I think both Intel and AMD reps in this were very good. I mean they never spill the beans on upcoming products, but they both know their own platforms and clearly were trying to be as open as they were allowed. its not easy !
When I was selecting memory for my 7700x I just cheated and used the exact same kit they supplied to the reviewers. No way they would give anything but the most optimal kit for that. That turned out to be CL30 6000, which matches up nicely with your charts for AMD.😲 I would be interested to see how the RAM effects 3d cache chips, my guess is that it will have little/no difference on performance.
I bought some G Skill Trident Z5 6000 CL30 for my 7600x rig, and it’s been great so far. I don’t get into the nitty gritty of trying to OC RAM. As long as it’s running at EXPO speeds, I’m happy
Thank you so much for making this video, i am gathering parts for a new high end build for gamin and have been racking my brain over lower Cl or higher M/T for DDR5 gaming.
When I was searching for an ideal kit on pcpartpicker, I started by setting the minimum speed to the maximum that my processor was rated stable for (5200mt for my 7600x), then the lowest latency within that range (28cl), then the highest speed at that latency (5600mt at the time; I believe there are 6000 cl28 kits available now). Worked out pretty well.
@@Pyreleafcas latency is in clock cycles remember. 5600 cl28 is the exact same cas latency as 6000 cl30. As long as that CL figure is half of the first 2 digits of the memory's MT/s, they're the same. Ie 6400 would be cl32, 7200 cl36 etc.
@@Frozoken not *exactly* the same; if memory (heh) serves, latency can't really be adjusted up and down in the same way as speed. When I still used my old ddr5-6000 cl40 kit before a recent upgrade, I couldn't run it at the full 6000mt on my 7600x, so I had to settle for 5600mt, and I think the cl stayed at either 40 or dropped slightly to 38, so my fwl actually increased. In other words, a 5600 cl28 kit may have the same fwl as a 7200 cl36, but if you can't use the full 7200mt, you're still stuck with the higher cl36.
did a smol upgrade, switched to a MSI Z690 Pro-A Wifi DDR5 mobo from the ddr4, installed G.skill cl32 6400m/t 32gb ddr5 ram, wasnt a major upgrade but did notice a difference in performance :) running a i5-13600k+RTX4080
hello there very informative video... but i would like to know , did you use xmp or manual overclocking for the intel cpu and higer frequncy ram kits....
It really is not surprising to have supposedly slower kits beating faster ones because sadly all we see from the kit is the primaries but other settings can increase much more performance. For example, I can easily beat a 6400 c32 kit that is using xmp with a tuned 6000 kit.
It would be so much easier to compare memory latency if it were advertised in nanoseconds rather than cycles, so we could directly compare latency, independent of the speed we ran the memory at. Unfortunately, absolute CAS latency hasn't improved since the earliest days of DDR, and that doesn't fit the "newer is always better" narrative beloved of the industry. Ordering the reviewed kits by absolute CAS latency, adding in some legacy specs for comparison we get: CL16 @ 4266 = 7.5ns DDR4-4266 CL9 @ 2400 = 7.5ns DDR3-2400 CL4 @ 1066 = 7.5ns DDR2-1066 CL34 @ 7000 = 9.7ns DDR5-7000 XMP £195 Dominator Platinum RGB 32GB (2x16GB) CL30 @ 6000 = 10ns DDR5-6000 EXPO £310 Dominator Titanium First Edition 64GB (2x32GB) CL2 @ 400 = 10ns DDR-400 CL40 @ 6800 = 11.76ns DDR5-6800 XMP £130 Vengeance 32GB (2x16GB) CL36 @ 5600 = 12.86ns DDR5-5600 XMP £220 Vengeance RGB White 64GB (2x32GB) CL38 @ 5200 = 14.62ns DDR5-5200 XMP £105 Vengeance 32GB (2x16GB) CL2 @ 100 = 20ns PC100 DDR5-6000 would need to be down at CL23 to get close to CL16@4266. Why does the Vengeance RGB (12.86ns) do so well in so many tests? That would require more testing, but it wouldn't be the first time that CPU/cache issues resulted in 'slower' speeds being 'faster'. It would certainly be interesting to see what running the Dominator sticks with CL28 @ 5600 (10ns) did to the benchmark results. This may be one of those situations where lower latency at a lower MT/s works better overall than higher latency at a higher bandwidth.
