From what i gathered there seem to be a hope that X870 X870E motherboards might be able to run DDR5 8000 with BIOS update.
Currently i am planning to buy AMD 9900X and *ASRock X870 Pro RS WiFi with G.Skill DDR5-8000 CL40-48-48-128 1.40V 48GB (2x24GB) F5-8000J4048G24GX2-TZ5NR kit (AMD EXPO) that listed in the motherboard QVL.
What the chances of this happening and which motherboard should i buy? I prefer ASRock, MSI and try to avoid Gigabyte, ASUS. I don’t really need the connectivity that X870E offer but I heard that beefier VRM give more room for memory overlocking, should i buy more expensive X870E board that has better VRM to get a better chance of getting DDR5 8000 working in the future (BIOS update) or stay with the cheaper X870 board like the RS Pro Wifi.
There a few options like:
ASRock X870 Pro RS WiFi : Has the F5-8000J4048G24GX2-TZ5NR listed in QVL but not as good VRM
ASRock X870E Nova WiFi : Does NOT have the F5-8000J4048G24GX2-TZ5NR listed in QVL but has better VRM, a bit cheaper than the X870E Taichi Lite
X870E Taichi Lite : A bit too expensive for my budget, much better VRM
I do programming (heavy C++), want to learn 3D modeling, lightweight video/audio editing, not a hardcore gamer.
Build: pcpartpicker (.) com/list/MjpfrM (can’t include link for first post, sorry). the GPU is just placeholder for NVDIA 5000 series. total budget around 2800$~3000$, if i can save money on motherboard i could spend it else where like audio equipment or better GPU.
I think DDR5 8000 will help with these tasks ?. or should i play safe and go for 24Gb x4 DDR5 6000 kit instead. Will having DDR5-8000 working reliability motherboard/memory lottery? I am not a pc builder for a very long time, much appreciate if you guys help out with my questions, thank you.
This is kinda depending what is actually important to you.
For example do you need ecc support or not really?
Because last time i checked when the X870E / X870 boards were released,
the Asrock boards seem to struggle a bit with 8000mhz memory speeds.
However this could already been addressed with bios update by now.
Or that definitely will be addressed in the near future.
The Msi X870 Tomahawk wifi does seem to work well with 8000mhz memory.
However Msi does not support ECC memory on any of their main stream boards.
But of course it´s not only the motherboard but also the said ram kit that actually matters.
That is why there are QVL lists.
AFAIK DDR5 8000 is still a bit of a lottery. If you get a kit rated at 8000 it should be capable, but the board and CPU might still be a limitation. It’s supposed to be better on x870 boards but it is not a foolproof guarantee. You should get better success on higher-end boards (generally) and on 2 Dimm boards (better signal integrity). I would check out the overclocking forums (e.g. overclock.net) to see how different boards OC.
The performance advantage depends on the workload, and sometimes 6000/6400 might even be faster:
The reason is because the memory controller runs at the same clock as RAM at 6000-ish (up to 6400 if you win the silicon lottery but 6000 is safe) but at half speed for higher speeds. You still gain bandwidth but lose a bit of latency. Spending some time tuning can offset this difference though.
Given that phoronix’ compile benches seem to run better with 6000/6400 that seems like a safer bet IMO, but it all depends on how much time you’re willing to spend tuning and how critical stability is for you.
How much memory do you actually want/need? 96 or 48? You won’t be able to do 8000 any way with 96. Additionally 2x48 will be faster and more stable than 4x24 since having 4 dimms makes the signal integrity worse.
Part of the answer here’s X870(E) boards generally support DDR5-8000 at 1DPC 1R per OP query. As @quilt pointed out, the CPU may be the limiting factor.
As to whether it’s worth it, TechPowerUp’s also looked at Zen 5 memory scaling with similar results. There’s little data on associated power scaling but all of what I’m aware of is on Level1 as well.
MSI’s clearer with their specs than ASRock, Gigabyte, and Asus.
If there’s data comparing quad 1R to dual 2R I’m not aware of it, though MSI specs no difference between the two. As others have pointed out, dual 2R’s the accepted practice.
Probably not intrinsically. Quad DIMM terminates what would otherwise be open stubs from the unused DIMM sockets. Available quad data’s almost all 2DPC 2R, though and, despite lack of an evidentiary basis to do so, it’s common to conflate that with 2DPC 1R. The type of DIMM socket used likely also affects comparisons between 2DPC 1R and 1DPC 2R.
Another question I have here’s whether CUDIMM helps any with 2DPC. Haven’t seen anybody look. FWIW my guess (coming from layout experience of prior DDR gens) is there might be some benefit despite the lower clocks in 2RPC configs. Doesn’t mean CUDIMM’d be useful in making the jump from 6400 to 8000, though.
I don’t know if i need ecc or not. I am writing my first game engine, so my workload is similar to compile UE5 or Chromium. I don’t think i heard anyone need ecc in game industry, and UE5 recommended memory is 64GB and not mention ecc.
