Low Time Spy Scores - RTX 4090 STRIX OC - 30K Graphics

I am super confused now.
I was referring to 3dmark with single dimm on 7950x3d

Did it score 38K graphics score?

My hunch is that you have 2 bad parts based on what you have said.

Asus motherboard
7950x3d

I would RMA both.

Here is the order of how I would’ve tested.

7950x3d, Asus board, known good stick of mem in known good slot
7950x3d, Asus board, known good stick of mem in known bad slot
7800x3d, Asus board, known good stick of mem in known good slot
7800x3d, Asus board, known good stick of mem in known bad slot

The ones that produce 30K should be the culprits.

If 2 and 4 produce 30k it is board
If 1, 2 and 4 produce 30K it is both board and 7950x3d processor
 
Last edited:
I am super confused now.
I was referring to 3dmark with single dimm on 7950x3d

Did it score 38K graphics score?

My hunch is that you have 2 bad parts based on what you have said.

Asus motherboard
7950x3d

I would RMA both.

Here is the order of how I would’ve tested.

7950x3d, Asus board, known good stick of mem in known good slot
7950x3d, Asus board, known good stick of mem in known bad slot
7800x3d, Asus board, known good stick of mem in known good slot
7800x3d, Asus board, known good stick of mem in known bad slot

The ones that produce 30K should be the culprits.

If 2 and 4 produce 30k it is board
If 1, 2 and 4 produce 30K it is both board and 7950x3d processor
Thank you.

7950X3D - all configs - 30K
7800X3D - 38K

Mem slot closest to CPU is confirmed bad.
 
Damn, tough luck. Well RMA both then and see what you want to do later once RMA parts arrive. At least there is no downtime since you have a new proc and board lying around.
 
Damn, tough luck. Well RMA both then and see what you want to do later once RMA parts arrive. At least there is no downtime since you have a new proc and board lying around.
Yeah, after reading up on more Cinebench benchmarks it looks like I have two things slowing down my number:

- 280MM AIO cooler (most with higher numbers use 360MM, 420MM or custom wc)
- Single DIMM (DDR5 6000 MHz) versus two

So I think the 7950X3D is fine. Still baffled by 3DMark and the 7950X3D but everything else looks fine. Working on an RMA with ASUS for the X670E GENE now.

EDIT: ASUS approved RMA. I guess the "nightmare" part would be when they receive it...hopefully that goes smooth!
 
Last edited:
- 280MM AIO cooler (most with higher numbers use 360MM, 420MM or custom wc)

That's kind of weird. The X3D chips are pretty low wattage. 7800X3D in particular can run off of an air tower cooler trivially. I believe the 7950X3D is a bit higher, but I would think a 280MM would absolutely be plenty. Make sure your 280mm is actually working properly. I think if it hits 90C, it is throttling. The throttle point for X3D chips is 89C.

For what it's worth, I have a Kraken X73 (or similar) which is I believe a 360mm AIO. I put it on my 7800X3D and the temperatures were worse than they were with a Noctua single tower air cooler, and a Thermalright Phantom Spirit. The latter is a $34 or so cooler. So I think that's something else you can try. I think you might be throttling on the 7950X3D. Your AIO might be bad with its contact surface or something similar.
 
That's kind of weird. The X3D chips are pretty low wattage. 7800X3D in particular can run off of an air tower cooler trivially. I believe the 7950X3D is a bit higher, but I would think a 280MM would absolutely be plenty. Make sure your 280mm is actually working properly. I think if it hits 90C, it is throttling. The throttle point for X3D chips is 89C.

For what it's worth, I have a Kraken X73 (or similar) which is I believe a 360mm AIO. I put it on my 7800X3D and the temperatures were worse than they were with a Noctua single tower air cooler, and a Thermalright Phantom Spirit. The latter is a $34 or so cooler. So I think that's something else you can try. I think you might be throttling on the 7950X3D. Your AIO might be bad with its contact surface or something similar.
Yeah in HWiNFO it says "No" in the throttling section. If the AIO was toast it would definitely be a different story. The 7800X3D is a far cooler, chip - obviously. So it may just be better suited for my system since a 280MM AIO is all that fits. We're talking Cinebench R23 for 10 minutes so it is not exactly just playing a game - it's far more intensive.