Yeah, I would have liked the first word latency to have been included in the video, but nice of you to do the work here, so I don't have to calculate it myself.
I picked up a couple kits of GSkill CL14@4000 = 7.0ns last year, Waiting for the 14900ks to come out and going to pair them together for a low latency system
With all due respect to you Leo, I have a sneaking suspicion motherboard sets loose any secondary and tertiary timings that don't fit into XMP when going for higher frequencies. If so, then it might be taking a look if those were at least minimized if you can't keep the same values which are set at 5600MT/s. Also I assume it might be worth finding other voltage sweetspots, I doubt auto settings allow for training tight values at higher speeds. Love the channel, don't mind me suspecting your testing methodology, those ARE realiatic scenarios for plug n play type of people, but for anyone who has knowledge to manually set overclocks and timings I think it lacks in info. Would love to see either BIOS inputs or at least zentimings and asrock timing software (or whatever is used on intel to check timings) screenshots.
I assume he used the XMP/EXPO profiles thus the primary, secondary and tertiary timings have been set that way. I have no problem with that (manually set overclocks are not for ordinary users), but since those XMP/EXPO timings can be still not in line proportionally to one another, I agree that he should have displayed them. They may explain some of the unexpected results which right now are mystery.
@@vilimtustanovski1270 Ok, but still I would be interested in to calculate the ratios between the (XMP/EXPO) timings of the kits. That has been the main thing for me.
I appreciate the hard work as I’m sure the benchmarking took forever, but a lot of this data is useless unfortunately and certainly not definitive by any means. Testing different speeds with RAM timings all over the place makes no sense for buyers/viewers. For example you can get a 5600MHz *CL28* 16x2 kit for $103 on Newegg. Yet you’re testing a 5600MHz CL36, which for a 16x2 kit is $88 on Newegg. That latency makes a massive difference on the AMD side of things (lesser so for Intel) as seen on LTT’s and Hardware Unboxed’s video. The extra $15 is well worth it and matches the more expensive 6000MHz CL30 kits. Really the only definitive and ultimately useful test would be testing 16GB x 2 kits (since it’ll be the most common setup for buyers) and picking the lowest latency kit at every speed. I realize this would be an expensive endeavor, you’re probably looking at $2000+ in memory alone, let alone the sheer man hours. Though you could skip basically anything below 5600MHz as both Intel and Ryzen performance has a significant drop off regardless of latency at anything below 5600MHz
He helped me too, i honestly struggle to understand half of what he says at times, but he has some good tips for boards and settings which can really work wonders.
If your total RAM is under 128g, you want to use a set (2) identical RAM memory sticks because using four will cut your mhz... For example, if you want 64g RAM, you want to use two identical 32g sticks, not four 16g identical sticks... You will see a noticable difference by trying it both ways... 💥
You need to do some testing with command rate and number of ranks vs locked memory timings then optimized memory clock speeds to what each setup (1T/2T, 1Rx8*2, 2Rx2*2, 2Rx8*4) can max out at. That would show much cleaner why some kits will perform better than others.
I love the Intel guy trying not to loose his job as he talks about the guy who repeatedly has said Intels latest memory controllers are sh!t and he doesn't even want to try overclocking on Intel anymore until they correct it. I'm sure they are both equally frustrated by this, he just can't say that in a interview lol.
@@philipjfry628 yeah, the intel guy looked very relaxed, the AMD guy was definitely in a more PR style delivery, but I am sure they know what not to talk about. I wonder if Kitguru had to cut parts of the dialogue under an NDA and get approval before this went live.
