Try this DDR5 Subtiming Tweak for Extra Performance
HTML-код
- Опубликовано: 12 июн 2024
- Check out Seasonic MagFlow Fans:
seasonic.com/magflow-120mm
---------------------------------------------------------
Support me on Patreon:
/ der8auer
---------------------------------------------------------
Music / Credits:
Outro:
Dylan Sitts feat. HDBeenDope - For The Record (Dylan Sitts Remix)
---------------------------------------------------------
Paid content in this video:
- Seasonic Spot
Samples used in this video:
- V-Color Memory Sticks
der8auer Advertisement Policy:
der8auer.com/advertisement-po...
---------------------------------------------------------
Timestamps:
0:00 Intro
1:09 Seasonic Mag Flow (Advertising )
1:50 RAM kit specifications
2:14 AIDA64 Cache & Memory Benchmark
3:57 tREFI Tweak
5:13 Gaming benchmarks
8:17 Summary/Conclusion
10:39 Outro 
Наука
![PADDINGTON IN PERU - Official Trailer [4K] - Paddington Bear is back!](http://i.ytimg.com/vi/lKgitu25ZAg/mqdefault.jpg)
9:23 cat meows in the background
Hah! good catch! Love me some cats and meows!
Please do an in depth video on all the DDR5 timings, what they do, and how to optimize them. Increasing TREFI is a good tweak, but, there is much more that can be done.
go watch Buildzoid
Tweaking trefi and trfc will exhaust 80% of the overclock potential, an average user really shouldn't bother with other secondary and tertiary timings.
Look up Actually Hardcore Overclocking / buildzoid - he's got quite a few guides for DDR5 timings and I've been using his "easy ddr5 timings" on my 7950x for over a year with no issues (well no issues that weren't caused by this mobo initially sucking at training but a bios update fixed that).
@@beachslap7359 Unless you have Ryzen, then tweaking the Secondary timings makes huge difference in performance😉
I just tested your suggested setting on my AM5 Ryzen 7500f. It does give some improvement to memory performance. There is about a 5% increase in Read, Write, and Copy from the previous setting of auto (12346) to this exact number, 65535.
Latency is a little bit better too. Thanks for sharing.
If you want a test that'll *actually show a difference* try using Starfield or Star Citizen. Both have massive improvements from memory latency, even on X3D CPUs.
I feel like you should've mentioned more about making sure TREFI is stable, not all kits will just work.
It's one of the harder timings to be sure is stable and can corrupt your OS slowly if it's not. Too many noobs will watch this, max it and start gaming.
Worse performance with higher Trefi is an STRONG indicator that the ram tune is not stable! The test then is invalid and should be rerun after Karhu 6000%+ testing.
I had a 2x32GB DDR5 6400 CL32 Hynix A-Die kit and 65K TREFI is unstable (errors on VT3 and N63) unless I raise voltage from 1.4V to 1.435V
Yeah, and the heat from the GPU can cause instability too from high case temps. I'd ideally test with some GPU load to check if the RAM would be stable given additional heat.
I also want to know HOW it could make it warmer
😮
Can you test the same kits on an AM5 platform, those Ryzen CPUs tend to favor lower latency more than clock speed, maybe you would get a much wider degree of results.
I second the request for 7800x3d testing. Would be great to see how frequency and latency affect the AMD chip.
X3D chips don't care thta much about memory
@@MaxIronsThird that's what I understand. It's one of its advantages. Still, I'd like to see it too.
Tell AMD to make better memory controllers compared to Intel's. AMD's IMC's have been crap shoots for a while.
My 2c, when in break creeping up on every frame,learning where to eke out the performance is just not for today but also future systems. I too would like to see the 7800x3d testing. Given that testing, there are games that really are sensitive to memory, that being DCS AND MSFT 2020 both the flightsims , with dcs being more difficult to get consistent results. Totally free with one of the pov in earlier comments that hours of testing for a few frames is not well seen when you are talking in the hundreds of FPS realm but when we NEED those xtra few frames in VR , THEN YES IMHO it could be with it. 7800x3d is the king of the hill for the above 2 titles and memory timing /performance is the next hill to climb.