I think in the end I am just going to keep my 7950X3D and wait for the board to be returned from RMA. Get everything back together and assess from there. I've got my gaming laptop and stuff that I can sub in - versus using all of these parts I just bought temporarily.
 
Yeah, after reading up on more Cinebench benchmarks it looks like I have two things slowing down my number:

- 280MM AIO cooler (most with higher numbers use 360MM, 420MM or custom wc)
- Single DIMM (DDR5 6000 MHz) versus two

So I think the 7950X3D is fine. Still baffled by 3DMark and the 7950X3D but everything else looks fine. Working on an RMA with ASUS for the X670E GENE now.

EDIT: ASUS approved RMA. I guess the "nightmare" part would be when they receive it...hopefully that goes smooth!
It's good the approval went quickly. The nightmare would be them making up some crap about you damaging the board and the replacement cost is more than a new one would cost. I think you will be okay though in this case.
 
It's good the approval went quickly. The nightmare would be them making up some crap about you damaging the board and the replacement cost is more than a new one would cost. I think you will be okay though in this case.
Don't count those chickens. They haven't received the board yet.
 
Yeah in HWiNFO it says "No" in the throttling section. If the AIO was toast it would definitely be a different story. The 7800X3D is a far cooler, chip - obviously. So it may just be better suited for my system since a 280MM AIO is all that fits. We're talking Cinebench R23 for 10 minutes so it is not exactly just playing a game - it's far more intensive.

I think in the end I am just going to keep my 7950X3D and wait for the board to be returned from RMA. Get everything back together and assess from there. I've got my gaming laptop and stuff that I can sub in - versus using all of these parts I just bought temporarily.

The 7800X3D hits about 89-90W during a stress test like that. I believe the 7950X3D is rated at 120W. It's a difference for sure, but not quite as much as you might think (33% extra watts, or 30W total). I didn't mean the AIO was totally toast. My Kraken still technically operates. But I think there might be something defective about its contact surface, it's not transmitting heat nearly as well as a much cheaper tower cooler. I'm not sure how much trust I would put into HWInfo's throttling indicator. Maybe try watching clocks manually? If it's throttling, that would explain an awful lot. Though, you were having some issues in gaming as well, before.

Could obviously also be that ASUS board, though.
 
Hopefully ASUS takes care of things quickly. My experiences with parts vendors range from good (Corsair, EVGA, MSI) to downright horrendous (Gigabyte). Never dealt with ASUS before, but they have a large enough US presence that I bet they're at least passable.

Never bothered with Cinebench before, but while running all of my office programs (Teams, Chrome with 6-7 tabs, Adobe CC, etc.) my 7950X3D pulled 35K. I have a feeling that I'd be looking at the "normal" 38K if I ran things clean.
 
Hopefully ASUS takes care of things quickly. My experiences with parts vendors range from good (Corsair, EVGA, MSI) to downright horrendous (Gigabyte). Never dealt with ASUS before, but they have a large enough US presence that I bet they're at least passable.

Never bothered with Cinebench before, but while running all of my office programs (Teams, Chrome with 6-7 tabs, Adobe CC, etc.) my 7950X3D pulled 35K. I have a feeling that I'd be looking at the "normal" 38K if I ran things clean.

Thanks! I think 35K versus my 33K is fair - since I was rocking only 1 memory DIMM. Makes me feel better the CPU is not the issue.

The 7800X3D hits about 89-90W during a stress test like that. I believe the 7950X3D is rated at 120W. It's a difference for sure, but not quite as much as you might think (33% extra watts, or 30W total). I didn't mean the AIO was totally toast. My Kraken still technically operates. But I think there might be something defective about its contact surface, it's not transmitting heat nearly as well as a much cheaper tower cooler. I'm not sure how much trust I would put into HWInfo's throttling indicator. Maybe try watching clocks manually? If it's throttling, that would explain an awful lot. Though, you were having some issues in gaming as well, before.

Could obviously also be that ASUS board, though.

You can see the speeds here in my post: https://hardforum.com/threads/low-t...c-30k-graphics.2032750/page-2#post-1045818501.

Fresh Kryonaut, NZXT Kraken X63 280MM AIO...
 
You can see the speeds here in my post: https://hardforum.com/threads/low-t...c-30k-graphics.2032750/page-2#post-1045818501.

