I was introduced to this channel a few months ago and I've found myself continuing to come back for more. The interviews are truly the bread and butter of this channel, the breadth and depth of the questions asked are obviously thought about beforehand, and it leads to an excellent video. Keep it up!
Journalism has moved away from asking key people good questions and listening to what they say. Simple formula, but it works. Gordon's great because he still does that, without any agenda to push (other than maybe to push back against "the internet" 😁). It's not as common with newer tech journalism, though Gamers Nexus and a few others are starting to do it, thankfully.
Patriot is underrated. They produced some really good DDR3 and 4 kits in the past for an affordable price. The 4000+ Viper B-Die for DDR4 or higher-end Sector 5 DDR3 PSC kits in particular come to mind.
When I started my recent build, I bought the CPU and mobo as part of a bundle first. Then I checked the QVL for the mobo and bought my RAM off that list. That worked out really well for me :)
I'd always recommend to check the QVL list too. If you find a RAM kit that's a good value, with decent speed and timing and then check it against the QVL for the MoBo you want and it's not on there, at least you can then find something similar on the list
Good to know Patriot has redundant memory profiles and both XMP and EXPO build in... Will definitely consider them for my next build, as memory timings are a bit daunting for me as a fairly new PC enthusiast
The interview teaches you the most important high level things that you would learn over 3 hours worth of Buildzoid RAM overclocking videos, but only takes 15 minutes. Great interview!
To all enthusiast ram manufacturers: Please can we get the module rank listed as standard? In DDR4 era there wasn't much difference running 1DPC 2R vs 2DPC 1R, but DDR5 seems to have a bigger clock hit from running 2DPC. 1DPC 2R seems to be the better way to go. This info is commonly available for JEDEC modules, so why not on the higher end? I'm not looking for the most extreme ram and want to turn on XMP and done, but 2R modules does make enough of a performance difference to seek out vs 1R modules.
they want to retain the ability to swap bins based on availibility but dont want the informed people to be able to make a quick reference to their cheap choices. a standard in marketing and spec disclosure would be nice, agreed.
Most DDR5 16GB & 24GB sticks are single rank and 32GB & 48GB sticks are dual rank. But yes showing how many ranks AND die type on memory packaging would be extremely helpful.
@@n8spL8 Yup, I fell for that before. I have an early Ripjaws V kit 3200C16 2x8GB that was 2R. Performed great. Bought another kit later. It moved to 1R, and performance tanked. At DDR4 8GB the standard is 1R for a very long time, and worse 1Rx16 exists as I found when getting a laptop a couple years back. It's the same deception going on with some SSDs. Same marketing name, but could be very different hardware.
@@corsairsloop3234 For now, but it can and will change over time. Early DDR4 8GB could be 2R but it moved to 1R not long after. 16GB were 2R for a long time, but also moved towards 1R. Absent specs to say otherwise, the only safe way to get 2R DDR4 might be going for 32GB modules.
Patriot memory sure comes across as genuine. I have heard about their kits before, I have a G.Skill B die kit in my AM4 system running custom timings as fast as they can run stable at 1.45v (3733MHz CL14 / tuned sub timings) but I know that it can run higher speeds since I ran it on Intel 10900KF before at 4266MHz. Ended up having to turn it down a notch with the 5800X3D which is fine due to the insane L3 cache available and I'm happy I made the switch. Also like Crucial a lot. I don't know which memory is being used by Kingston but at my day job which is mostly PC Boutique and solving ICT end user problems / sale stuff I go with Kingston Fury Beast memory but that's just XMP profiles and it just works on ANY motherboard which is quite impressive. So keep in mind that if you build systems for people you want a safe bet, but if it's your personal PC you can go a bit crazy and mess with custom timings at insane speeds if you like! So Patriot memory might be something I look at in the future when I upgrade my platform with DDR5 in a couple of years. But 5800X3d is still very fast in my PC @ 4K even with a 4090 that I UV at 0.945V with stock performance still. Only runs at 4080 wattage or less depending on the game which is kind of insane. But I always turn off raytracing since it's a huge fps drop and it hammers the CPU. (running custom water cooling loop but a fan on top of the gpu aimed at memory is always an option)
@@rjScubaSki I bought a kingstom fury beast kit 5600 cl36 kit and it runs stable at 6400 cl30, and that just blows my mind. Never had memory thats been so flexible before.
This guy is ON POINT. I *_REALLY_* wish some some mainboard manufacturers would offer *_cheaper,_* two-DIMM boards. I've tried FOUR Z790 boards and they were all limited to 7200MHz DDR5
I want a Rosetta Stone for DRAM overclocking because I'm just too afraid to do it. There's so many settings that I have no idea what they do, and motherboard setting names aren't consistent. Then there's offset values vs total values which can confuse people into frying their hardware. DRAM overclocking is so weird it makes it hard to want to play with it.
I went with AMD AM5 this time around as Intel CPU I currently have has lost a core and never really lived up to the hype of the CPU. Plus the energy consumption is lower on the AMD. AM5. EXPO was a huge head ache when trying to buy the DDR5 ram. Ended up going with a Corsair 6000MHZ at 64 GB at 32GBx 2 wanted 6400MHZ, but could not find any info about compatibility. I looked at Patriot and many other companies DDR5 but wanted a certain look that would work with my Corsair ICUE setup for my keyboard/ mouse and headset.
I buy highend hardware because I can crank extra perf out at its end life and enjoy a little extra power during its life. I like to build a system roughly every 8-9 years
I needed this video 3 days ago. I just purchased Ram for my new AMD 7950x3d with Asus Crosshair extreme build. I picked up 2 modules of Corsair Dominator 64GB @ 32GB by 2 PN CMT64GX5M2B6000Z40 I really wanted to go with the 6400MHz verses the 6000MHz but the QVL only listed 1 6400MHZ and support in the Forum said to stay away from it as expo is not there yet.
EXPO and AGESA is getting better... That being said, a good EXPO profile can really even be a good starting point to manually tune your memory in to run how you want them. I expect as a few more UEFI/AGESA updates come through you will see the memory compatibility and performance improve. Much like AM4, the memory kept scaling (to a point) once they started to get the updates figured out. Just be aware that AMD IMC normally for most users is most comfortable and compatible with a tighter 6000 - 6200 kit, rather than trying to force 6400, as that can be problematic for many chips/setups.
@@punx223 That is why I settled with the 6000MHz DDR5. Stability over over clock ability. I can always replace the 6000MHz DDR5 in the future. Because I both game and do design work for my personal woodshop I needed stable setup.
He hasn't convinced me. He's taking high speeds and how lucky he was testing 7800-8000 on 13th gen. You can get memory running. But what actual testing he's doing - is it stable? Anything above 7200 on 13th gen requires fine tuning and may not work just because of weak memory controller.
We have run HCI Memtest, and YCruncher for over 72hrs without fault on 13900k/790 Apex with our 8000s with no custom tuning other than setting the XMP profile. I dunno about luck in binning, that was more said tongue in cheek as I tested a ton of CPU, and the amount I have see. That cannot hit 8000 was a small percentage, like single digit. I will say that when you get to these higher speeds, pcb quality, IMC quality, literally everything can impact the capability.
In my first build a few months ago I first bought the MSi B450-A Max Pro Motherboard since it was inexpensive and had big coolers over the mosfets. Then I aimed at the CPU for the 5950X but since it was to expensive I got the 5900X instead. Then I looked for the most inexpensive AM4 RAM but decent one and I first got an pair of 2 x 16 GB 3200 Mhz and when I assembled the PC it worked immediately flawless even when not in the QVL I later ordered anothe same pair for now 64 GB of RAM. I just reaett the speed in the Bios to XMP off and sticked them in. It ran without any problem. Then I put the XMP 2 profile again on and it works flawless. When I opened the case to stick the 2 second sticks in on an just switched off PC I noticed the RAM coolers where noticeable warm. My RAM is Teamgroup Expert 3200 Mhz. I recommend that brand and this particular RAM model. It just works as well with Cinebench full run. The 5900X however has to be undervolted to 0.1 Volts since it tends to run allways close to 66 degrees Celsius which keeps thw case fans running all the time annoyingly. I never would want an 7000 series CPU which runs near 95 Celsius on standby. All your case fans will run allways. Undervolting keeps the 5900X between 50 and 56 degrees Celsius on normal tasks. Just go for Teamgroup as RAM. It just works and is more economic while claiming more durable.