Marginally
Love your videos, Short and very informative. Thank you Sir
i would love to see a similar video with the 7800X3D. Love your videos :)
With any X3D it matters even less.
No comment on CAS latencies are in the unit of clock cycles, and cannot be compared straight off when you have different speeds?
To compare them you need to frequency normalize. The 5600 kit has cl 38-52.7-52.7-111 when normalized to the 8200 kit. Values are multiplied by (8200/5600) in this example.
This was not a test of low latency vs high speed and which one of those matters the most. Those two kits have quite similar latency in ns.
This was a test of high speed vs low speed with similar latency. With one option of both low speed and high latency, clearly performing the worst.
I would have loved to see a test of actual high latency but fast transfer speed. Something like 8200 with 40% higher cl values, 56-72-72-180, where it actually is slower than the 5600 kit.
All I was able to learn from this test is that MT/s matter a little, but not much in gaming. And that using both bad timings and bad transfer speed has a noticeable performance hit.
pretty much in all tests derbauer was gpu limited, valhalla is infamous for being gpu bound game, esports was tested at 4k (huh?), ram speed absolutely matters, latency and bandwidth, im not sure why he tested at such weird settings, he should have used 720p for all the tests.
@@h1tzzYT CS2 and Valorant even with 4K are extremely CPU limited. The GPU load is 50-60% and I never test 720p because it's a useless and only theoretical scenario. I want people to see what it could actually do for them and not what it might do in a scenario they will never see
The primary timings make virtually no difference to the latency with DDR5 when compared to DDR4. tREFI is probably the single handedly most impactful timing, and (For Intel), I'd argue tWRWR_sg is the next most important timing.
@@der8auer-en But that makes the entire test, and this video, irrelevant. I greatly enjoy your content, but I strongly agree with @ealmen. You missed the mark with this video, by a wide margin.
@@Zhunter5000 Be careful with tREFI, overdo it and you will experience SERIOUS memory corruption ***that RAM tests will NOT pick up on***!
I speak from experience. It's a good way to corrupt your OS and, worse, your data.
This is a very extreme example comparing 6000 CL30 and 8200 CL40 because the latency is actually lower for the 8200 at about 9.76ns vs 10ns for 6000 CL30 so you're getting the additional bandwidth and lower latency.
This is because the timings are in clock cycles and every cycle of the 8200 kit is 0.244ns while the 6000 kit is 0.333ns. Clock cycles are not a measure of time and lower number doesnt always mean better.
Good way to tell is actually taking the frequency and dividing it by the CAS Latency. The higher the number, the lower your latency. Dont buy 6400 CL40 over 6000 CL30 just because the clocks are higher, your performance might actually be worse with 6400 than 6000.
6000 boot so slow mine still 4800
Could you please give me your opinion on klevv ddr5 7200mhz CL34?
all those dry and theoretical calculations are pretty much meaningless as CAS timing is only small part of the total ram latency, subtimings are just as important if not more. Derbauer used pretty much AUTO subtimings which will effect the latency by a massive margin.
CAS latency doesn't matter at all, It's trcd that you would need to take into account
also higher frequency means IMC clocked higher which leads to lower latency from the IMC
With RAM are are only 2 things where higher is better absolutly everything else lower is better
Frequency and tRefi
Thats it those 2 higher is better
Absolutely everything else you want lower not higher
Thank you! Much appreciated:)
You lowered the intro volume! Thank you!
Great video as always. I would always recommend the Tgroup sticks btw. 7200+ will always be A die and the stock heatsinks actually cool the best compared to other brands. Also, you can generally tighten the timings up a lot more than demonstrated in this video.
Please keep up the good work on your WID-eos.
