Poll: Ryzen 7950X3D, 7900X3D, 7800X3D

[Eddie99]

Well for me as my machine is otherwise stable, it’s not something I’m going to RMA over.

When you get random bugs and glitches I bet you'll be blaming the software rather than your faulty hardware.

Yeah just an update to say I managed to get a CPU that passes AVX2 at stock, both in OCCT and Corecycler and this is with the BIOS defaulted. So they do exist. Makes it frustrating therefore that AMD is pushing out CPUs they must know have cores that won't make these tests at stock.

Like yourself I would say if something is crashing at stock and it's repeatable across multiple stability testing apps it's not really acceptable. Especially if you get some random reboots.

I think this is just what happens when there's competition and they are pushing clocks to the absolute limit out of the box, most people aren't going to test it and it will work fine most of the time so return rates will be low even if they're faulty.

 

[Audioboxer]

When you get random bugs and glitches I bet you'll be blaming the software rather than your faulty hardware.


I think this is just what happens when there's competition and they are pushing clocks to the absolute limit out of the box, most people aren't going to test it and it will work fine most of the time so return rates will be low even if they're faulty.

Yeah it's definitely what has happened, the frequency wars with Intel. Because it's CCD1 that ends up having issues and that's the core AMD pushed hard in the 7950x (CCD0 as well on non-3D cache)/7950x3D to get the frequency up a lot higher than the 5950x.

AVX2 is obviously pretty heavy on the CPU and the average person won't care about it/know about it, but for years now it has long been accepted Prime95 is an acceptable industry stability test. Heck, AMD can even ask you for evidence from Prime95 if you're RMAing with them directly around an unstable CPU/crashes/reboots.

 

[McCarthy]

Yeah just an update to say I managed to get a CPU that passes AVX2 at stock, both in OCCT and Corecycler and this is with the BIOS defaulted. So they do exist. Makes it frustrating therefore that AMD is pushing out CPUs they must know have cores that won't make these tests at stock.

Like yourself I would say if something is crashing at stock and it's repeatable across multiple stability testing apps it's not really acceptable. Especially if you get some random reboots.

My experience is a positive one thankfully. My 7950x3d has been flawless, passed all stress tests (even though it took months to finalise) and is rock solid in everything I’ve thrown at it with negative curve optimiser.

I understand people’s frustrations. I’ve had many problems over the years.

If it fails stock on a number of tests, it’s faulty. I would want an exchange in this case.

 

[Audioboxer]

My experience is a positive one thankfully. My 7950x3d has been flawless, passed all stress tests (even though it took months to finalise) and is rock solid in everything I’ve thrown at it with negative curve optimiser.

I understand people’s frustrations. I’ve had many problems over the years.

If it fails stock on a number of tests, it’s faulty. I would want an exchange in this case.

Yeah, this one I have now passes AVX2 at complete stock, so it just proves what I already knew that some chips can do it.

Between AVX2 and y-cruncher 00-x86 FFT only test this is my current PBO values

The 00-x86 FTT in Corecycler with y-cruncher absolutely decimates the 3D cache chips. As in, while AVX2 is a great test to run for finding your curve values, 00-x86 FFT is 100% going to bring down the cache only cores curve further. I recommend everyone tries it on their cache cores. Core 0 for example came all the way down from like -33 with AVX2 for me

Pretty much got my memory tuned in as well now

So I finally got there even though my 'welcome to AM5' was quite a rough ride!

The last CPU was a bit better with core frequency boosting, but it was failing AVX2 at stock and was not FCLK 2200 stable in the slightest. One of the best ways to check your FCLK other than looking out for USB issues is to run LinpackXtreme and see if you get performance regression during the test.

 

[nlel1975]

Be interested to see your Cinebench 2024 results at stock with the 7950X3D, I got 2003 multicore points with a max core temp of 75.7C.

 

[Audioboxer]

Be interested to see your Cinebench 2024 results at stock with the 7950X3D, I got 2003 multicore points with a max core temp of 75.7C.

Stock as in the whole BIOS at stock or just the CPU stock (no PBO) with the above memory settings?