I always say, the only dumb question is the one you don't ask. Always take the chance to educate yourself further. I don't know everything and always open to learn
B650 Riptide, DDR5 6000 CL30. I had no issue getting it to train memory. Where I am uncertain is, Ryzen 9 7900. It's a 65watt chip but MSI shows it pushing 90 watts all the time. I don't know if MSI is off or if my chip is pushing that much past tdp, or if it's something I even need to pay attention to.
i wouldnt worry abot it, it is likely the opportunistic boost (Precision boost) at work. Keep an eye on thermals, as long as you are well within thermal limits, getting good performance, and happy with the cooler/fan noise.... let it run. TL:DR the opportunistic boost feature regularly oversteps the rated TDP value to use extra power/thermal headroom for better boosting/performance. IF you truly want to lock it down you can likely disable this activity but it would be a solid detriment to performance.
Shannon, I am buying an Asrock B650m HDV/M.2 motherboard (which only has two dimm slots) and a Ryzen 7700 (non-x) CPU. I want to get a 64GB 5600MT/s ram kit but, in ya'lls QVL the only 64GB kits listed for Asrock motherboards are your 4800MT/s "signature series" kits. Even for the X670E Taichi motherboard. I really want to get a Venom non-rgb kit to put into the HDV mobo. Is there a problem with running a Venom 64GB kit at 5600MT/s on AM5?
I will be honest, you can try it and likely it may be ok with newest UEFI updates, but being the current state of AMD, I can tell you that the density, same as slot population puts extra stress on the memory controller in most scenarios. The only thing I can say is you can try it, and in my experience as the UEFI matures it should be possible. However right now as I'm sure you have seen there have been massive changes to AGESA and their usage/limits to vSOC, and therefore I think you will need a few patch/updates for everything to settle. Worst case, for now, enable the profile, if not stable, manually turn the speed down to 4800 and with each succession in UEFI update you can test again, as these small update/steps will likely see more performant RAM results.
Ram is the only part I did upgrade in my system. Ignoring additional storage. Also the free graphics card I got, but that was a downgrade for some games and workloads.
What platform and memory did you upgrade to? Sometimes chasing speed especially for gaming can end up being a losing effort, or at the most a marginal/menial result. You really do have to find the right spec to match your platform and usage.
08:10 - When Should You Buy RAM? When you just find the cheaper mb and ram combo that will suit your needs. All the money you save you'll have to spend in other things, no need to pay a fortune for 8000 ram than performing at max 2.5x faster than ddr4 3200 will have a first read latency (CAS) so much worse than 10 ns (+20 to 50% at ease) than PCIe or cpu BW limitation will make become a 2.5 times faster a 2.5 to 25% max FPS increase...don't waste your money
Samsung B die and DDR5 need cooling to be performant. The performance increase on lows is honestly underreported, especially with Ryzen systems. That being said, if you can run DDR4 3800C14, it's much less of a headache than DDR5 tuning. I'll take the possible 5% loss of performance from top end DDR4 to 7200mts+ DDR5 for the days of headache it is to tune.... Shoutout to Patriot Steel Vipers. Best value SR B-die kits in Canada.
Yeah, Patriot Viper Steel DDR4 kits with B-die are the best value. Unfortunately for me, I could not get my kit to complete stress tests without a fan on the memory. They got too hot. I decided to do surgery on one stick to replace the thermal pads on the modules and replace the foam on the back with a thermal pad and destroyed the stick when my PC failed to post. I took the loss as a great learning experience.
Patriot Viper Steel are great sticks and some great B-Die with the 4400s being pretty sick. Much like the Xtreme 5 though it's gonna be rough on some IMC/mobo
Hi Gordon and thank you very much for this video on rams and l also wouldlike to thank Shannon Robb l never heard his name before but after watching the video he is a cool DUDE and really love the way how honest coments he made so Gordon thank you so much if it wasn't for you we would never heard proper educational video like this about ram! keep up the good work!👍👍👍
I know you touched bases on DDR5 temperatures but I am curious. I am running overclocked Hynix dual ranked DDR5 in my AM5 build that is the stock 2x32GB 6000 CL30-40-40 kit. For the most part they are under 40C but when I'm really using my PC, HWInfo64 will show the SPD hub temperature reach 64C for a moment. Which is odd because often times it's just one of the two sticks and if its not reaching 64C, it mainly stays under 50C. There doesn't seem to be really an in between. Is there a cause behind this or is this too much for my sticks in any way? I don't seem to be having any issues with performance in any way but I've not seen ram ever reach this high of temperatures before even if it is only for a second. Heck for all I know it's a HWInfo64 fluke.
My 7950X3D runs best at 6400 Mhz (G.SKILL Trident Z5 Neo - 2x32GB). I tried 6000 and 6200 with better subtimings, but nothing that ran stable was faster than the 6400 Mhz settings, so I left it like that...
That's a solid setup, but some small tweaks to vSOC may help with tightening up your sticks, but that would take some work... As long as you have a good FCLK/DRAM/UCLK you will be in a good place either way, even if leaving a modicum of latency on the table
00:36 - 2 DIMM vs 4 DIMM gamer/PC versus Dual, Quad, Octa, Dodeca channel memory Bandwidth o server/workstation, 128 bit vs 768 bit, 25 GB/s DDR4 3200 (10 ns CAS latency) Dual, 38 GB/s DDR5 4800 (15 ns CAS latency), 28 GB/s DDR4 3600 (cl16 10 ns CAS), 50 GB/s DDR5 6000 (cl28 10 ns CAS), just cpu limited to 50 to 100 GB for AM4/AM5/1700 sockets cpus... vs server 200GB to 460 GB/s...
First time i tried memory ocerclocking i did my research, i saw that AMD's APU like the 4750G and 5700g had an IMC on TSMC 7nm and was generally pretty good compared to AMD CPUs and Intel CPUs, then i got a cheap ITX 2 slot board, and then some single rank samsung bdye rated for 4400C19 (super cheap, under $100) Within probably 2 hours i had that up to 4933C17-17-17-28 without changing anything on the CPU other than manually setting IMC, FCLK, and UCLK to 2466 and disabling gear-down for ~$500 i had a system running faster, with lower latency, than any DDR5 kit at the time, including those $1000+ kits Its just too bad the 5700G only has PCIE 3.0, and has 1/2 the cache per core compared to a 5700X. Sure in some use cases that memory did make my 5700G faster than a 5800X3D, but in most cases it only slightly bested the 5800x and sometimes lost to the 5700x With one of the 5700G i was able to eventually hit 5000C16, but the 4750G(the only one i have) was only ever able to get 4933C18, or 5000C22, maybe it was related to the added out-of-band-management, maybe it was related to having 2 CCXes instead of a single unified CCX/CCD
On the subject of heat, yes, very much so, after i put my best 5700G into a micro PC with only that noctua horizontal cooler with a slim fan, it dropped to not being able to even hit 4933, instead only acceptably stable at 4600C18, still good, but nowhere near 5000C16
@@foxdart Part of it was accudentally reducing the time betweem memory refreshes or something along that line, making it slower than it could have been at the same spec, but it was still faster than DDR5 at ~75GB/s and 49ns of latency after a chat with buildzoid probably 2 months later i found that with proper config i might have been able to get down into the low 40 upper 30ns if i had not reduced one of the sub timings. It was... surprizing for me to achive that first try, but buildzoid clarified that it probably would not have been stable had i not reduced that refesh delay or whatever it was. Now we have kits of DDR5 that are near or above 100GB/s and some down to below 45ns, though often not in the same kit
what about timings, i think it's way important than the speeds but almost no one talks about it except AHOC x HW unboxed getting results with almost no perf increase from 6000 to 7200 mhz while getting ~ 10% boost from tuning the timings
Depends on your goals, and totally agree. Those are not mutually exclusive, there are workloads where high speed will deliver better results than a lower speed tight configuration. However I also agree that super tight kits in the 6400 - 7000 range can be rockstars if you have a good board/IMC combination. If just gaming, crazy high memory clocks will not benefit much, there's a tipping point where your speed stops delivering noticeable improvements. Alot of times it depends o. Hardware combination and workload.
