with integrated graphics. Sure the iGPU would only be useful for the setup until media servers are actually willing to address its media encoder, but a 6 core 12 thread CPU is only $240 on the platform. More than enough for a dedicated NAS. I've been running truenas in a VM with 4 threads and 8GB of RAM for 2 years, in a PC otherwise used for more virtualization and gaming. You should be doing way more with it to need Epyc.
@@blkspade23brought bro back to reality
well done video!
for illustration purposes, I'd rather recommend using trains instead of cars in way like this:
motherboard is a town,
cpu is main train station,
chipset is secondary train station,
each lane is a railway line,
PCIe generation is amount of wagons a train can deliver
;)
If your analogy of pcie gen is widening of the hi-way, then what about pcie lanes 1x, 4x, 8x,16x ?
PCIe 6 spec was established years ago. In 2022 in fact. Heck, PCIe 5 was released in 2019...
Why is the Chip-Set type and features only emphasized with PC Motherboards?
When I look at Laptops, I might see if they support Thunderbolt 4, and such, but they never mention the Chip-Set, like they do with Motherboards.
This video is very under-rated! I am now looking at my MSI MEG X570S ACE MAX like it's a child lol now I feel the need to move up to a ThreadRipper/Workstation because I am bottlenecked with my 4 m.2 NvME Drives x 2 M.2 running in RAID. Its time for me to upgrade anyways as I am on Gen 3/4 components.
Great visual reference for the PCI lanes! Now whenever I say or hear it I'm going to see Tron Legacy!😅
Very well explained sir. Thank You.
I see now that the difference between Server Xeon processor and Workstation Xeon processors are the number of pcie lanes (server platinum xeon 80 lanes, xeon w-2400 64 lanes for workstation, xeon x-3400 112 lanes for workstation). But with a motherboard that is designed for workstation purpose but accepts the Xeon Server (which can be found cheaply than workstation Xeons) we get better performance at lower cost.
Hey Sarge, it seems that even with the best consumer level stuff if you plan to do any serious 4k+ 6k+ workflows, rendering etc. One should probably look at saving$ & going straight for a workstation type setup ? I know that sends the budget into the stratosphere. With YOUR HELP (thank you)...I have built a nice I91300k w/Asus 790mb Intel ArcA770 16 gb card...but do get hitching, bottlenecks, rubberbanding from time to time working with 5.3k off the go pros etc. It is a huge improvement working on the rig compared to the 8 year old laptop I had..
Very well explained, love it ❤
What about laptop's lanes? for example Ryzen 7 5800H?
hi can you make video pc build with intel xeon w 3400 😄
Good presentation!
Ryzen 9 having 24 lanes only and dual channel config is straight criminal!
Those who game are fine with their R5-R7 but R9 is meant to be a creator CPU and those specs are laughable..
Most comments: want more lanes? move to Threadripper or EPYC..
Dual channel means?
Thanks for the video!
Is the reason that the Thread Ripper and Xeon processors have more pins because of all the PCIe and memory lanes. Meaning that we won't get anymore PCIe lanes until CPU manufacturers change the CPU socket and at the moment they won't do that.
I was hoping to get a question answered by this video, but I’m afraid it didn’t. I have a 5900x on a phantom gaming x570 MoBo and I’m running a 3070 and 2 NVMe drives, am I short on lanes? If so, what should I be experiencing, such as slowing down somewhere? This is really my first PC and I don’t quite understand the fine aspects of a PC. I know my CPU has 20 lanes, but then he confused me by saying MoBos can alleviate bottlenecks some how with their chipsets and I got confused.
When the CPU has inbuilt graphics, then is the PCI free lanes used by other peripherals or those 16 lanes are still kept for graphics card?
Why consumer cpu have so low amount of pcie lanes? What if I want a second gpu? Or add some other stuff
It’s not clear to me whether or not the chipset acts as a traffic-cop when dealing with all the extra peripherals attached to it. Can someone please clarify?