Fresh Kryonaut, NZXT Kraken X63 280MM AIO...

Is that while it's running? Unless I'm missing something, this just shows "Current, Maximum, Minimum, and Average", after you already had your score. That's not showing what it reaches steady state while running, and what your frequency fluctuations look like. For the record, my X73 had the same issue. My CPU would go very high, very quickly. With the current Thermalright Phantom Spirit, it goes up to a point during Cinebench quickly, and then rises a bit more slowly to reach a steady state (around 78C or so). With the Kraken, it would basically smash up to 88C almost instantly and barely try to remain cool.

I'm not sure what's wrong with my Kraken, but I do notice a dent in its surface, towards the center. It's a miniscule dent but either way I don't think it's making good contact, even with the thermal paste (which is some Grizzly variant as well). Again, just something to keep in mind. I've read of people having issues with just how much they tightened the screws on AM5, it's an odd platform. Since all X3D chips are very low wattage, if you have a Phantom Spirit lying around (again, it's like $34-37 cooler), I would give it a whirl. It might do better, it might not. I think there was a topic where they were discussing offsets on the cooler due to the hotspot on the 7950X3D being odd?
 
Is that while it's running? Unless I'm missing something, this just shows "Current, Maximum, Minimum, and Average", after you already had your score. That's not showing what it reaches steady state while running, and what your frequency fluctuations look like. For the record, my X73 had the same issue. My CPU would go very high, very quickly. With the current Thermalright Phantom Spirit, it goes up to a point during Cinebench quickly, and then rises a bit more slowly to reach a steady state (around 78C or so). With the Kraken, it would basically smash up to 88C almost instantly and barely try to remain cool.

I'm not sure what's wrong with my Kraken, but I do notice a dent in its surface, towards the center. It's a miniscule dent but either way I don't think it's making good contact, even with the thermal paste (which is some Grizzly variant as well). Again, just something to keep in mind. I've read of people having issues with just how much they tightened the screws on AM5, it's an odd platform. Since all X3D chips are very low wattage, if you have a Phantom Spirit lying around (again, it's like $34-37 cooler), I would give it a whirl. It might do better, it might not. I think there was a topic where they were discussing offsets on the cooler due to the hotspot on the 7950X3D being odd?
That’s after the run. I’ll check it out - thank you! I have plenty of time as I wait for RMA.

My preferred cooler was an Arctic Liquid Freezer II 280 but it didn't fit by a MM or something. :( So I bought that X63 when I built the rig back in October 2022.

Your post (and the fact that I have swapped CPUs about 10 times during the AM5 generation) makes me think I need to tighten less. I get big pump out even when it has only been a day or two. Actual thermal paste on the CPU itself is razer thin upon removal.
 
Last edited:
Yeah, after reading up on more Cinebench benchmarks it looks like I have two things slowing down my number:

- 280MM AIO cooler (most with higher numbers use 360MM, 420MM or custom wc)
- Single DIMM (DDR5 6000 MHz) versus two

So I think the 7950X3D is fine. Still baffled by 3DMark and the 7950X3D but everything else looks fine. Working on an RMA with ASUS for the X670E GENE now.

EDIT: ASUS approved RMA. I guess the "nightmare" part would be when they receive it...hopefully that goes smooth!
When you run the 3dMark bench on the 7950X3D, It will show a graph with all of the measured results graphed out for those users with the same CPU and GPU, The graph will show an "Average" line, and a line for your results. Where does it fall in that graph, very bottom?

Suspect that the results you are getting, might just be typical of a 7950X3D. And you can check if that is the case, if the above graph shows your results close to the average of other users results. If your results are markedly lower, could be a bad 7950X3D. Could be the mobo too. RMA'ing it isn't a bad idea. And you might have had 2 bad parts at the same time, sometimes happens.
30K results could be: Typical of a 7950X3D, and if not the 7950X3D might be defective.
The Ram issue is probably the new 7800X3D, also a chance it is the mobo.

I guess now at least you have sufficient spare parts for any future problem diagnosis... :)
 
Just got an email from ASUS saying my replacement mobo has shipped! Nice...
 