We have run our 8000s on HCI for 72hr, and YCruncher among other stability, number crunching and stress tests before evr validating them to release. We even test some workstation applications under heavy burn loading just to test them at actual work. On cpu render, etc.
I found out the hard way i got a 7600mt 32gb kit and my 13900k or z790 aorus elite mb can only get 6800mt stable with no crashing. I could have got a lot cheaper kit and had the same results. MY kit cost £500 4 monts ago
That's pretty rare, which bios version? I've had the aorus elite to over 7200 quite easily. I mean it's possible IMC may just be kinda bad but that is quite low from my experience.
@@punx223 buildzoid did a full breakdown of the aorus elite and found that 7200mt would be the max stable he could get because it's got 4 dimms and it's only a 8 layer board. plus it looks like i lost the lotto on the 13900k my chip only has 92 biscuit score which is pretty bad. my setup maxes out my alienware 34" oled ultrawide so its not a big deal. just a waste of money
@@benprice1984 I have seen his coverage, but BZ seems to not be able to get stable above 7000/7200 much at all, which is very different than my experience. If I get the opportunity to work with him, I will be happy to explore it and see if there's something weird/amiss I can help sort for him, or he just has some of the worst luck in 13th gen silicon I have seen to date. BZ seems like a great dude, but our paths have not crossed thus far, but I am always open to work with people and help debug/tweak to find whats the root cause of issues.
Just like stickers on cars my friend, every sticker can add "up to" 10HP, with computers though you have to decide on whether the RGB adds better latency or speed.... choose wisely.
14:45 - Does RGB Make RAM Go Faster? MAtching your RGB with the RAL colour code of your case or hardware parts, may go faster the light so much you don't see any difference. IN fact who cares that a 13th gen INtel CPU memory bus is limited to 78.6 GB/s even for a Celeron or Pentium from 12 th gen because you can overclockit using 8000 faster expensivest ram to 128 GB/s, while AM5 only doing 78 to 100 GB/s and INtel DDR4 dual channel 4400 only goes 2x34GB/=68s but losing a lot of CAS latency, so 3200 or 3600 for AMD and 4000 for INTel only while cas latency goes below 10 ns.
I have seen alot of his coverage on this, and I'm open to discussing and working with him. I just honestly have not had the chance to do so. That being said, my experience has definitely been wildly different than his. No way to know why without alot of data, but I would still love to sink my teeth in and try and help, this is what I do.
Just to be clear my main rig is running a 13900k, retail from Microcenter. 4-DIMM board gigabyte Z690 waterforce and it runs 2x16GB 7400 24/7 since I've built it.
I have a feeling that there are differing standards of stability. Sure, it can run 7400. But does it fail in certain corner cases? He was like "one error after eight hours".
@@punx223 It runs and maybe even stable, but the “7400 is easily” is the thing. I believe he says that 6800 is the safe speed to run without headaches (XMP). His is a sample of I believe one 13900k and a 13600k so there is that. I think he is taking it that his 13900k is the worst bin there is. I am just perspective buyer and don’t have much experience with any of this. I would fall into the “Run XMP now, and if a someone like him could give me some safe, better than XMP subtimings I would start there and do some testing but probably not 12 hours worth of it, and likely with whatever I could do it for free with. I have only used OCCT and the old testmem that makes you open line 10 instances.
@@TheGreenReaper Also 4 DIMM or no, isn’t a water force board really good? Maybe it doesn’t matter but is it a 8 layer or 10 layer, and does the water part cool the Ram?
The question the manufacturers are too scared to answer is why they refuse to validate ECC kits for OC. Gamers want ECC, too, and not just at DDR5-4800
Not Afraid to answer this at all, and honestly the two are kinda at odds most times. TBH DDR5 has a minor hardware level of what I call "passive ECC" but in reality ECC has always been made for absolute stability. ECC hardware on DIMMS (Like real ECC/server level) can add induced latency as it has hardware on board checking for any bit flips to avoid data corruption or stability issues. In many circumstances these ECC modules are strictly controlled on their signaling/timing, and therefor allowing clock speed modulation/overclocking can cause chaos for the IMC/memory system as a whole. This is not to say it is impossible but largely improbable. Also, the platforms have to have ECC enabled which most mainstream/enthusiasts desktop platforms they do not, so that topic would first need to be addressed on the platform level.
@@punx223 Literally all modern Ryzen CPUs come with ECC support, lol. My small server is running a Ryzen 5700X with ECC UDIMMs. The chips on ECC memory are ostensibly the same as on non-ECC chips, they just get extra chips for parity. All manufacturers would really need to do is evaluate the individual chip's overclocking potential, bin the chips accordingly, and then place them onto an ECC DIMM. But they don't. Because the server market needs to be protected at all costs, I guess.
@@insu_na yes I am aware Ryzen and TR offer it, but my explanation still stands that overclocking and ECC run into each other. It's not as simple as "test the dies" I mean, yes we could probably (cannot say for sure) setup an XMP SKU with ECC... But once you were to try and tweak or tune it the timings can fall out of sync or cause errata, and that's why nobody really does OC ECC DIMMs. Again, that being said, I'm not saying it will never happen but there would need to be alot more demand, and the underlying tech would need some work to make it scalable. Sorry, that's just reality.
Don't you get used to random crashes and having to act as if they not existed or starting a research carreer on maintaining your own PC? They do so user become involved and don't fall sleep and extending your time in front on the display, just how first home computer were entertaining while waiting for the casette tape to load and eventually crash because a random analog read/write error, or a magnet or heat on floppies or an airbone virus that just landed in your partition table... is just for fun and laughts...
DDR5 actually has a low level of ECC within the DIMMS but nothing like you see on server/enterprise targeted modules. ECC is a strong component for scalable datacenter/.server solutions, but as you notice the speeds are tightly controlled. This is because the ECC tech within the modules are sensitive to detect things like bit flips, etc. Overclocking ECC DIMMs is always a weird prospect as you are taking a module designed for very precise signal timings and trying to modulate the signals to give better performance, this as long as ECC has existed has been a tough combination to resolve. It is not purely impossible, but you also must remember an important component of this is also the platform support, and most consumer platforms simply do NOT offer ECC compatibility at all, and wont even boot with ECC modules installed. If we want enterprise level ECC support, it has to start with the CPU/platform level, then from there we would have to see how far they can be pushed before causing false positive errors within the ECC.
@@ericneo2 I totally understand, the issue really comes down to platform support mostly. Not to mention as I have said before, overclocking and ECC can be pretty problematic as overclocking, you are pushing performance and getting really good stability, but even the IMC can become unstable, which is why you notice ECC enabled platforms, or mostly, always run at the JEDEC for the platform.
Like every hobby it pays to not go too overboard. If you're at the point of buying single slot motherboards, 8000 ram and binned processors you should probably stop and take a deep breath.
10:04 - Should You Care About DRAM Die? No because you may realize DDR embedded in Mobile phones, SoDIMM or Apple Silicon it's is much better faster, lower cas latency and lower power usage and made with state of the art Node process technology isntead of the dimms you buy for desktop on older nodes wasting 1 or 2 digits Watts.
Not sure how much SARCASM in your 5 words. Just in case your worried to... @@-eMpTy- 12:52 - Should You Worry About DDR5 Temps? NO. You'll know you have worse problems to worry about once you realize how that metallic smell like ozone or blood is due to vaporization and air ionization and plasma wich started just a little before you heard that first crack. What crack? then a repeated mini fireworks will distract you about the fact you already lost your work or gaming progress...At that moment is when you start to worry about ddr5 temps and how unfortunated is the fact that you can't search a video tutorial on how to deal with your problem...Tranquility begins when you grab your phone...