Github MemTestHelper / DDR4 OC Guide
Keep in mind that running max tREFI can corrupt files, so tread with caution.
DDR5 can run a much higher tREFI than what he ran. 65535 is a safe value for 99% of sticks, and the performance dropoff makes higher values irrelevant.
@@Zhunter5000depends, my 32x2 6000 cl30 kit causes corruption even at 25k.
@@DaltonP386 that's most likely because of your sticks getting too hot, not refresh rate itself.
I had similar thing going when i overclocked my hynix m-die kit. When i set 65535 trefi, ycruncher vt3 would error out in 56 mins on 28th loop because sticks got too hot by that point. Simply installing a 120mm fan, held by the zip-ties above the ram sticks solved all instability issues (6h of vt3 + 4 months of daily use w/o problems)
@@Zhunter5000he did run tREFI too in the gaming benchmark.
@@DaltonP386 Your RAM temperatures are probably too high and that's causing the issue. I saw something similar happen in a PC that didn't have good airflow over the RAM. I put a fan near the RAM and all of a sudden 65535 was stable and passed multiple passes of Testmem5 and y-crucher 5b.
Thank for this!
Very curious how well this works with APU's, like the 8600G etc.
Since the RAM on an APU system is also working as VRAM, you should aim for memory that's as fast as possibe.
Awesome content, thank you !
Can you test the same for AM5, as Ryzens till date were more latancy then frequency dependent?
How does the extreme memory tweaker part of the bios @ 2:57 looks like that? Is it a new bios version?
It would be really interesting to see what you could get from properly tuning those kits. Just how tight could you get them manually.
Thank you!
0:59 -- thanks. Never had thought about lower latency ram being more easy going with non-ultra mobo's and cpus.
I gravitate towards a little upbuying of RAM latency before upbuying clock speed. But without going totally nuts.
I do know that if you hit the upper vram limit on a gpu, the system will look to ram for extra storage. So, at higher resolutions is where the biggest differences are. That is where having more ram with a lower end card is good because you will run out of vram faster. Even if you use dlss or fsr, I have noticed similar vram usage to that of running the higher resolution natively. So this test is interesting and the results make sense.
A cool benchmark to add would be running local LLMs, especially CPU only, as it is really sensitive to RAM performance in a multi threaded workload.
DDR5 is 2 years old but still a pain in the neck from time to time. Running 4 sticks to get 128GB with XMP enabled is still a problem. Reboot delay and slowness with expo enabled on AMD processors are still going on. The failure rate is much higher. Compatibility issues are still a problem as processor IMC is something we can only find out after testing with actual memory kit. The last thing a normal person requires is an unstable system that crashes randomly and corrupts OS to get from 179 to 189 FPS in some select games.
DDR5's a complete shitshow on both AMD and Intel when you put in 2 DIMMs per channel though. Definitely wouldn't recommend going with a total of 4 DIMMs for just gaming
while I agree it is unacceptable that 4 dimms are almost unusable with ddr5 but I'm not sure who's to blame. I think it would be safe to go for a 4 dimms kit if you need more than 128gb ram. in other cases just use two dimms. but the way I see it 2x64 should cover most power users
those fans that you can daisy chain directly or with the cable are sick. nice to see seasonic in the fan market
The 48gb 7200clv34 kits out there offer the best of both worlds in speed and latency imo and a pretty good sweet spot for people who don’t have a super binned board or cpu to run 8000
+1. The video is misleading and doesn't properly cover the topic. 7200CL34 is 9.444ns, versus 8200CL40 at 9.756ns and 5600CL26 at 9.285ns.
@@noxious89123 Cas latency is not a relevant thing to compare with DDR5 as it in of itself does not act the same as DDR4.
All those hours going in and out of the BIOS for 3 more frames
Been there done before. Lol.
Don't really recommend
Got like 15 more frames but was it worth a week of testing every single subtiming ? I'm sure I don't want to do this again.