  • Like
Reactions: Void
like this
New mobo solved the issue:

https://www.3dmark.com/spy/45407245

I went with an X670E ASRock Steel Legend a forum member was selling used - had to reconfigure my case but with the X670E GENE in high demand I'll just sell it off. Thanks for all of the help everyone. It's a relief to finally have normal performance.

View attachment 632862

(STRIX OC is for sale - sticking with the FE for the size/price)
 
Last edited:
New mobo solved the issue:

https://www.3dmark.com/spy/45407245

I went with an X670E ASRock Steel Legend a forum member was selling used - had to reconfigure my case but with the X670E GENE in high demand I'll just sell it off. Thanks for all of the help everyone. It's a relief to finally have normal performance.

View attachment 632862

(STRIX OC is for sale - sticking with the FE for the size/price)
Glad to hear it. Enjoy!!
 
Was just skimming through this thread and noticed what might have been the issue with the ASUS mainboard. Shouldn't those be A2 and B2.
ASUS board.png
 
Lots of good info in this thread. After reading through it, I ran Time Spy on my system (5900X and 4090) and discovered my graphics score was also a bit low.

Here's my starting point:

Timespy 2 displays 2024-02-07 133125.png


After some tweaking, I ended up here:

Timespy overclocked 2024-02-07 221817.png


I did the following:
  • disabled virtualization
  • enabled resizable bar
  • set PBO limits to motherboard defined
  • set auto-overclock to +200
  • set all-core curve at -15
  • bumped RAM to 1.5 volts
  • reduced timings from 16-16-16-36 to 14-15-15-34 (1T) (3600Mhz)
    • tried relaxed timings and faster speeds, but Time Spy liked the lower timings best

I'm pretty happy with this result. I'm still not seeing 100% GPU utilization during the GPU specific tests, so I expect the 5900X is still holding the 4090 back a tiny bit, but it's good enough I'm content with the 5900X until Zen 5.
 
Last edited:
Lots of good info in this thread. After reading through it, I ran Time Spy on my system (5900X and 4090) and discovered my graphics score was also a bit low.

Here's my starting point:

View attachment 633552

After some tweaking, I ended up here:

View attachment 633553

I did the following:
  • disabled virtualization
  • enabled resizable bar
  • set PBO limits to motherboard defined
  • set auto-overclock to +200
  • set all-core curve at -15
  • bumped RAM to 1.5 volts
  • reduced timings from 16-16-16-36 to 14-15-15-34 (1T) (3600Mhz)
    • tried relaxed timings and faster speeds, but Time Spy liked the lower timings best

I'm pretty happy with this result. I'm still not seeing 100% GPU utilization during the GPU specific tests, so I expect the 5900X is still holding the 4090 back a tiny bit, but it's good enough I'm content with the 5900X until Zen 5.
All-Core OC, Turn SMT off, you can max out a 4090 on a 5xxx series this way, and also remove stuttering in games. If you need the link to where i talked about this elsewhere with graphs, let me know! ;)
 
All-Core OC, Turn SMT off, you can max out a 4090 on a 5xxx series this way, and also remove stuttering in games. If you need the link to where i talked about this elsewhere with graphs, let me know! ;)

Very interesting results with SMT off. CPU score takes a significant hit, but GPU score gets a boost. Most exciting, my GPU utilization is hitting 100% so I think you're on to something with the stuttering. Played a couple hours of Cyberpunk after turning SMT off last night and it was nice and smooth. Would have to switch back and forth to say for sure if I notice any hitching with SMT on.

SMT on: 28,829 overall score; 35,530 graphics score; 13,939 CPU score.

SMT off:

Timespy SMT Off 2024-02-08 215154.png
 
Very interesting results with SMT off. CPU score takes a significant hit, but GPU score gets a boost. Most exciting, my GPU utilization is hitting 100% so I think you're on to something with the stuttering. Played a couple hours of Cyberpunk after turning SMT off last night and it was nice and smooth. Would have to switch back and forth to say for sure if I notice any hitching with SMT on.

SMT on: 28,829 overall score; 35,530 graphics score; 13,939 CPU score.

SMT off:

View attachment 633795
FWIW, you can overcome the slight dip in CPU score pretty easy if you overclock all cores. Also, make sure CPPC is OFF. What I proved, and have used for well over a year now is the absolute best way to run a Ryzen 5xxx CPU (for gaming anyway).
 