Great! do you have a part number for 4x32GB and 8x32GB kits? I searched the Patriot store for RDIMM, but all the results were UDIMMs. Also what motherboards are QVL for W5-2465X and W5-3435X?
everyone has different build priorities, and realistically I usually buy all at once, but if I were to take steps, I would likely pick GPU last as prices are far more volatile.
@@punx223 I only do a build once about every 5 years, so I tend to just buy top of the line parts, I did wait to buy my monitor though because I had seen it on sale before and knew it would be again soon
I would have more respect for memory manufacturers if they listed all of the timings instead of hiding most of them because they all want to sell us "speed" instead of trying to teach people how speed for their product actually works.
I think the issue is 'marketability' of memory timings. Unfortunately, consumers are tuned for higher clocks as that's what the CPU vendors also market on. Consumers also understand 'more MHz' and it moves them. While a small subset of tuners will understand and target timing, the vast majority will only focus on the clocks. The good news is most vendors do also provide the module timings in specs too.
I'm here, what would you like to see dude? Our sku sheets for each model of DDR5, show the primaries, which timings are you wanting to see? Unfortunately doing something like a TB dump would not really make sense and even be information overload for many people.
@@punx223 Solution is quite easy. Add full info links next to current ones (or as separate segment, web dev should be able to figure it out) with all XMP/expo timings. I am quite sure you are already aware that even 5000 DDR4 kit can be worse then JEDEC 2666 and that kind of information should be easily available to calculate. Obviously, this kind of info matters since it is relevant for a product and most of the purchases that involve me are for others so I can't OC memory for everyone. We do not get reviews for memory like for CPUs/GPUs, so it would be nice from any company to get out of their way to break "only bare minimum" standard.
I guess I can sum it up like this: Is it worth dumping an existing LGA1200/AM4 system with DDR4 for DDR5? For the most part, the return on investment will be very small and most could not justify it. However, if you are building new, or selling old to build new, the advantage with DDR5 in density and clocks (more RAM, higher clocked RAM) with current DDR5 prices probably means (if you're performance focused) you should indeed "switch." Yes DDR5 has a very real advantage (for example, QuickSync performance with Intel has a very sizeable increase in performance on DDR5 vs. DDR4 due to the bandwidth advantage) but it's going to be situational based on your own personal situation.
Well is you mean the dual rank visibility to OS or overall design, it essentially is somewhat true but its more involved than that. this is why single DIMM DDR5 runs quite well, and you dont take as hard of a perf penalty vs running single DIMM DDR4.
@@facegamefps6725 nice! I have tested in the triple digits, so I have a pretty wide dataset to pull from, while it's not all encompassing, this is my experience between K, KF, KS
That's a good question but I believe Shannon's parameters, like most module vendors, is hitting highest clock speeds while being stable in X, Y and Z metrics. One problem with 'stable' RAM is stable is different to everyone. But this all gets to the problem we have where most of us assume 100% stability in all things in modes are dependent on so many different factors with so many variables. CPU lottery, how well you cool that CPU, motherboard, module selection and even quality of power supply can impact memory stability. Throw in the fact that the hundreds of thousands of different use-case scenarios and I'm not sure we should really expect the same 'stable' use as running JEDEC clocks and timings.
The target of the Xtreme 5 is definitely to try to break records, we are very honest about this. It's similar to why companies enter F1 and spend a ton on R&D. It's because we are performance guys and we want to do it because 1) it's fun and 2) because we can, we wanna show what we can do. If looking for a set of DIMMs just for gaming and whatnot, then there are several models to choose from that are at more reasonable speeds that will run fine as well, as the crazy high end is more for tweaking and tuning, while it can game, it's excessive for that use Aka taking a Ferrari to get the groceries.
I was introduced to this channel a few months ago and I've found myself continuing to come back for more. The interviews are truly the bread and butter of this channel, the breadth and depth of the questions asked are obviously thought about beforehand, and it leads to an excellent video. Keep it up!
Journalism has moved away from asking key people good questions and listening to what they say. Simple formula, but it works. Gordon's great because he still does that, without any agenda to push (other than maybe to push back against "the internet" 😁). It's not as common with newer tech journalism, though Gamers Nexus and a few others are starting to do it, thankfully.
They are pretty rad!
What a nice and knowledgeable guy
You should meet me
@@Akkbar21 Allahu
This guy is great! He knows the stuff in and out! This video clarifies a lot of questions on RAM
🤘
Patriot is underrated. They produced some really good DDR3 and 4 kits in the past for an affordable price. The 4000+ Viper B-Die for DDR4 or higher-end Sector 5 DDR3 PSC kits in particular come to mind.
I am a computer technician and I work more than 15 years, this Ram guy is really honest and not lying to the people. Congratulations 👏👏👏
When I started my recent build, I bought the CPU and mobo as part of a bundle first. Then I checked the QVL for the mobo and bought my RAM off that list. That worked out really well for me :)
QVL is anyways one of the safest ways, so good call
I'd always recommend to check the QVL list too. If you find a RAM kit that's a good value, with decent speed and timing and then check it against the QVL for the MoBo you want and it's not on there, at least you can then find something similar on the list
Good to know Patriot has redundant memory profiles and both XMP and EXPO build in... Will definitely consider them for my next build, as memory timings are a bit daunting for me as a fairly new PC enthusiast
The DRAM crash course I didn’t know I needed. Thanks, guys!
Shannon is great. Gordon thanks for making interviews fun.
🤘
This guy is highly knowledgable and damn awesome 👍
That Shannon Robb guy is a pure natural in from of the camera! Loved his energy!!!
Every RAM Vendors need a Shannon on their team for sure, great interview :D
❤
The interview teaches you the most important high level things that you would learn over 3 hours worth of Buildzoid RAM overclocking videos, but only takes 15 minutes. Great interview!
Wish this guy was my dad instead, knows a lot and is quick to just give it all
These more casual interviews are the best.
To all enthusiast ram manufacturers: Please can we get the module rank listed as standard? In DDR4 era there wasn't much difference running 1DPC 2R vs 2DPC 1R, but DDR5 seems to have a bigger clock hit from running 2DPC. 1DPC 2R seems to be the better way to go. This info is commonly available for JEDEC modules, so why not on the higher end? I'm not looking for the most extreme ram and want to turn on XMP and done, but 2R modules does make enough of a performance difference to seek out vs 1R modules.
they want to retain the ability to swap bins based on availibility but dont want the informed people to be able to make a quick reference to their cheap choices. a standard in marketing and spec disclosure would be nice, agreed.
Most DDR5 16GB & 24GB sticks are single rank and 32GB & 48GB sticks are dual rank. But yes showing how many ranks AND die type on memory packaging would be extremely helpful.
@@n8spL8 Yup, I fell for that before. I have an early Ripjaws V kit 3200C16 2x8GB that was 2R. Performed great. Bought another kit later. It moved to 1R, and performance tanked. At DDR4 8GB the standard is 1R for a very long time, and worse 1Rx16 exists as I found when getting a laptop a couple years back. It's the same deception going on with some SSDs. Same marketing name, but could be very different hardware.
@@corsairsloop3234 For now, but it can and will change over time. Early DDR4 8GB could be 2R but it moved to 1R not long after. 16GB were 2R for a long time, but also moved towards 1R. Absent specs to say otherwise, the only safe way to get 2R DDR4 might be going for 32GB modules.