@@club4ghz its fun if you like to tinker
But definitely does not help much for 4k gaming
why even visit this page if that is not something you are after ?
@@AcidRP probably because he likes Debaurzz content? Just a random guess
I have a Gigabyte B650 Aorus Elite AX, and if you enable "High Bandwidth Support," it does this tweak for you. Mine's was set to 53032 because I enabled it earlier.
Does AMD still have larger performance swings based on memory kits? Maybe this revisit on 7000 series could have different results.
It's a funny comparison, because when translated to actual time, the kits you compare are:
DDR5-5600 (9.3 ns - 12.9 ns - 12.9 ns - 27.1 ns CR2)
DDR5-8200 (9.8 ns - 12.7 ns - 12.7 ns - 31.2 ns CR2)
So the 5600 MT/s kit doesn't even respond faster on all the subtimings, it's actually a little slower on the trCD times, even though the difference is more than compensated by the shorter CAS latency.
But yeah, in generally I don't think the memory timings have ever been massively relevant since the DDR2 times, when latency difference were massive and CPU branch predictions were bad or non-existant. Modern CPUs will just keep the caches filled with all the likely data most likely to be needed next, so the actual ALUs are rarely waiting for a RAM access.
And I guess nobody is even bothering to get to CR1 anymore, I think this still could give you a few tenth of a percentile of performance on DDR3 systems, but the overhead due to double-commanding is probably entirely irrelevant now when latencies are well into the double-digit numbers.
Hello Der8auer, I have a technical question to you. A few years ago Steve from GN made a test on different RAM configurations, specifically 4x16 vs 2x32, DDR4. I remember he was saying in the conclusion that by running 4x16 configuration you generally get about 10% more performance. Do you think this is still relevant today for DDR5? If yes, do you think it would affect the results of the memory bandwidth tests?
Hi der8auer I see you left out tuning of trefi x9 at same time when tuning trefi, I tuned trefi x9 and trefi on my 8000-40-48-48-128 kit and saw a further reduction of latency again, I have a 8000-40-48-48-128 48gb kit doing 52-53ns latency with a trefi x9 and trefi tweak whereas they will only do 58-60ns when tuning just trefi alone.
I'm running 64GB GSkill 6000 C30 in a z790 Dark Hero/14900K and honestly, I'm quite happy where I'm at. I don't see any compelling practical uses in my case for pushing any faster.
Just a small remark, but moving the mouse while Aida64 is testing, skews results for the worst.
Given the almost rage inducing stability issues I recently had using baseline settings, I won't get anywhere near this.
It's a nice and educative video though, that gives a pretty nice use case scenario for enthusiasts. I myself find stability, much more enticing than one digit performance boost which case, to be honest, I wouldn't be able to notice any difference.
Thank you!
nice, but why those high settings. even with high end gpus I dont use high settings in e-sport titles. medium at max and low without any res scaling.
righ now am using 7800c38-46-46-96 with 12700k/z790itx.
The ram timing become much more important on games/programs with a very heavy RAM usage for many small data sets and where the CPU cores (for the program) are getting maxed.
effectively, the CPU/GPU busy thing also happens to RAM, so lower latency effectively means it can handle those memory calls faster and move to the next computation sooner.
this is especially true in physics simulation games that peg only a few cores like KSP (1), where the engine is handling complex math with many interworking variables that the CPU has to calculate a few times every frame.
if the games simulation rate happens at say, 200hz, that means on average we're making 3x simulation passes per frame at 60fps. if the game has to run gravity, velocity, rotation, collision/interaction, aero, thermal, thrust/thrust vector, lift, and power calculations on EVERY PIECE of the aircraft 3x every frame... yea, thats a LOT of math that handles floating point numbers stored im memory. latency is much bigger here.
The most important subtimings:
tREFi - 65535 (If it's unstable can be used 52224 or 32767)
tRDRDSG - 16
tWRWRSG - 16 (or 8 if it's stable) everything between 8 and 16 gives worse performance than 16
PowerDown - Disable
Love the vids
👍🏆, ty
I bought some G Skill 6600mghz DDR5, and i can't even enable XMP without BSODs. I'm moderately confused.