Would love to see you go back and try this out with a X3D chip as it seems memory speeds have a much smaller impact on performance with them as the L3 cache provides that low latency memory required and picks up a lot of the slack from slower memory kits. Also think it would be good to see you do some longer burn in tests as you said your a fan of Buildzoid, he has pointed out that Intel does not remain stable on the higher memory settings starting around 7000 or so. Finally I've seen some 24g and 48g kits out there that are hitting 8000mt/s which if I understand correctly, due to how memory speeds are tied to the infinity fabric clock, should be much easier to get stable then anything from 6200 up to 8000 so that could be very informative for the consumer as well, especially as we see these speeds becoming more common.
Hardware Unboxed's video on DDR5 scaling are way more extensive and a lot more applications are benchmarked. To those who want to know more I recommend you go there. This video tells very little story.
@@saricubra2867 They do. What were you saying? They even took Buildzoid's DDR5 timing to use it as a performance benchmark. Although they do specify the brand, who cares about the brand anyway? The chip inside the RAM are the one that mattered, brand doesn't. as long as it's Hynix A die it'll all perform the same with the same timing & frequency potential.
I would have liked to have seen DDR4, say 4000c15 in this line up for the Intel part! Many people are still on DDR4. Not much use comparing without the previous gen.
Bcos I plan to upgrade my aging Intel rig I have been searching for info but it is damn tough to find smthg useful about mem scaling. I never cared about zip/unzip performance. I am not a file compressing enthusiast, I do not zip or unzip files often enough. And what if it takes 5 seconds or whatever more? Per Buildzoid the AIDA64 mem bandwidth benchmark is quite useless. It measures ideal scenario when the data is perfectly sequentially arranged in the mem which never happens in real life. The gaming benchmark section raises more questions for me than it answers. First, only 2 games are tested which is very little sample data. Then I can see that in both games the red team delivers FPS below the blue one and AMD scales with mem speed but Intel does not. For me what it means that AMD could not hit the engine limitation or cap (contrary to the conclusion of the guy). Ok, but the FPSs are practically equal at the same settings at both 1080p and 1440p which is normally an indication of engine limit. Therefore I can see a contradiction here on the AMD side and I am not able to make sense of it. Additionally in FC6 the AMD performs at 1080p a tiny bit worse than at 1440p (and Intel too on the 1% lows). Even if the differences are only a few FPSs it still bothers me that something might be wrong here. Or this is just how FC6 is, right? But other games? (Afaicr Doom Eternal scales with bandwidth, for example.) Or future games? I simply cannot trust that this material puts me into the right direction. I am open to go either way, blue or red and can afford to spend on mem reasonably (up to ~6400 MHz). I plan to use the XMP/EXPO profile, no manual tweaking of the timings. On AMD’s side the ~10% gain showed here by faster but more expensive mem is acceptable for me. But this guy suggests that I should not care about the mem. Is that really a good advice? (Especially based on a 2 game size sample?) Any thought?
If you decide to go to AMD and you are ok with the price then I think a decent 6000 kit is good, but pay attention to buy something with tighter timings because that also matters more on AMD. You are not exactly right about that Intel does not scale with memory speed. The 5200 kit (which probably has the worst timings as well) is slightly slower in all 4 test scenarios than the others. So I think what it means is that the 5200 is not hitting the engine limit, but going upward it happens soon. Generally amongst the PC testers results within 2-3% range are deemed to be equal due to measurement tolerance. Therefore I consider the ranking of Intel results in these charts above 5200 to be meaningless, and I think you certainly cannot expect the 5600 CL36 kit to perform better than the 6000 CL30 or 7000 CL34 if the game is not engine limited. Another weird thing what I can see here is that usually at 1080p the FPS numbers are 20-30% higher than at 1440p. If we accept the 1440p data to be correct then on both Intel and AMD the 5200 kit should produce higher FPS at 1080p than at 1440p and hit the limit. But that’s not what we can see thus I believe that all the presented data as 1080p are actually 1440p. Some kind of mistake must have happened. Unfortunately I do not know as much about Intel as about AMD, so instead of saying anything I would say try to do further research on the subject. I hope so this piece can still help.