Patriot memory sure comes across as genuine. I have heard about their kits before, I have a G.Skill B die kit in my AM4 system running custom timings as fast as they can run stable at 1.45v (3733MHz CL14 / tuned sub timings) but I know that it can run higher speeds since I ran it on Intel 10900KF before at 4266MHz. Ended up having to turn it down a notch with the 5800X3D which is fine due to the insane L3 cache available and I'm happy I made the switch. Also like Crucial a lot. I don't know which memory is being used by Kingston but at my day job which is mostly PC Boutique and solving ICT end user problems / sale stuff I go with Kingston Fury Beast memory but that's just XMP profiles and it just works on ANY motherboard which is quite impressive. So keep in mind that if you build systems for people you want a safe bet, but if it's your personal PC you can go a bit crazy and mess with custom timings at insane speeds if you like! So Patriot memory might be something I look at in the future when I upgrade my platform with DDR5 in a couple of years. But 5800X3d is still very fast in my PC @ 4K even with a 4090 that I UV at 0.945V with stock performance still. Only runs at 4080 wattage or less depending on the game which is kind of insane. But I always turn off raytracing since it's a huge fps drop and it hammers the CPU. (running custom water cooling loop but a fan on top of the gpu aimed at memory is always an option)
Kingston is usually Hynix, sometimes Micron. Supposedly never Samsung.
@@rjScubaSki I bought a kingstom fury beast kit 5600 cl36 kit and it runs stable at 6400 cl30, and that just blows my mind. Never had memory thats been so flexible before.
This guy is ON POINT. I *_REALLY_* wish some some mainboard manufacturers would offer *_cheaper,_* two-DIMM boards. I've tried FOUR Z790 boards and they were all limited to 7200MHz DDR5
Which board and cpu?
That is definitely weird
What a fun vibe off you two gents. Would love to see more videos of you both.
I want a Rosetta Stone for DRAM overclocking because I'm just too afraid to do it. There's so many settings that I have no idea what they do, and motherboard setting names aren't consistent. Then there's offset values vs total values which can confuse people into frying their hardware. DRAM overclocking is so weird it makes it hard to want to play with it.
There's plenty of information out there, unfortunately there is not one all encompassing guide, and even I'm afraid to try and attempt one.
Frequency and trefi go high. CL, tRCD and tRP go low. Everything else go auto. And you're set.
So cool to see people thst are pasionate about there work. Dude was great.
Much apreciated, and hope I added some morsel of knowledge for you 🤘
That was so helpful. I’m wanting to go with AM5 but I thought I needed to buy expo certified ram which doesn’t have as good a choice
I went with AMD AM5 this time around as Intel CPU I currently have has lost a core and never really lived up to the hype of the CPU. Plus the energy consumption is lower on the AMD. AM5. EXPO was a huge head ache when trying to buy the DDR5 ram. Ended up going with a Corsair 6000MHZ at 64 GB at 32GBx 2 wanted 6400MHZ, but could not find any info about compatibility. I looked at Patriot and many other companies DDR5 but wanted a certain look that would work with my Corsair ICUE setup for my keyboard/ mouse and headset.
I buy highend hardware because I can crank extra perf out at its end life and enjoy a little extra power during its life. I like to build a system roughly every 8-9 years
I needed this video 3 days ago. I just purchased Ram for my new AMD 7950x3d with Asus Crosshair extreme build. I picked up 2 modules of Corsair Dominator 64GB @ 32GB by 2 PN CMT64GX5M2B6000Z40 I really wanted to go with the 6400MHz verses the 6000MHz but the QVL only listed 1 6400MHZ and support in the Forum said to stay away from it as expo is not there yet.
EXPO and AGESA is getting better... That being said, a good EXPO profile can really even be a good starting point to manually tune your memory in to run how you want them. I expect as a few more UEFI/AGESA updates come through you will see the memory compatibility and performance improve. Much like AM4, the memory kept scaling (to a point) once they started to get the updates figured out. Just be aware that AMD IMC normally for most users is most comfortable and compatible with a tighter 6000 - 6200 kit, rather than trying to force 6400, as that can be problematic for many chips/setups.
@@punx223 That is why I settled with the 6000MHz DDR5. Stability over over clock ability. I can always replace the 6000MHz DDR5 in the future. Because I both game and do design work for my personal woodshop I needed stable setup.
He hasn't convinced me. He's taking high speeds and how lucky he was testing 7800-8000 on 13th gen. You can get memory running. But what actual testing he's doing - is it stable? Anything above 7200 on 13th gen requires fine tuning and may not work just because of weak memory controller.
We have run HCI Memtest, and YCruncher for over 72hrs without fault on 13900k/790 Apex with our 8000s with no custom tuning other than setting the XMP profile.
I dunno about luck in binning, that was more said tongue in cheek as I tested a ton of CPU, and the amount I have see. That cannot hit 8000 was a small percentage, like single digit.
I will say that when you get to these higher speeds, pcb quality, IMC quality, literally everything can impact the capability.
Good deep drive! He mention Apex! Awesome rig!
Love Patriot, got a few of their Viper DDR4 Bdie kits and my old DDR2 kit.
Biggest surprise in this video is that patriot still sells memory.
Yep, we still do
This guy has his brain running in ultra fast timings!
Finely tuned memory and brain can help performance 🤣
that wendell cameo at 4:05 🤣
LOL yea Wendell was hanging out and he, gordon and I were chatting pre/post this shoot.
In my first build a few months ago I first bought the MSi B450-A Max Pro Motherboard since it was inexpensive and had big coolers over the mosfets. Then I aimed at the CPU for the 5950X but since it was to expensive I got the 5900X instead. Then I looked for the most inexpensive AM4 RAM but decent one and I first got an pair of 2 x 16 GB 3200 Mhz and when I assembled the PC it worked immediately flawless even when not in the QVL I later ordered anothe same pair for now 64 GB of RAM. I just reaett the speed in the Bios to XMP off and sticked them in. It ran without any problem. Then I put the XMP 2 profile again on and it works flawless.
When I opened the case to stick the 2 second sticks in on an just switched off PC I noticed the RAM coolers where noticeable warm.
My RAM is Teamgroup Expert 3200 Mhz. I recommend that brand and this particular RAM model. It just works as well with Cinebench full run.
The 5900X however has to be undervolted to 0.1 Volts since it tends to run allways close to 66 degrees Celsius which keeps thw case fans running all the time annoyingly. I never would want an 7000 series CPU which runs near 95 Celsius on standby. All your case fans will run allways. Undervolting keeps the 5900X between 50 and 56 degrees Celsius on normal tasks.
Just go for Teamgroup as RAM. It just works and is more economic while claiming more durable.
Thermal pad and paste conductivity changes can also tune further
If you mean that better paste/pads can lead to better thermals which allows more tuning, then yes totally agree, great point.
slept on channel. Gordon is perfecto
This dude is exploding with knowledge 😅👍
15:48 - Get that man some water!
I'm alive don't worry 😅
I'm never afraid to ask more about tech. That said, I am often a dumbass. Thanks for the info~
I always say, the only dumb question is the one you don't ask.
Always take the chance to educate yourself further. I don't know everything and always open to learn
I have a pair of Viper Gaming (Patriot brand) memory but it doesn't have RGB. I can feel that my computer is slower.
Loved this interview, Shannon is very knowledgable and well-spoken.
Thank you, I try
I will never not appreciate people like this designing and building the tech I use.
😊 Well we love doing it
04:01 - How Much cooler is having memory passive cooler with heatsink?
Very nice to see Mr. Robb, very cool!
Happy to be seen, I hope I can offer some morsel fo knowledge that is helpful.
Ok I don’t have much time left , thermal watch with sensors cool to stabilize to all same temp even if some get heatsinks and some don’t need it
I was curious about the 24gb dimms outclocking the 16gb ones
Great interview. I’m gonna look at Patriot first next time I need ram because I like those answers (-8
B650 Riptide, DDR5 6000 CL30. I had no issue getting it to train memory.
Where I am uncertain is, Ryzen 9 7900. It's a 65watt chip but MSI shows it pushing 90 watts all the time. I don't know if MSI is off or if my chip is pushing that much past tdp, or if it's something I even need to pay attention to.
i wouldnt worry abot it, it is likely the opportunistic boost (Precision boost) at work. Keep an eye on thermals, as long as you are well within thermal limits, getting good performance, and happy with the cooler/fan noise.... let it run.