Surprised you didn't clarify there's a bigger benefit to low latency on AMD. For X3D it doesn't matter much in games but vanilla Zen4 sees decent gains. At least DDR5-6000 C28 (Hynix) with FCLK 2000-2100 should work on most systems, even with geardown mode disabled.
Zen4 can run up to 8000 in 1:2 mode now but it's not worth it for gaming and not every board can. Even tuned 8000 C34 on my 7950X is only marginally lower latency in AIDA than the ~53ns of a tight 6200 C28 config but ~10% higher bandwidth.
i saw something interesting. the name on the kit says Xfinity, there is a usa internet company named that, its Xfinity by Comcast. hope that company doesnt get in trouble if its in USA
When I was interested in DDR4s, I preferred Micron-E die chips because they overclock very well. What's the situation like with DDR5?
For ddr4, I believe Samsung B-Die was the best one. For DDR5, I believe the current best performers are Hynix A/M-dies, though I haven't kept eye on the oc scene of DDR5...
Hynix A die, M die is very close but A die is still the best
@@renerant thank you. Where can I find list/pdf of which brand and model use which chip?
Hynix A or M die. I think I've seen better results with M die but haven't tested them yet myself
@@GoldPunchI know that G.Skill at least shows who’s SDRAM they’re using on the kit’s PDF. I can’t comment on any other brands. You usually need to look at the timings and could check Builzoids what RAM to buy.
8200 C40 already is a low latency kit. It's the closest you can get to the same as 6000 C30 and 5600 C28 without being the exact same. Thus the C26 5600 kit is only marginally better. Though, sub timings and tertiary timings would have an impact on hierarchy as well as raw megatransfer speed.
Curious how these kits behave with Ryzen 3D v-cash
Does frequency over latency hold true for ryzen?
Honest question... shouldnt AMD benefit more of the lower latency kit?
Not tested with the lowest graphic setting? I thought it'll help huge difference in performance when tuned RAM as long if it's fast as CPU tuned.
@9:24 Shiek offers their advice
It was in disagreement.
More of a "where are you" than disagreement
@@JohnSmith-zu2sy They just want to know if tREFI can be used to adjust time between pets/feeding
Very interesting video on especially with the memory Trefi setting. What’s the latest on the Intel mycro die kit has it been released?
Would be nice to see what the difference is if any on 7800x3d.
Test it out for yourself
@@elitepauper7400 We don't have money to purchase multiple of the fastest kits.
X3D cpus benefit less from memory latency cause they have to reach memory less frequently because of the extra cache .
But non X3D Ryzen CPUs benefit a lot from ram latency , higher than intel cpus .
I was hoping he tests 7600x or 7700x for this video .
Intel cpus like to have higher bandwidth ram .
The 7800X3D will always be limited by the Infinity Fabric speed. For example, at 2000Mhz it will only do 64GB/s of read speed.
So even if the memory can do much higher speeds, it won't matter.
Because of that, with Zen4 CPUs that have only one CCD, it's better to focus on memory latency.
The difference isn't much, but it is better. however, I did find a sweet spot of 58880 tREFI on my 7800x3d + 6000CL30 vengeance. I found both lower and higher tREFI to increase the latency.
FCLK 2166
i wish you made the amd version too, it is more sensitive to memory timings
6000 CL30 is what you need on am5
@@ODIOPOWERunless you’ve got an 8000 series APU, then you can run much faster memory
Great all the speed we can handle. Yet, I thought there was a relationship between pin burn-out and xmp profiles, mixed with high performance ram - while supported on the motherboard QVL will actually void the warranty of the chip (Intel) I am a system integrator, while I desire the best performance possible for our high end workstations creating failure isn't an option. Was purchasing K sku's for a long time until the recent problem - mixed with the frame issues is a giant mess indeed.
i don't think higher trefi will increase temperatures if you don't also increase vddq.