That’s not why AMD systems can now support higher speeds, there was an agesa update from AMD that significantly expanded memory compatibility on AM5 … you will also need to switch on high speed or low latency mode in bios. Also, the Z790 refreshed motherboards from all manufacturers have QVL’s that include blatant lies … none of those 4-dimm motherboards can run 8000+ stable. You need to update your bios on your AMD motherboards and update the chipset drivers … the higher speed memory will then work.
@@haremofprocessors6954 The audience who follow buildzoid would, but most people just hit the BIOS settings default and leave it at that. I can understand why. a lot of it makes little sense to most people.
correct, and this lack of tuning/setup is the reason so many people fall into the ryzen is good trap a tuned intel setup is so far ahead its not even funny
that's a side effect of V-Cache, not an indication of whether faster memory is worth it or not. take a look at AM4 CPUs, they were well known to love fast, low latency RAM, but the 5800X3D was an exception in that RAM choice didn't affect it all that much compared to regular AMD CPUs.
That is just untrue. I saw up to 14% better performance in games going from 6000CL30 Expo settings to manually tuned 6200 MT/s memory with my 7800X3D. Most media do not test memory performance with actually overclocked memory, almost everyone is using the shit Expo/xmp profiles that come with the DDR5 kits. Most kits have terrible timings. Also, single-CCD Ryzen chips, like the 7800X3D are limited by the infinity fabric clock and cannot fully utilize the bandwidth that even DDR5 6000 offers, without overclocking the infinity fabric. With better chips you can get 10% more bandwidth from the memory without even touching the memory itself, just with an fclk overclock. This becomes apperant in open world games where memory bandwidth is more important than latency. Reduce the latency on the memory and you get more efficient use on the bandwidth available, and you get higher performance in games as well. Although with the 7800X3D, overclocking the cache itself helps in all gaming workloads just as much as memory OC.
Intel's Alder Lake chips and similar ones (13th and 14th gen) use Ring Bus design and are monolithic, the IPC is higher and they can communicate better with RAM.
My old 7800x3d couldn't post at 6000mhz with Expo guaranteed Qvl listed ram sticks. Through many tests, I concluded it's the CPU problem like the mem controller not the ram sticks. I replaced my CPU with AMD warranty. I heard there are a lot of CPUs with low capacity just above the threshold. For the guys wondering my issues, I listed what I experienced. 1. No post at 6000mhz, but can post lower clocks with lower latencies. 2.Even though it makes to the window, it's unstable like random rebooting and crash. It's very similar to symptoms with badly timed ram. So I ran the computer at the default speed of 4800mhz, but noticed random freezing persists usually after using 2 hours. 3. It fails all the ryzen master cpu performance enhancement. If I click on the apply button, then system tries to reboot, but it always ends up to no post and bios reset. 4. I'm not sure if it's related but the cpu temperatur on the overlay monitor provided by Radeon adrenaline is stuck at the random number and stopped working randomly. If it happens I usually expected crash or freezing would take place. I got the new cpu and now it runs like silk. But the last one month struggling with the bad cpu was really nightmare.
If you set the game to a low resolution, lower settings etc and still only get a few frames extra when the game alreday run with 150+ FPS. Pointless, since more money into the GPU is what matters.
really good video kitguru thanks. I have still no idea about RAS and CAS, but I use XMP and just hope it all works. usually goes. thankfully. Went over to see buildzoid as I saw you give him a shout out. Went over my head.
I Purchased Ryzen 7 7800x3d and pre ordered a Corsair Dominator Platinum 6200 Mhz CL36 to upgrade my Am4 system to Am5. but lately I discovered that the Ram I bought was for Intel, SKU: cmt32gx5m2x6200c36w Question is: Will it work? or what will happen if I combine these two? I hope to get answers from you! Big Thanks!
one of the best mass produced memory videos yet... but, corsair ram sucks, and the ability to tune your system with one of the higher end kits wasn't addressed
You could argue does anyone tune their memory anymore ? I know a lot of enthusiast users who just run at xmp. I used to do it. Not worth it now for most people
Interesting comment I personally think that it’s a very very niche audience now who go in and tune ras and cas etc. it was always niche but xmp does the job for almost everyone.