TL:DR the opportunistic boost feature regularly oversteps the rated TDP value to use extra power/thermal headroom for better boosting/performance. IF you truly want to lock it down you can likely disable this activity but it would be a solid detriment to performance.
I heard those speeds then remembered I have DDR4 lol
Get some good B-Die 😊 some of the best IC late stage in DDR4
Patriot is so underrated
❤
good video I got info and a few laughs
For a 'marketing' guy, he's very reasonable
Great info. Please write this up in an article. It would be a great resource for reference.
Hell yeah love this
Noise and thermal assistance
Shannon, I am buying an Asrock B650m HDV/M.2 motherboard (which only has two dimm slots) and a Ryzen 7700 (non-x) CPU. I want to get a 64GB 5600MT/s ram kit but, in ya'lls QVL the only 64GB kits listed for Asrock motherboards are your 4800MT/s "signature series" kits. Even for the X670E Taichi motherboard. I really want to get a Venom non-rgb kit to put into the HDV mobo. Is there a problem with running a Venom 64GB kit at 5600MT/s on AM5?
I will be honest, you can try it and likely it may be ok with newest UEFI updates, but being the current state of AMD, I can tell you that the density, same as slot population puts extra stress on the memory controller in most scenarios. The only thing I can say is you can try it, and in my experience as the UEFI matures it should be possible. However right now as I'm sure you have seen there have been massive changes to AGESA and their usage/limits to vSOC, and therefore I think you will need a few patch/updates for everything to settle. Worst case, for now, enable the profile, if not stable, manually turn the speed down to 4800 and with each succession in UEFI update you can test again, as these small update/steps will likely see more performant RAM results.
@@punx223 Oh, there ya go. I hadn't thought of that. Very good, thank you.
Awesome interview! Shannon has my respect, cool dude.
🤘
My pick for RAM guru is Buildzoid
BZ is very talented, for sure
Guys if you are not a professional overclocker. Dunt bother with it. Just buy memory with the maximum your processor com handle officially.
Ram is the only part I did upgrade in my system. Ignoring additional storage. Also the free graphics card I got, but that was a downgrade for some games and workloads.
What platform and memory did you upgrade to? Sometimes chasing speed especially for gaming can end up being a losing effort, or at the most a marginal/menial result. You really do have to find the right spec to match your platform and usage.
Some damn good info there thank you, you gave me an idea
Great vid Shannon
good info. thank you I needed this
08:10 - When Should You Buy RAM? When you just find the cheaper mb and ram combo that will suit your needs. All the money you save you'll have to spend in other things, no need to pay a fortune for 8000 ram than performing at max 2.5x faster than ddr4 3200 will have a first read latency (CAS) so much worse than 10 ns (+20 to 50% at ease) than PCIe or cpu BW limitation will make become a 2.5 times faster a 2.5 to 25% max FPS increase...don't waste your money
Shannon!
❤
Patriot has good customer service (The Netherlands) . Cyberpunk 2077 killed one of my 16GB Patriot dims. within the same week i had a new one.
Thats awesome to hear and thank you for sharing. We have a great team out there and I will share this feedback with them.
Samsung B die and DDR5 need cooling to be performant. The performance increase on lows is honestly underreported, especially with Ryzen systems. That being said, if you can run DDR4 3800C14, it's much less of a headache than DDR5 tuning. I'll take the possible 5% loss of performance from top end DDR4 to 7200mts+ DDR5 for the days of headache it is to tune....
Shoutout to Patriot Steel Vipers. Best value SR B-die kits in Canada.
Yeah, Patriot Viper Steel DDR4 kits with B-die are the best value. Unfortunately for me, I could not get my kit to complete stress tests without a fan on the memory. They got too hot. I decided to do surgery on one stick to replace the thermal pads on the modules and replace the foam on the back with a thermal pad and destroyed the stick when my PC failed to post. I took the loss as a great learning experience.
Aren't there issues with overclocking (at least with XMP) DDR5 RAM once you get into the higher capacities as well? Like, 128GB+ or something?
Patriot Viper Steel are great sticks and some great B-Die with the 4400s being pretty sick. Much like the Xtreme 5 though it's gonna be rough on some IMC/mobo
Hi Gordon and thank you very much for this video on rams and l also wouldlike to thank Shannon Robb l never heard his name before but after watching the video he is a cool DUDE and really love the way how honest coments he made so Gordon thank you so much if it wasn't for you we would never heard proper educational video like this about ram! keep up the good work!👍👍👍
I know you touched bases on DDR5 temperatures but I am curious. I am running overclocked Hynix dual ranked DDR5 in my AM5 build that is the stock 2x32GB 6000 CL30-40-40 kit. For the most part they are under 40C but when I'm really using my PC, HWInfo64 will show the SPD hub temperature reach 64C for a moment. Which is odd because often times it's just one of the two sticks and if its not reaching 64C, it mainly stays under 50C. There doesn't seem to be really an in between. Is there a cause behind this or is this too much for my sticks in any way? I don't seem to be having any issues with performance in any way but I've not seen ram ever reach this high of temperatures before even if it is only for a second. Heck for all I know it's a HWInfo64 fluke.
memory likes to live around 60c, and tends to have a really high red line. i know GDDR has a redline around 110c, and wont throttle til 90-100.
It's a bug of the sensor. The actual temperature is the one that stays on most of the time, so don't trust the max reading of HWiNFO column
Nice guy with a lot of knowledge. Too bad the brand doesn't seem to really exist in my corner of the world
Which Corner of the world? I am the guy who may be able to help with that.... depending
My 7950X3D runs best at 6400 Mhz (G.SKILL Trident Z5 Neo - 2x32GB). I tried 6000 and 6200 with better subtimings, but nothing that ran stable was faster than the 6400 Mhz settings, so I left it like that...
That's a solid setup, but some small tweaks to vSOC may help with tightening up your sticks, but that would take some work...
As long as you have a good FCLK/DRAM/UCLK you will be in a good place either way, even if leaving a modicum of latency on the table
I love RAMguy
❤
00:36 - 2 DIMM vs 4 DIMM gamer/PC versus Dual, Quad, Octa, Dodeca channel memory Bandwidth o server/workstation, 128 bit vs 768 bit, 25 GB/s DDR4 3200 (10 ns CAS latency) Dual, 38 GB/s DDR5 4800 (15 ns CAS latency), 28 GB/s DDR4 3600 (cl16 10 ns CAS), 50 GB/s DDR5 6000 (cl28 10 ns CAS), just cpu limited to 50 to 100 GB for AM4/AM5/1700 sockets cpus... vs server 200GB to 460 GB/s...
First time i tried memory ocerclocking i did my research, i saw that AMD's APU like the 4750G and 5700g had an IMC on TSMC 7nm and was generally pretty good compared to AMD CPUs and Intel CPUs, then i got a cheap ITX 2 slot board, and then some single rank samsung bdye rated for 4400C19 (super cheap, under $100)
Within probably 2 hours i had that up to 4933C17-17-17-28 without changing anything on the CPU other than manually setting IMC, FCLK, and UCLK to 2466 and disabling gear-down
for ~$500 i had a system running faster, with lower latency, than any DDR5 kit at the time, including those $1000+ kits
Its just too bad the 5700G only has PCIE 3.0, and has 1/2 the cache per core compared to a 5700X. Sure in some use cases that memory did make my 5700G faster than a 5800X3D, but in most cases it only slightly bested the 5800x and sometimes lost to the 5700x
With one of the 5700G i was able to eventually hit 5000C16, but the 4750G(the only one i have) was only ever able to get 4933C18, or 5000C22, maybe it was related to the added out-of-band-management, maybe it was related to having 2 CCXes instead of a single unified CCX/CCD
On the subject of heat, yes, very much so, after i put my best 5700G into a micro PC with only that noctua horizontal cooler with a slim fan, it dropped to not being able to even hit 4933, instead only acceptably stable at 4600C18, still good, but nowhere near 5000C16
Congrats bro on an epic overclock. Sad about the cut down cache on those monolithic dies but damn dude that's still an achievement 👍
@@foxdart Part of it was accudentally reducing the time betweem memory refreshes or something along that line, making it slower than it could have been at the same spec, but it was still faster than DDR5 at ~75GB/s and 49ns of latency
after a chat with buildzoid probably 2 months later i found that with proper config i might have been able to get down into the low 40 upper 30ns if i had not reduced one of the sub timings.