I ran it there with my PBO on

Sometimes with CB23 I'd run it a few times due to variance but this is the first time I have ran 2024 and given how slow it goes, 1 run for now

 

[nlel1975]

Stock as in the whole BIOS at stock or just the CPU stock (no PBO) with the above memory settings?

I meant CPU stock, no PBO and no settings in BIOS which enhances CPU performance.

 

[Audioboxer]

I meant CPU stock, no PBO and no settings in BIOS which enhances CPU performance.

Do you still want my memory as is? So 8000 and the system is in 2:1 mode? I assume memory/latency will still play a role with CB2024 scores.

 

[nlel1975]

Do you still want my memory as is? So 8000 and the system is in 2:1 mode? I assume memory/latency will still play a role with CB2024 scores.

Prefer memory at 6000 like mine if possible, thanks.

 

[Audioboxer]

Prefer memory at 6000 like mine if possible, thanks.

This is without PBO and CPU at complete stock (nothing turned on that influences it)

I don't have a 6000 memory profile at the moment, I'll get back to you on that.

FCLK is at 2200 as well which I believe helps things be a bit faster overall.

 

[nlel1975]

This is without PBO and CPU at complete stock (nothing turned on that influences it)

Thanks that’s a bit higher than what the reviewers get. I choose Cinebench 2024 as it uses AVX2 and therefore requires a AVX2 compatible CPU. I don’t think Cinebench r23 does.

 

[Audioboxer]

Thanks that’s a bit higher than what the reviewers get. I choose Cinebench 2024 as it uses AVX2 and therefore requires a AVX2 compatible CPU. I don’t think Cinebench r23 does.

That's really interesting to know given all the issues that can arise with AVX2 and these chips. So while CB is not a stability test it's an added benefit to be running it if it uses AVX2 now.

Any additional score for me is likely down to cooling and/or FCLK at 2200 and RAM at 8000 with tight settings. Reviewers will likely be air-cooling, FCLK 2000 and running ram at XMP. XMP settings for RAM use jdec specs for pretty much all secondary timings which leaves lots of potential performance on the table. Tuning RAM is the dark arts though and can take a lot of time so many just don't see the small performance gain as worthy of the pain.

I'll drop down to 6000/2000 later today with a basic profile and test again.

 

[McCarthy]

Yeah, this one I have now passes AVX2 at complete stock, so it just proves what I already knew that some chips can do it.

Between AVX2 and y-cruncher 00-x86 FFT only test this is my current PBO values

The 00-x86 FTT in Corecycler with y-cruncher absolutely decimates the 3D cache chips. As in, while AVX2 is a great test to run for finding your curve values, 00-x86 FFT is 100% going to bring down the cache only cores curve further. I recommend everyone tries it on their cache cores. Core 0 for example came all the way down from like -33 with AVX2 for me

Pretty much got my memory tuned in as well now

So I finally got there even though my 'welcome to AM5' was quite a rough ride!

The last CPU was a bit better with core frequency boosting, but it was failing AVX2 at stock and was not FCLK 2200 stable in the slightest. One of the best ways to check your FCLK other than looking out for USB issues is to run LinpackXtreme and see if you get performance regression during the test.

I had a very similar result when learning how to tune this CPU. Corecycler in YCruncher and Prime is the best way to get stability. At first I thought something was wrong because previous testing I could have certain cores at -20 say, then YCruncher In Corecycler took the same core to -2. Hours I wasted with other tests only to find I could have saved time just using these tests. Even in the dreaded SHA3 I had to use the programme WhoCrashed to find the exact core that was causing the problem.

In terms of memory @LtMatt helped a great deal, because this is not something I have spent much time on I’ll be honest.

Like I say this system has been rock solid in everything, and I’ve had 0 errors or crashes.

Sometimes a game that has just been released might not have updated in gamebar but it usually happens fairly quick.

Glad you finally got this sorted. Nothing worse than the pulling your hair out stage haha.

 

[Dreadi]

Anybody give me an idea of safe voltages to be pushing through my ram...?