(might be necessary but didn't explain why in the video)
My two z690-i systems' latency @ 55-53ns: 14900K + 3090OC + 96GB @ 6600CL30 9.09ns, 12700K + 3070OC + 64GB @ 5800CL26 8.97ns. You want system latency under 60ns to feel good, 55ns better, 50ns best, 45ns- golden. DDR5 is typically 10ns, Windows 60ns. DDR5 can usually drop to around 9ns before heat becomes an issue. I saw demo videos on how to make hardcore edits to Windows timing structure, basically cutting OS latency down to a Linux-like level around 30ns; I have not even attempted this as it is really well beyond my understanding at the moment, but is an incredibly interesting idea.
"you want system latency under 60ns to feel good".
You are living in your old false reality. A make believe world in which you are able to see the difference between 10 nanoseconds and not.
You don't perceive any difference. Even if it was 58ns vs 102ns, you wouldn't notice anything.
Interesting video and unsurprisingly the conclusion is that it isn't worth it except for the ultra enthusiast who must have the highest end gear and also win the IMC lottery in order to actually hit 8000+ MT/s.
Also anyone with x3d just not giving a f lol
I have an AM5 motherboard, 7950X3D, and 8000MT/s DDR5 RAM and it was a two week long trial-and-error period of getting my 8000MT/s ram stable with tightening primary and subtimings. The increase in performance is about 18-20% in almost every scenario, but I also overclocked my timings to the absolute limit.
In order to achieve 8000 MT/s, you have to buy hardware that can support it and can push those frequencies in addition to providing support.
It's just not worth it. The difference between 7600MT/s and 7800MT/s is 2% and 7600MT/s and 8000MT/s only being 4%.
getting 104gbs on 6400 cl32 and 58.2-59ns with memory try it and msi memory enhance can get slightly better with manual timings but pretty good for board settings.
I recently built a high end system. My RAM came with the combo I got. I am running it at 6000Mhz 36/36/36/96... It's average I guess but it works fine for my 1440p gaming. My motherboard is rated for up to 7200 with dual rank but do I really want to go messing around for 5 more FPS
This tweak can work with ddr3? Thanks
Most important timings as always tRCD, tREFI
Laughs in tRRD and tFAW
Higher frequency then manually adjusted timings for best of both worlds. Just takes time to dial in
👏👏👏
I would love to see this test for an AMD iGPU (i.e. 8700G)
ty
I know Valorant is extremely light and cpu dependant but 947fps at 4k is crazy. Asus launched a 500hz monitor when we already need a 1000hz one
Why did you test cs2 and valorant at 4K max settings? That isn’t how people play those games. 1080 or 1440 low/med would have been a better test.
Showing the cpu bottleneck i think
@@PaulBlartGamingPlaying 4k passes the bottleneck to the GPU more often than not.
If he had done 1080p low settings test, you'd see more of a CPU impact.
I get playing maxed out and seeing if the RAM can make an impact, but to make a larger CPU impact, run low settings.
@@MyersComputerServices if he had tested on amd you would see a larger impact, also in cs2 and val its a cpu bottleneck anyway so it doesnt matter. %dif stays the same when you are testing things to make the cpu faster if the cpu is the bottleneck the whole time.
There wasn't said this is valid for Intel only...for Ryzen it is the other way around, lower timings make much bigger difference than higher frequency. Especially tweaking not only the Primary but especially the Secondary timings for Ryzen makes huge difference in performance😉I would like to see the comparison also for Ryzen with those low latency sticks....
@der8auer-en I would love to see this test on an AMD system. What's good for the Goose is not always what is good for the Gander..
Why does the temperatur of the ram increase with higher tREFI?
Where did he say that? The issue i
with high trefi is that tolerance for high temps decrease
@@mircomputers1221 4:55
The ram refreshes more and thus generates more heat.