It was... surprizing for me to achive that first try, but buildzoid clarified that it probably would not have been stable had i not reduced that refesh delay or whatever it was.
Now we have kits of DDR5 that are near or above 100GB/s and some down to below 45ns, though often not in the same kit
what about timings, i think it's way important than the speeds but almost no one talks about it except AHOC x HW unboxed getting results with almost no perf increase from 6000 to 7200 mhz while getting ~ 10% boost from tuning the timings
Depends on your goals, and totally agree. Those are not mutually exclusive, there are workloads where high speed will deliver better results than a lower speed tight configuration. However I also agree that super tight kits in the 6400 - 7000 range can be rockstars if you have a good board/IMC combination.
If just gaming, crazy high memory clocks will not benefit much, there's a tipping point where your speed stops delivering noticeable improvements.
Alot of times it depends o. Hardware combination and workload.
Gordy shouldn't worry too much about EXPO or XMP. The only thing that he should care about is WIG, a new tech that is much faster than EXPO and XMP.
How to define “will run”? Y-crunch vst overnight?
Y-cruncher vst overnight, testmem5 extreme anta 777 preset, hci memtest 1000%.
Credit to Buildzoid (here named: Actually Hardcore Overclocking
We have run our 8000s on HCI for 72hr, and YCruncher among other stability, number crunching and stress tests before evr validating them to release.
We even test some workstation applications under heavy burn loading just to test them at actual work. On cpu render, etc.
I found out the hard way i got a 7600mt 32gb kit and my 13900k or z790 aorus elite mb can only get 6800mt stable with no crashing. I could have got a lot cheaper kit and had the same results. MY kit cost £500 4 monts ago
That's pretty rare, which bios version? I've had the aorus elite to over 7200 quite easily. I mean it's possible IMC may just be kinda bad but that is quite low from my experience.
@@punx223 buildzoid did a full breakdown of the aorus elite and found that 7200mt would be the max stable he could get because it's got 4 dimms and it's only a 8 layer board. plus it looks like i lost the lotto on the 13900k my chip only has 92 biscuit score which is pretty bad. my setup maxes out my alienware 34" oled ultrawide so its not a big deal. just a waste of money
@@benprice1984 I have seen his coverage, but BZ seems to not be able to get stable above 7000/7200 much at all, which is very different than my experience. If I get the opportunity to work with him, I will be happy to explore it and see if there's something weird/amiss I can help sort for him, or he just has some of the worst luck in 13th gen silicon I have seen to date.
BZ seems like a great dude, but our paths have not crossed thus far, but I am always open to work with people and help debug/tweak to find whats the root cause of issues.
Gordon, you know Frame Chasers is just going to laugh at this one.
I welcome their feedback, I'm more than happy to discuss this topic with them.
Shoot I thought RGB was only for looks, now I learned it is performance booster! 😛
Just like stickers on cars my friend, every sticker can add "up to" 10HP, with computers though you have to decide on whether the RGB adds better latency or speed.... choose wisely.
Dang, I knew I should have installed the amd cpu sticker on the cpu before I put thermal paste on it. I could have gotten more MHz!
Much like cars, stickers add 10HP or 100MHz, or 5C, you have to choose 😂
14:45 - Does RGB Make RAM Go Faster? MAtching your RGB with the RAL colour code of your case or hardware parts, may go faster the light so much you don't see any difference. IN fact who cares that a 13th gen INtel CPU memory bus is limited to 78.6 GB/s even for a Celeron or Pentium from 12 th gen because you can overclockit using 8000 faster expensivest ram to 128 GB/s, while AM5 only doing 78 to 100 GB/s and INtel DDR4 dual channel 4400 only goes 2x34GB/=68s but losing a lot of CAS latency, so 3200 or 3600 for AMD and 4000 for INTel only while cas latency goes below 10 ns.
Buildzoid would definitely have something to say about 7400 being easy to run. So confusing
I have seen alot of his coverage on this, and I'm open to discussing and working with him. I just honestly have not had the chance to do so.
That being said, my experience has definitely been wildly different than his. No way to know why without alot of data, but I would still love to sink my teeth in and try and help, this is what I do.
Just to be clear my main rig is running a 13900k, retail from Microcenter.
4-DIMM board gigabyte Z690 waterforce and it runs 2x16GB 7400 24/7 since I've built it.
I have a feeling that there are differing standards of stability. Sure, it can run 7400. But does it fail in certain corner cases? He was like "one error after eight hours".
@@punx223 It runs and maybe even stable, but the “7400 is easily” is the thing. I believe he says that 6800 is the safe speed to run without headaches (XMP). His is a sample of I believe one 13900k and a 13600k so there is that. I think he is taking it that his 13900k is the worst bin there is.
I am just perspective buyer and don’t have much experience with any of this. I would fall into the “Run XMP now, and if a someone like him could give me some safe, better than XMP subtimings I would start there and do some testing but probably not 12 hours worth of it, and likely with whatever I could do it for free with. I have only used OCCT and the old testmem that makes you open line 10 instances.
@@TheGreenReaper Also 4 DIMM or no, isn’t a water force board really good? Maybe it doesn’t matter but is it a 8 layer or 10 layer, and does the water part cool the Ram?
subtitles broken for anyone else?
The question the manufacturers are too scared to answer is why they refuse to validate ECC kits for OC. Gamers want ECC, too, and not just at DDR5-4800
Not Afraid to answer this at all, and honestly the two are kinda at odds most times. TBH DDR5 has a minor hardware level of what I call "passive ECC" but in reality ECC has always been made for absolute stability. ECC hardware on DIMMS (Like real ECC/server level) can add induced latency as it has hardware on board checking for any bit flips to avoid data corruption or stability issues. In many circumstances these ECC modules are strictly controlled on their signaling/timing, and therefor allowing clock speed modulation/overclocking can cause chaos for the IMC/memory system as a whole.
This is not to say it is impossible but largely improbable.
Also, the platforms have to have ECC enabled which most mainstream/enthusiasts desktop platforms they do not, so that topic would first need to be addressed on the platform level.
@@punx223 Literally all modern Ryzen CPUs come with ECC support, lol. My small server is running a Ryzen 5700X with ECC UDIMMs. The chips on ECC memory are ostensibly the same as on non-ECC chips, they just get extra chips for parity. All manufacturers would really need to do is evaluate the individual chip's overclocking potential, bin the chips accordingly, and then place them onto an ECC DIMM.
But they don't. Because the server market needs to be protected at all costs, I guess.
@@insu_na yes I am aware Ryzen and TR offer it, but my explanation still stands that overclocking and ECC run into each other.
It's not as simple as "test the dies" I mean, yes we could probably (cannot say for sure) setup an XMP SKU with ECC... But once you were to try and tweak or tune it the timings can fall out of sync or cause errata, and that's why nobody really does OC ECC DIMMs.
Again, that being said, I'm not saying it will never happen but there would need to be alot more demand, and the underlying tech would need some work to make it scalable.
Sorry, that's just reality.
Don't you get used to random crashes and having to act as if they not existed or starting a research carreer on maintaining your own PC? They do so user become involved and don't fall sleep and extending your time in front on the display, just how first home computer were entertaining while waiting for the casette tape to load and eventually crash because a random analog read/write error, or a magnet or heat on floppies or an airbone virus that just landed in your partition table... is just for fun and laughts...
This is a truly great interview. Gordon had great questions and we got great answers.
Treat ram like ram on a graphics card helps Apus act like graphics cards
Why does DDR5 not support ECC by default?