@@lucifersheadscarf6988 that's always been the argument but I'd like to think that if someone wants to do the work they should be praised for it instead of someone saying what's the point....tuning memory has been an argument since the late 90s and I've never not gotten a benefit from doing it... For it to be an option is something worth mentioning, and some of these lower speed kits you can't do anything with
Not really, you would have to do your voltages abd timmings manually as XMP might not work properly out the gate. But all 2 dimms mobos and every 13/14900K will do 7600Mhz quite easy without breaking a sweat @MGK195
@@alexc5564 "But all 2 dimms mobos and every 13/14900K will do 7600Mhz quite easy without breaking a sweat" buildzoid and basically everyone with experience in memory overclocking will highly disagree with that statement. just running y-cruncher for 30 minutes without errors is already Apex territory. igor's lab has binning results for ~600 CPUs and 7600 was mostly the absolute peak with a few golden samples going beyond that.
I get scared with the advanced settings for memory in the bios. Feels safer just to leave everything at "auto". An unstable system is just a nightmare.
XMP is useful for all that. I dont tinker much with bios settings anymore, I used to play with RAS and CAS all the time. now I just leave it with the settings in the BIOS. sad, but true.
The hours spent tweaking a system for the odd percent - if you're lucky - is time better spent working to buy better components in the first place imho. But then, some people like playing around in the BIOS - and that's fine - whatever floats your boat !
Your testing shows that the 5600 has more aggressive timings pushing it to out perform the 6000 and 7000. But you never compare the timing of the separate kits of memory.
Hi, I’m going to build my own PC with a 7800x3D and I had a question. I am about to buy 2 sticks of Corsair Dominator Titanium DDR5 RAM, but from an aesthetic viewpoint, I want them to be white and those are only XMP. The EXPO ones are only available in grey. Can I use XMP memory instead of EXPO? Would I be sacrificing performance or stability?
Loved everything aside from the game selection. Those are very poor titles to test this kind of stuff, even though the difference in performance was measurable.
I wantto see you do this same video but for music/audio production, since everyone always focuses on gaming pcs rather than single-core-oriented tasks like music production.
Thank you for the video! Good content. On my end i just built a new PC with an X670e Mag Tomahawk, 7800x3D and was able to install 64gb (2x 32gb) of 6400 MT at CL 32. Everything is running very smooth, and the infinity fabric clock is 1:1 using the Expo profile. I was worried by others comments that the clock would be affected to 2:1 and the memory would slow down. My RAM kit was XPG Lancer DDR5 RGB
I also just picked up the same kit, running at 6400MT, 2133 Fclk Cpu - 7800x3d Getting 61ns latency on aida 64 memory benchmark best ive got and tested multiple kits
Cas Latency * 2000 = X / Advertised Speed = True speed of RAM (smaller is faster). True speed for DDR4 RAM of 3600 speed with 16 CL; 16 * 2000 = 32000 / 3600 = A true speed of 8.8888. Now lets do DDR5 RAM of 6000 speed with a 36 CL; 36 * 2000 = 72000 / 6000 = A true speed of 12. In this example, the DDR4 ram would run faster then the DDR5 ram.
What do old people need, RTX 4090 system, frames, or just a PC ??? 1600 is fast enough for these 14900k CPU's ? What do old people need ? 2 X Dimm ! NOT MORE !
DDR5 is a complete mess! Sometimes frequency matters, sometimes latency. Sometimes almost neither. 7700X didn't look good at all compared to that 14600K with 5800MT/s and over....
why is the FPS practically no difference between 1080p and 2k res during gaming? that makes no sense... was it fps capped or? Vsync? or other reason???
TL/DR: No, it's not really relevant. Obsessive types who chase: "bigger numbers" may like to mess with DRAM and all the instability and hassles it can bring, but from a human perception perspective, it's all completely meaningless. Something like a Raptor Lake 13600k or Zen3 X3D chip and DDR4 running at JEDEC timings will be perfectly fine for just about any non-outlier game/task and actually _noticeable_ "improvements" from messing with anything above that baseline will be entirely confirmation bias, end of story.