@@happymountain2050 well that is just not the issue, increase in trefi makes memory more sensitive to temperature, doesn't make it hotter. But let's say your ram can run at 70c with no errors at default trefi, you might need to get it to 50c to have stable at 32640
@@Zhunter5000 refresh less
I have 5200 jedec ram, i overclock it to 5800 30-36-36-72 at 1.3V and its stable for stress testing and lot of games,
however when i try 6000, its unstable even i put loose latency 40-40-40-96, im so confused
My specs Ryzen 5 7600 and Asrock b650m hdv
Try 1,4-1,5v.
This only dips slightly into the art of memory tuning. If you really want to get into it, be prepared to spend a lot of time and effort. tREFI looks a quick win, but it’s not like you can set it to 65K and call it a day. As mentioned, increased temperatures can cause stability issues with this setting. And then you’ll end up in the arduous process of trial and error that memory tuning is anyway.
And latency tests?
Good video...I really need to learn how to OC ram...I know few of the basics, but I need to take the time to really understand what all the tertiary timings actually do...
It’s not worth the hassle believe me 🥲
Adjusting secondary tertiary timings on an already powerful system gives a small boost.
The main increase comes from frequency, primary timings and tREFI. In total they give about 10-15% increase.
Everything else affects but within 5% of the overall CPU performance in games.
If you need it for sport, or you want to get the most out of the system, then try it. But remember, without powerful cooling of memory strips, good timings cannot be achieved. tREFI is especially sensitive to temperature, despite the fact that it also gives one of the most significant gains from secondary timings.
To get tREFI 130560 or higher, at 8000+ memory clocks, you will need to liquid cool your memory.
For air cooling it is better to stay at 7200-7600 MHz, but it is good to tighten the timings. This way you can get the highest performance in games.
@@MrJB327 I'm not wanting to learn just for more performance...I enjoy just tinkering! More performance is just icing on the cake! I bought a couple of super cheap kits of ddr4 to play with and actually ended up getting almost 20% more performance out of one kit and I didn't have much idea what I was doing...DDR5 should be fun, but I just want to learn the more technical aspects of it instead of just poking around in the dark so to speak...
@@IIHydraII I just want something new to learn and have fun with...I'm sure the performance gains aren't worth the time invested, but that has been the case with almost every CPU and GPU i've overclocked in years! lol
@@brucepreston3927 Then good luck to you. DDR5 requires more fine tuning when overclocking to get stable operation. So it's an interesting experience.
I'd honestly expect the CPUs to be designed to take into account the relatively high latency with all of the prefetching they do to fill up cache.
Hilarious that you say that bigger tREFI warms up modules.
EDIT: After trying this for almost a week I found it creates massive stability issues. Graphical glitches on desktop. Chrome crashing. Programs not loading properly or loading but not running properly. I wouldn't recommend this.
I tried this with my A-Data DDR5 4800Mhz (2x 16gb) running XMP1@6000Mhz. XMP1 timings are 76-40-40-40. AIDA64 reported 73.1ns with stock timing. Then 64.5ns with the tREFI tweak. I'll try this setting seeing as I see a big difference and see if theres any problems with daily use. Using a 12900k and 3080Ti on a Z690 Gigabyte Aorus Ultra MB. .
I wanna see a video of DDR4 last gen cream of the crop, 4000mt cl14, overclocked vs DDR5. Personally by my math my 4266 CL14 will cream that 5600 cl26 except for bandwidth. There will be a cap in bandwidth for games, you don’t need 90gb/s and latency starts to matter more. It’s not loading the whole game in 1s. The 8200 CL40 was the best because of latency not everything else. So it really goes back to what has been true since ram was invented, latency, I mean that’s why it exists in the first place and that’s how software wants to use it. This is a “tweak” for gaming where the data is constantly refreshing, try solidworks or 50 chrome tabs and see how fast the BSOD will get you
My next system will likely be AMD based, so if Zen 5 follows Zen4 then 6000/CL30 is the only worthwhile option.