DDR5 actually has a low level of ECC within the DIMMS but nothing like you see on server/enterprise targeted modules. ECC is a strong component for scalable datacenter/.server solutions, but as you notice the speeds are tightly controlled. This is because the ECC tech within the modules are sensitive to detect things like bit flips, etc. Overclocking ECC DIMMs is always a weird prospect as you are taking a module designed for very precise signal timings and trying to modulate the signals to give better performance, this as long as ECC has existed has been a tough combination to resolve.
It is not purely impossible, but you also must remember an important component of this is also the platform support, and most consumer platforms simply do NOT offer ECC compatibility at all, and wont even boot with ECC modules installed. If we want enterprise level ECC support, it has to start with the CPU/platform level, then from there we would have to see how far they can be pushed before causing false positive errors within the ECC.
@@punx223 I just come across so many products that would be perfect for SMB but because they lack ECC support they're a no go.
@@ericneo2 I totally understand, the issue really comes down to platform support mostly.
Not to mention as I have said before, overclocking and ECC can be pretty problematic as overclocking, you are pushing performance and getting really good stability, but even the IMC can become unstable, which is why you notice ECC enabled platforms, or mostly, always run at the JEDEC for the platform.
I soak the whole ram pcb until it changes color then clean and alcohol copper slot pads only
uhhh, soak in what? I mean something like IPA shoudl not have much issue, but please be cautious soaking any component.
RGb shaves off 5ns latency
Mr ram man says air is hot over there. How come he is not wearing heatspreaders intead of his shirt ?
Still in prototype, not ready for public release 🤣
@@punx223 I hope we see man heatspreader wear in future fashion shows 😁
@@SupremeMasterr no guarantees 😂
Like every hobby it pays to not go too overboard. If you're at the point of buying single slot motherboards, 8000 ram and binned processors you should probably stop and take a deep breath.
It's not for everyone, that's why there's options. But for the people looking to push the limits, it opens some doors.
10:04 - Should You Care About DRAM Die? No because you may realize DDR embedded in Mobile phones, SoDIMM or Apple Silicon it's is much better faster, lower cas latency and lower power usage and made with state of the art Node process technology isntead of the dimms you buy for desktop on older nodes wasting 1 or 2 digits Watts.
least lunatic PC world viewer
Not sure how much SARCASM in your 5 words. Just in case your worried to...
@@-eMpTy- 12:52 - Should You Worry About DDR5 Temps? NO. You'll know you have worse problems to worry about once you realize how that metallic smell like ozone or blood is due to vaporization and air ionization and plasma wich started just a little before you heard that first crack. What crack? then a repeated mini fireworks will distract you about the fact you already lost your work or gaming progress...At that moment is when you start to worry about ddr5 temps and how unfortunated is the fact that you can't search a video tutorial on how to deal with your problem...Tranquility begins when you grab your phone...
You'll never know if your system RAM is stable without ECC reporting. When are the DDR5-8000 RDIMMs?
They are for sale now, and we have run extremely in depth validation on 8000 and the profiles before ever considering releasing them.
Wait, you mean like 12V RDIMM? I don't think Sapphire Rapids will really push that yet, give those platforms time.
Great! do you have a part number for 4x32GB and 8x32GB kits? I searched the Patriot store for RDIMM, but all the results were UDIMMs. Also what motherboards are QVL for W5-2465X and W5-3435X?
@@punx223 Yes, I thought all RDIMM are 12V? If not Sapphire Rapids then which platforms are on the QVL?
@@shanent5793 our DIMM are 5V UDIMM on DDR5, we have not released DDR5 12V RDIMM yet
8:20 not me, graphics card first
everyone has different build priorities, and realistically I usually buy all at once, but if I were to take steps, I would likely pick GPU last as prices are far more volatile.
@@punx223 I only do a build once about every 5 years, so I tend to just buy top of the line parts, I did wait to buy my monitor though because I had seen it on sale before and knew it would be again soon
I love when guys are super knowledgeable and passionate about what they do.
I am glad, because thats what we do 😊
I would have more respect for memory manufacturers if they listed all of the timings instead of hiding most of them because they all want to sell us "speed" instead of trying to teach people how speed for their product actually works.
I think the issue is 'marketability' of memory timings. Unfortunately, consumers are tuned for higher clocks as that's what the CPU vendors also market on. Consumers also understand 'more MHz' and it moves them. While a small subset of tuners will understand and target timing, the vast majority will only focus on the clocks. The good news is most vendors do also provide the module timings in specs too.
I'm here, what would you like to see dude? Our sku sheets for each model of DDR5, show the primaries, which timings are you wanting to see?
Unfortunately doing something like a TB dump would not really make sense and even be information overload for many people.
@@punx223 Solution is quite easy. Add full info links next to current ones (or as separate segment, web dev should be able to figure it out) with all XMP/expo timings.
I am quite sure you are already aware that even 5000 DDR4 kit can be worse then JEDEC 2666 and that kind of information should be easily available to calculate. Obviously, this kind of info matters since it is relevant for a product and most of the purchases that involve me are for others so I can't OC memory for everyone.
We do not get reviews for memory like for CPUs/GPUs, so it would be nice from any company to get out of their way to break "only bare minimum" standard.
How about that ram?
I still don't see any comprehensive explanation of the benefits of switching to DDR5.
I guess I can sum it up like this: Is it worth dumping an existing LGA1200/AM4 system with DDR4 for DDR5? For the most part, the return on investment will be very small and most could not justify it. However, if you are building new, or selling old to build new, the advantage with DDR5 in density and clocks (more RAM, higher clocked RAM) with current DDR5 prices probably means (if you're performance focused) you should indeed "switch." Yes DDR5 has a very real advantage (for example, QuickSync performance with Intel has a very sizeable increase in performance on DDR5 vs. DDR4 due to the bandwidth advantage) but it's going to be situational based on your own personal situation.
@@FakeGordonMahUng Thats what I think.
@@teekanne15 Yup. There is no "right" answer despite what some people insist. Base it on your situation and budget.
Ddr5 is emulation ddr4 raid
Well is you mean the dual rank visibility to OS or overall design, it essentially is somewhat true but its more involved than that. this is why single DIMM DDR5 runs quite well, and you dont take as hard of a perf penalty vs running single DIMM DDR4.
You get ks they all hit 8000
Actually from my testing KS is normally a bit stronger core, but IMC is still similar pass/fail rate.
@@punx223 non of mine ks failed. They all do 8000 + easy on 2 dimm z790 tachyon
@@facegamefps6725 nice! I have tested in the triple digits, so I have a pretty wide dataset to pull from, while it's not all encompassing, this is my experience between K, KF, KS
@@punx223 cool! I have to say all of my K once maximum I could do was 7800 tuned stable. Overall they are much better than AMD MC
Feels like he’s speaking as if your trying to hit the fastest speeds, not what is actually going to stay stable for daily use
That's a good question but I believe Shannon's parameters, like most module vendors, is hitting highest clock speeds while being stable in X, Y and Z metrics. One problem with 'stable' RAM is stable is different to everyone. But this all gets to the problem we have where most of us assume 100% stability in all things in modes are dependent on so many different factors with so many variables. CPU lottery, how well you cool that CPU, motherboard, module selection and even quality of power supply can impact memory stability. Throw in the fact that the hundreds of thousands of different use-case scenarios and I'm not sure we should really expect the same 'stable' use as running JEDEC clocks and timings.
The target of the Xtreme 5 is definitely to try to break records, we are very honest about this. It's similar to why companies enter F1 and spend a ton on R&D. It's because we are performance guys and we want to do it because 1) it's fun and 2) because we can, we wanna show what we can do.
If looking for a set of DIMMs just for gaming and whatnot, then there are several models to choose from that are at more reasonable speeds that will run fine as well, as the crazy high end is more for tweaking and tuning, while it can game, it's excessive for that use
Aka taking a Ferrari to get the groceries.
An open memory slot acts as an antenna for interference, choose a board that fits your choice of memory kit.
If pushing for max performance, 1DPC is your friend