FWIW, you can overcome the slight dip in CPU score pretty easy if you overclock all cores. Also, make sure CPPC is OFF. What I proved, and have used for well over a year now is the absolute best way to run a Ryzen 5xxx CPU (for gaming anyway).
By overclock all cores do you mean set an all-core negative offset? I have a -15 offset on all cores, and +200 set on the boost frequency.
 
By overclock all cores do you mean set an all-core negative offset? I have a -15 offset on all cores, and +200 set on the boost frequency.
No, I mean turn off that PBO garbage and legit OC via voltage and multiplier the "old school" way. For example; 45x multiplier with 100 BCLK at say 1.25Vcore or something. Your milage my vary obviously, 1.35Vcore is the "accepted" 24/7 max voltage on the 5xxx, but you could probably go higher. I simply set mine to 1.35v and saw how far I could go temps and stability wise and called it a day there.
 
No, I mean turn off that PBO garbage and legit OC via voltage and multiplier the "old school" way. For example; 45x multiplier with 100 BCLK at say 1.25Vcore or something. Your milage my vary obviously, 1.35Vcore is the "accepted" 24/7 max voltage on the 5xxx, but you could probably go higher. I simply set mine to 1.35v and saw how far I could go temps and stability wise and called it a day there.

Sweet, looks easy enough. I'll give it a shot and report back.
 
FWIW, you can overcome the slight dip in CPU score pretty easy if you overclock all cores. Also, make sure CPPC is OFF. What I proved, and have used for well over a year now is the absolute best way to run a Ryzen 5xxx CPU (for gaming anyway).
Turned off CPPC, set VCORE to 1.35V. Highest I could get the multiplier was 42.75, which produced nearly identical results as with PBO on:

1707537317583.png


With a better binned CPU it may be worthwhile to do the all core OC, but in my case I think I'll just switch back to PBO and forget it. I'm not too worried about the dip in CPU score, I think getting GPU utilization to 100% is the main thing to resolve any possible stuttering issues, and turning off SMT seems to be the solution there.
 

Attachments

  • 1707535611273.png
    1707535611273.png
    730.3 KB · Views: 0
Turned off CPPC, set VCORE to 1.35V. Highest I could get the multiplier was 42.75, which produced nearly identical results as with PBO on:

View attachment 634005

With a better binned CPU it may be worthwhile to do the all core OC, but in my case I think I'll just switch back to PBO and forget it. I'm not too worried about the dip in CPU score, I think getting GPU utilization to 100% is the main thing to resolve any possible stuttering issues, and turning off SMT seems to be the solution there.
For sure... OC'ing is always luck of the Silicon Gods as we all well know over the last 30+ years... lol. :) If you can keep your GPU at 100% with much less stutter, you are doing better already with your configuration. I lucked out with a 4.8Ghz OC w/ 1.35V on a 5950X I think... lol. However, keep in mind, an all-core OC can reduce stutter more, so if you have test which measures stutter, it is worth running at least to see even if that minor OC you got is better or worse than PBO. ;)
 
  • Like
Reactions: Epos7
like this
What program are people using to measure CPU stability these days? Cinebench, AIDA64, Prime95? Ditto with how long or how many runs before you're happy?
 
SMT = AMD Hyperthreading right? Turning that off on Intel can also help.

Your results are pretty good either way:
SMT On
1707695450759.png

SMT Off
1707695494855.png

SMT Off was a 2% increase in overall score, 8.3% cpu score increase, 6.4% GPU score increase.

No idea how 3dmark is calculating the overall score. Maybe the number of threads dropping affects overall score.

I like the above graph the best in the 3DMark results. I wish they would show those for the GPU and CPU as well.

The average dropped by 1 point on your second run, you are affecting the bell curve :)
 
sk3tch Hey brotha idk if you ever solved this issue completely, but I was having the same one. Running Timespy with a 7950X3D and an ASUS ROG STRIX 4090 OC edition base clock 2610 but OCed it to just under 3GHz. Graphics scores never got above 31k. I found another forum post that completely solved my issue and thought I might let you know about it

In Windows if you go to the search bar and type in core isolation, open it up, and turn off memory integrity. After doing that, I ran Timespy again and my graphics score was around 35.5k. Right where it should be for stock. Try that out if you’ve still got the hardware for it and lmk what you get
 
Back
Top