 

[ph123uk]

I know I shouldn't have, given I play at 4k 120hz on a 65" LG OLED, but i've been looking at updating from my Watercooled 5800x / 7900XTX system, this weekend I've managed to bag:

Ryzen 7950X3D (2nd hand)
32GB of XPG Lancer RGB 6400mhz CL32 (New / Opened)
Asus ROG STRIX X670 E-A (2nd Hand)

After just reading the last few pages, I'm a bit worried now, the Board is here (and immaculate) but the CPU and RAM are on their way - Must admit, i normally just buy new, to save this hassle.

I suppose i'm asking, whats the surefire way of seeing if any of these bits are faulty, the CPU im worried about as I won the bid, and it was VERY low for that chip.

 

[McCarthy]

I know I shouldn't have, given I play at 4k 120hz on a 65" LG OLED, but i've been looking at updating from my Watercooled 5800x / 7900XTX system, this weekend I've managed to bag:

Ryzen 7950X3D (2nd hand)
32GB of XPG Lancer RGB 6400mhz CL32 (New / Opened)
Asus ROG STRIX X670 E-A (2nd Hand)

After just reading the last few pages, I'm a bit worried now, the Board is here (and immaculate) but the CPU and RAM are on their way - Must admit, i normally just buy new, to save this hassle.

I suppose i'm asking, whats the surefire way of seeing if any of these bits are faulty, the CPU im worried about as I won the bid, and it was VERY low for that chip.

I suppose testing the 7950x3d at stock would be the first thing to do. Just make sure you install the chipset drivers from AMD for your board, game bar is on etc and run some tests. If it fails more than one at stock it would indicate an issue.

Don’t worry too much before getting it though. Everything could be fine, and you’ve just grabbed yourself a bargain.

Good luck!

 

[ph123uk]

I suppose testing the 7950x3d at stock would be the first thing to do. Just make sure you install the chipset drivers from AMD for your board, game bar is on etc and run some tests. If it fails more than one at stock it would indicate an issue.

Don’t worry too much before getting it though. Everything could be fine, and you’ve just grabbed yourself a bargain.

Good luck!

Yep, I thought to myself "Why am I thinking glass half full here before its even arrived" So, everything is here except the CPU - We shall see!

 

[nlel1975]

FWIW mine does it too in OCCT doesn’t bother me as my machine appears to be stable otherwise.

Well I've got to the bottom of it guys. I did a fresh install of Windows 11 and installed all motherboard drivers, chipset etc. Now my machine passes the OCCT Extreme Variable Large Test on Corecycler. Left it running for the full hour and it passed with no errors.

 

[Audioboxer]

Think this is me close to done at 8000.

PBO curve is safe with AVX2 and 00-x86 FFT testing overnight, both of these are the curve killers with Ryzen 7000 it seems. IMO best things to run, configure Corecycler Prime95 section for AVX2 and Ycruncher section for 00-x86 and only the FFT test. Just be prepared for lots of reboots on your cache CCD with 00-x86 That setup seems to be the best way to narrow in on your CCD0 curve.

For example, I had Core 0 up at nearly -30 IIRC with AVX2 then 00-x86 knocked it way down. AVX2 seems to be the best test for the frequency core.

The joys of these 7950x3Ds, they are a lot harder to tune and from my posts earlier in this topic, IMO, even get stable from the factory (I will persist in saying every chip put out by AMD should pass things like AVX2 testing at BIOS stock).

 

[nlel1975]

The joys of these 7950x3Ds, they are a lot harder to tune and from my posts earlier in this topic, IMO, even get stable from the factory (I will persist in saying every chip put out by AMD should pass things like AVX2 testing at BIOS stock).

They all should and mine does at stock after a fresh Windows 11 install. It didn't on my 2.5 year old install which started out life on a different motherboard and CPU. The only reason why I did a fresh install was because Geekbench V6 was showing a much lower score than normal 13000 as opposed to 20000 which I now get.

I only tested OCCT again for the hell of it and it passed. In future when changing motherboards I will do a fresh install, it only took me about 4 hours to reinstall everything which is not to bad if only doing every few years.

Never realised that transferring an old Windows install to new hardware would cause this issues. But the advice has always been, do a fresh install after changing hardware.