I had 7950x with X670e Crosshair extreme an 96gb of Team FF4D596G6800HC36DDC01
I had endless issues with random reboots I struggled to get the memory to run at 6000MHz at 1:1 and the cpu performance was terrible at best I could manage 35000 on cinebench R23, but the performance was erratic and at times went as low as 32000. Single core I barely broke 2000. CPU wouldn't tolerate anything more than 5 for undervolting.
So I got a 9950x thinking I just got screwed by silicone lottery. My boot times became extremely long like 5-7min memory training times (the first time it booted I thought the CPU was broken). Still struggled to get 6000MHz at 1:1 but it was slightly more stable but CPU performance was still erratic with Cinebench R23 scores ranging from 34000 - 41000 and single core 2000-2100. This CPU did allow 15 for undervolting but it made no difference to performance though temps dropped slightly.
I tried some gskill memory but that made no difference.
So when the X870 motherboards were released and I saw ASROCK NOVA x870e had 5 M.2 with no sacrificing of graphics PCI-e lanes for a frankly ludicrously low price (in comparison to crosshair extreme) I jumpedat the chance.
What a difference, I am now running my memory at 6200MHz @ 1:1 with no problems at all. My CPU undervolts easily - currently running at -20 but will do -30 no problem. It boots quickly and best of all my cinebench R23 scores are steady around 44000 and 2240.
I thought I was just incredibly unlucky with CPUs but all that time it was just the motherboard that was garbage not everything else.....
My partner has 96GB on the Crosshair working fine.
1. If you don't plan on tuning manually, perhaps you should have referred to the QVL. ASUS tends to use tighter memory sub-timings than other vendors. Moreover, the X670 Crosshair doesn't have any 96GB kits validated at the frequency of the TG kit you were trying to run.
2. Most importantly, the ASROCK NOVA x870 is an entire generation newer than the Crosshair. Newer memory topology, chipset, etc. It's the progression of things that makes these things easier gen on gen...
> 6000MT @ 96GB on X670 vs X870 is apples to oranges...
3. You purchased a 9000 series CPU and put it into a previous gen board, only to then purchase a newer generation board from another vendor. Again, apples to orange (the board vendors design the board and memory topology around the CPU it launches with).
In short, don't be so quick to judge vendors for your own knowledge gaps.
@DKnell wrote:Yes of course - my knowledge gaps explains why both processors were so poorly performing despite using PBO, curve optimizer, curve shaper, infinity fabric optimization etc. It's my knowledge gap that explains that a cpu that will tolerate -30 undervolt on ASrock wouldn't run stable on anything more than -5 on the Crosshair extreme.
Yes, most likely. Did you compare load voltage? What about LLC settings? I know many people on the Hero who aren't experiencing difficulties with such things. I'm not sure why you think it makes sense to compare boards across generations. Do the media do such things and then chastise? No, of course not. Even they know better than to do this.
@DKnell wrote:
I tuned the memory using every resource known to man - even buildzoid's hynix timings didn't work.
Not sure what "Buildzoid's hynix timings" are. Memory timings are industry standard, and beyond JEDEC, module tolerances differ. He's not a memory vendor, he can't make assurances that any timings he suggests will work.
@DKnell wrote:
"the board vendors design the board and memory topology around the CPU it launches with" - your ignorance speaks volumes.
Not at all - this is a fact you will have to learn to accept. Board vendors constantly change the topology and layouts to better suit the CPUs they launch with, even in instances when it's simply a refresh. Might not be common knowledge but it's often the case. Sorry if this doesn't sit right with you for some reason.
@DKnell wrote:
I have managed to get the same memory kit working on my son's gigabyte aorus extreme x670e motherboard @ 6200MHz using my old 7950x. So does that satisfy your apples and oranges?
In short perhaps you shouldn't comment if you don't understand the problem......
You mean like this?
The fact is, manual tuning at these densities is often needed. Perhaps you can improve some of the latency on the Asrock by making use of some of these memory sub-sets if you can get them stable.
Hope it helps.
@Alpine_Alex wrote:You mean like this?
The fact is, manual tuning at these densities is often needed. Perhaps you can improve some of the latency on the Asrock by making use of some of these memory sub-sets if you can get them stable.
Can you show me those timings with a fresh run on AIDA 7.40.7100 & on bios version 2403, as your North Bridge Clock seems to be running a little slow there, this version of AIDA from early 2023 was buggy on AM5.
While you are at it show me with 7950x or 9950x which are the processors I was referring to as we don't want to compare apples and oranges now do we. I also see no Cinebench R23 run showing your CPU performance either, which was my primary complaint, as CPU performance was poor and extremely variable even across 2 different processors, including one it was supposedly "designed" for.
Looking through your timings there doesn't seem to be much gained except I haven't pushed my primary timings as far but other than that they look oddly similar so maybe I really don't know what I'm doing. I did bump up my Fclk to 2133 just so we are attempting to compare apples.
Clearly my crosshair extreme was garbage as it wouldn't run this memory at even very very conservative timings and also had extremely variable cpu performance.
This is stockfish, a very accurate and reliable benchmark that utilizes your CPU AND memory.
For this benchmark the difference is HUGE!!
Asus X670E Extreme.
Asrock X870E Taichi