2:27 59 "ms" would be uselessly slow.
It's a stock windows with no changes :) that's the most realistic scenario for normal users
@@der8auer-en I think you've missed his point. It was 59 nanoseconds, not milliseconds! You merely misspoke in the video, and @TuxikCE was picking up on that :)
What's a factor of 1000000 between friends.
Latency meens nothing in isolation of Read/write speed, dude
milli vs nano
I’ll stick with my 4000 cl14 ddr4 dual rank kit, latency of 45ns in Aida when tuned.
i thought the trefi change in bios was well known thing to do
You might weant to try on AM5. 8200C40 1:2 and 5600C26 1:1. Probably 1:1 scenario could be better since uCLK is faster. Regards
i don't think the amd cpu can handle that speed through?
@@Bu11sEyes newer AGESA versions allowed it, but as op said, 1:2 mode.
It was tested by buildzoid and was only slightly faster in maybe 1 synthetic benchmark. 6200 was best for most (not sure if there have been more tweaks since I last watched it)
@@richmanricho6400 1:1 was better than every 1:2 speed except 8000. 8000 is objectively much faster in gaming than 1:1.
Back in the 90's I was able to get my hands on some DDR 1 RAM with CL of 1.5. I could run it at 1.5, but was able to get more performance running a higher clock at CL 2.0. Latency always mattered, but speed(Hz) always overcame increased latencies. Still holding true!
Makes sense but clocks don't always win look at 3dvcach chips low clocks but 180GB per second bandwidth
@@mddunlap03 True, but cache is another story. Ever try burning a CD 2x with only 384kB of cache? You get a coaster. It's more like a capacitor keeping data flowing without interrupt due to a slower process in the line.
CL26 is approaching DDR4 server DIMM latency. That's pretty impressive for the speed.
Fwiw, you can't calculate latency from the timings alone. They're not an absolute value. RAM timings are counted by the number of cycles, and higher speed RAM completes more cycles in the same time versus lower speed RAM. 8200CL40 has a lower absolute latency than 6000CL30, despite 6000CL30 appearing to be a lower latency because of the lower number. 8200CL40 is 9.756ns, and 6000CL30 is 10.000ns.
But notice the second and third timings are 36. There are CL30 DDR5 kits that are 30-30 instead
what about AM5
that logo looks exactly like xigmatek
Instead of using 14900KS, should have used a CPU with less cache like 14600K or Ryzen 7 7700x which is known to heavily depend on RAM timings as shown by hardware unboxed tuned RAM testing
Could you test which is better with the multiple memory sticks to have 128GB or 256GB, low latency or higher frequence?
Should do the same test with amd cpu as it common knowledge that ram speed make a bigger difference also thay have a sweet spot on speed and latency
Not to disrespect, but I would expect from famous overclocker with 15 years experience (like Roman) to know about tREFI already.
I'm waiting for ddr5 10000
ddr5 7200mhz cl34
2:28 nanoseconds*
🖥 😁 👍 ⚡ 👌🖥
if you tweak tREFI, why no do tRFC at the same time? ;) tREFI for less refreshes and tRFC for shorter refreshes.
For desktop, Samsung or crucial or hynix with custom tweaked timings should have more value and performance. For notebook with limited BIOS tweaking options, Kingston hyperx,ripjaws and Corsair dominator worked very well with Dell, HP, Lenovo, Intel nuc, clevo out of the box with jedec or xmp or expo profiles.
One more thing, if standby memory is frequently cleaned it could give worse performance or stuttering with higher trefi values i.e less frequency of memory check
Crucial? Afaik they don't make memory ICs, but sell Micron IC based dimms. I know one of them owns the other, but still, micron is the IC.
@@renerant crucial is consumer side whereas micron is the parent and well known within the enterprise.
from my kit 3200mhz to 3600mhz incrase almost 10w on the system... on my little 35w cpu :S