I use a videocard that runs at X8 on my X570 ProArt with the 5900X - so I can have all my NVMEs running at X4 speeds .
Well that was disappointing. Very oversimplified and at times kind or wrong.
Intel 13gen cpu and x670 cpu have 28 lanes worth of bandwidth which are differently allocated. I am really missing information about bifurcation, how much bandwidth a gpu, a capture card, network or ssd storage really need. Which is much less that most people think. Most gpus are fine with 8 gen 4 lanes, 10gbe need technically 1 gen3 lane and so on
so lets say i use a x670e motherboard and want to use 2 gpus. 1st gpu is handled by the cpu and the 2nd gpu is handled by the chipset?
Let's say a gpu uses 16 Gen 5's. Can you maybe the just use 32 Gen 4's to get the same amount of "highway" since Gen 4 is half the size? Or does it not work like that?
how motherboard know which Gen is slot? example B450 is Gen3, but if we put 5000 series cpu, will GPU rtx4090 work in Gen3 standard or Gen4?? Because CPU support gen4??? lanes go directly to CPU. How that work??
Nice explanation!. Thank you!!🤙
Loads of tests have shown. The difference in pie 3.0 and 4.0 as far as GPU is minimal if not none.
Informative video 🎉🎉
how can i tell whast is using my lanes? for example ..after the GPU what is going on the remaining CPU lanes and what is using the chipset lanes.. can devices be assigned CPU lanes specifically? Assuming CPU lanes provide better data tranfers.. lets say i buy a faster m.2 and a slower one.. can i assign the faster m.2 to the remaining CPU lanes and leave the slower one to be used my the chipset? Thank you!!
question about M2 drives~ I'm building an AMD 7700x pc with ASRock B650 MB, it has a gen 5 pcie slot, also a gen 4 & gen 3 slots, but only the gen 5 has a heat sink... I have a 1 tb m2.0 gen 3, and getting a new one gen 4, which slots should i use for both?
But dont GPUs operate in the 16x pcie slot at only 8x band width?
I didn't understand one thing, does anything physically change between slot types?
Here's a question if I put in an expansion card for more storage and force my 4090 to run on 8 gen 5 lanes will it still work just as well? Since it normally runs on 16 gen 4 lanes.
Does the CPU have dedicated lanes to communicate with the chipset?
The GPUs are taking too much space on the motherboards. Could you show external connectors that will maximize the slots without loosing speed?
I have a question...if I have a 7900XT running on a PCIe 3.0 board, how much am I actually losing in performance without 4.0 available? I have a 9900K so it's probably getting time to upgrade soon, maybe later this year, but I'm just curious. I have 3 M.2's populated on the board too, which probably doesn't help lol
Nicely Done!👍
Would an high end AM5 APU benefit from having gen 5.0 pci-e lanes instead as AMD planned to have it 4.0, when it's putted on gen 5.0 capable motherboard ?
Now say your GPU had 16GB memory. Then 8 lanes of PCIe 4.0 can fill it in one second as it has speed of 15.754 GB/s. That is the magnitude of things.
To comparison typical AMD DDR5 memory bandwidth is 70GB/s.
trx50 amd 7990x 5,1ghz 96 core 192thred ddr5 qua canal
I still have 20 of gen 3
Do you know ANY motherboard that can do PCIe bifurcation (x4x4x4x4) for 2 of their x16 pcie slots?
thinking...
16 lanes of pcie (x4x4x4x4) - first card carrying m.2's
16 lanes of pcie (x4x4x4x4) - second card carrying m.2's
4 lanes of pcie - m.2 boot device
1 lane of pcie - video card
3 lanes of pcie- other motherboard things/NIC etc
===
40 lanes. Haswell XEON? 2650Lv3? or similar?
so, 8x 4tb m.2 drives with 2 redundant ones.. = 6 drives worth of storage. = 24tb?
Is this a reasonable thing? and also, is it a possible thing
Very well described 👍