I don’t know if Asrock boards has fixed it or not, the only indication that i aware of is that G.Skill DDR5-8000 CL40-48-48-128 1.40V 48GB (2x24GB) is in the QVL list if Pro RS WiFi and Taichi, Tomahawk WiFi. There is the video from Hardware Unboxed AMD X870/X870E Roundup, 21 Boards Tested where all of the ASRock failed and the Tomahawk WiFi passed but they said those tests weren’t conclusive because they didn’t run the them for long (1 hour Prime95), and i can’t find anywhere which memory kit they used.
If the kit is in motherboard’s QVL list does it mean they will work flawlessly once enable XMP/EXPO?
I hope that the memory kit will be good out of the box and i can just adjust the clock speed after a bios update if they haven’t fixed it, is it too good to be true?
Edit:
Thank you for your help, i much appreciate it. I have more information to make decision now.
Well memory support will be improved as the platform matures out.
There will be several AGESA updates there already is at least one update out by now.
Because of the release of the X3D cpu.
So i would not be to worried about that.
Asrock has a nice board line up good value for money.
Thank you, the benchmark for LLVM compile is the closest to my primary workload and it does seem 6000 vs 8000or within error range that doesn’t seem to be any significant boost.
I am not sure yet, i willing to sacrifice memory capacity with performance boost in compile time, but according to the benchmark above there is not any. I heard that dual rank memory will have some benefits or run more dimms would do the same. 2x48 probably be dual rank right. yes, that is good point that i overlooked, thank you for pointing it out.
Also this General Ryzen 7000/9000 AM5 CPU DDR5 and infinity fabric OC information says that CPU isn’t a big factor when run DDR5 8000 in 2:1 mode because the memory controller will run at 2000MHz in sync with the infinity fabric clock speed which a bad cpu can handle. So it largely motherboard and ram kit problem. I am not a overlocker i don’t know what is correct here.
Edit:
Thank you for your help, i much appreciate it. I have more information to make decision now.
Thank you for provide this information, where i live is quite hot and ambient temp usually 10°C higher than the 21°C testing environment. This thermal limit could be the deal breaker for me.
I don’t trust these spec support because they will fallback to “it technically overclocking, you do it at your own risk”. But i guess it not a unreasonable assumption.
@MisteryAngle reassure me quite a bit, thank you.
The thermal limit makes me even more worry now. Maybe 8000 isn’t a good idea for my use case after all, man i hope i can get some decent performance out of AM5 8000 “promise”.
Edit:
Thank you for your help, i much appreciate it. I have more information to make decision now.
Yeah, YMMV. In general, though, if the specs say it should work and it doesn’t that should be acceptable grounds for a return. If you want something that just works that’s DDR5-5600 dual DIMM.
There’s again next to no formal data but, from the DDR5 builds I’ve done, I wouldn’t expect ~31 C Ta to be an issue given an adequate DIMM cooling arrangement and corresponding choice of tREFI.
They say the extra heat coming the IO die not the memory stick , maybe i misread/misunderstood something?
your eyes are drawn to how high memory OC speeds (DDR5-7200 and DDR5-8000) cause the power draw of the I/O die to skyrocket, exceeding 20 W when under stress
All Zen 5 processors have their thermal limit set to 95°C, and when the 9950X is fully loaded it will reach that temperature, even with a big 360 mm AIO running at full speed.
Maybe it won’t be so bad since they tested 9950X (170W TDP) and not my target 9900X which only 120W TDP which still have some extra head room. I need to do more research if higher memory bandwidth helps with 3d modeling or video editing which are my secondary use cases. If there isn’t any performance boost in those use cases, i probably will go with 6000 2x32GB.
There’s some part to part variation with layout and core performance order but, in general, any current gen processor will thermal throttle with a workload that funnels enough boost power into few enough cores. Core thermal throttling’s also not initially of much significance as throughput’s pretty flat at high boost.
TechPowerUp doesn’t clearly report temperatures but, as far as I can tell, their comments on 95 °C refer entirely to cores. Recall their testing is with PBO (~18 W/core all core) while it appears you’re planning stock power limits (~12 W/core all core). As a data point, the 9900X I use is under a Phantom Spirit 120 and runs +21 °C ΔT on the IO die with 23.4 W SoC power and 65 GB/s of DDR traffic (DDR5-5600). IO die hotspot’s +31 °C. Even with AVX-512 putting 142 W into CC1 I don’t usually see any core going over +70 °C.
But, yes, an inadequate CPU thermal solution’ll back down DIMM thermals due to reduced utilization. TechPowerUp’s basically just saying they’re running a PBO that could use a delid.
Mind also the baselines used for comparison. Probably a majority of the accessible gain’s in tightening timings, which can be done at 5600 without actually overclocking. Turning up the clock from there often brings like another 2-3% which, depending on how active the build is, may not be worth additional idle power draw.
I probably will be quite active with my build so the idle power draw won’t matter too much and luckily the 6000 2x32GB kit is in stock where i live at a reasonable price. Thank you for the insight. this is a better option for my situation, save quite a bit budget for other equipment.