Hi. I´m building a "white PC". I chose the Ryzen7 7800X3D processor and the Asus ROG Strix B650E-A Gaming WiFi Motherboard. I Found the DDR5 Corsair Dominator Titanium in white, but unfortunatly only XMP. I could by the expo model but they only exist in grey. What would happen if i run the White XMP version on my AM5 setup? Will it still be compatible or will i run into some issues?
4200mhz c14 gear1 dual rank ddr4 gained me 20-30% fps in games that are hard on memory. Games like WZ, 2042, pubg. Maybe in a couple years there will be some reasonable priced mobo's and some ddr5 kits(best binned) that can push 8000+mhz c30ish speeds
There are ZERO games right know that use anywhere near the top of what DDR5 can utilize at its maximum. NONE. The problems right now all revolve around way too much power being needed to run some of these systems. Im sorry, but if you need a 1200 to 1600 watt Power Supply to run anything, someone needs to go back to the drawing board. There needs to be leap made in innovation.
Some games, simulators in my experience, lets say Star Citizen, loves fast RAM. It ridiculous how huge the difference in performance is with top shelf RAM and lower speed ones. Maybe its because of the bad optimisation, I dont know, not an expert,
I think I see why AMD constantly drops the ball, over-promises and under-delivers. Your one AMD guest seems to have way too many responsibilities/ titles. His business card must be huge. He had like five different titles.
As always RAM, once again, is shown to be not important for games. I mean its the same old story with testing in mostly lower resolutions, low settings and then still, in the worst case scenarios, only get a few FPS extra when the game already run at 150 FPS+. And mostly the results are all over the place for different speeds...
I'm a little disappointed at your CPU choices, to be honest. Would loved to have seen the performance numbers for the highest end CPUs, the 7800X3D and 7950X as well as the 14900K. Or did you test this and the performance uplift percentages are roughly the same?
@@POVwithRCI tested the 13900ks (14900k) with 6000 and 8000 (video & benchmarks are posted on my channel). Currently testing the 7800x3d with 6000 and 8000 too … very different behavior compared with intel.
People who buy i9s for gaming knowing they're pointless over the i7 are the same people to buy the most expensive ram kit they see. Why would you need a benchmark
Maybe i didnt hear but what motherboard was this tested on intel and am5. How can 13ns be equivalent to 10ns ram. I get 6000cl30 will be close to 7000cl34 within quarter Ns. But how can a 13ns kit keep on par with a 9.7ns kit. Something else is going on. Or am i somehow not understanding latancy over mts.
The AMD motherboard is Gigabyte X670E Aorus Master with BIOS F20a and AGESA 1.1.0.1 while Intel was used on Gigabyte Z790 Aorus Master X with BIOS F5g Leo
I just do not get these test at all. If you set the 5600mt ram the same timings as 7000mt ram ,there will be no real difference in PC games let's say.please try so E test with same ram timings for fun and especially higher resolutions,give a complete story on ram
When a new generation of DDR RAM is released, the sticks always perform the same as the best from the previous one until the technology improves, that hasn't changed in decades.
@@saricubra2867 noone care about benchmarks. For a majority of games DDR5 shows little to no improvement. Varying around 0-5%. In older titles, DDR5 is actually slower than DDR4 because older games prefer lower memory latency over transfer speeds.
From what I've seen in various tests, the memory seems to only make a worthwhile difference at lower resolutions where you're more likely to be CPU limited. I was also told that CAS latency appears to matter less with DDR5 than DDR4 and 3.
@@lucasrem Yeah, specifically the tests I saw were for DDR5 6000MT sticks with CL30-36-36-96 vs CL36-48-48-96. The difference was minimal, with the biggest gains coming on Intel platforms with higher MT sticks. And even then it wasn't that much.
i got the 7700 and run at 6000 right now, but i have been running the default at 4800 for a period because of stability issues and it doesn't actually feel too bad, at least in the games that i played at that time. but it is for sure a bit slower.
