[dono89]

NVIDIA released 572.75 hotfix driver, which finally addressed previously documented bug with voltage control API. NvAPI_GPU_ClientVoltRailsSetControl is now working as intended on RTX 5000 series graphics cards. Now you can safely unlock and use voltage control slider in MSI Afterburner or any other previously affected third party tools (including NVIDIA's own NVIDIA App) without risking to see unexpected performance degradation due to erroneously limiting maximum core clock frequency.


It seems nvidia fixed the issue with voltage slider. Still not working with higher voltage tho like 1,1V when the card at default boosts at 1V

 

[MrKenzie]

If his gaming trio is non oc version then it make sense as 2617 + 500 = 3117 is not far off from my own Astral (After days of gaming and reducing the clock by 7.5MHz every time it crashes it seems it settled at 3105).

If his gaming trio is oc verison on the other hand… 2700 + 500 = 3200 which make it great sample.

My TUF OC model will run games with +525 with the voltage slider at 100. It will only manage +450 game stable without the added voltage. When I ran Port Royal, there was no performance gain from +525 compared to +475. I'm trying not to run many benchmarks, I've only just removed my chiller from my system and I don't want to get caught up chasing numbers again 🙃.

 

[luthienxd]

That is a high overclock. You should test for stability now, and then decide how much power and heat you are comfortable with. The 5080 is great in that the performance scales very efficiently when overclocking however that extra performance is often not needed so generally better to run it lower in most games. A lot of people seem to run about 2.8 GHz at 900mV or 3 GHz at 0.95mV which is a nice balance of performance, power usage, and heat.

everyone has found stability with +2000 and +450 +500, I think it can go higher but there is no need for now

 

[sideeffect]

everyone has found stability with +2000 and +450 +500, I think it can go higher but there is no need for now

I don't think they have. Like was mentioned above it depends on the base clocks so +450 or +500 mean different things to different people. Most people seem to get somewhere between 3-3.2GHz with few cards going as high as 3.3 GHz, it is somewhat unusual to get 3.3 GHz stable with the available voltage with stock cooling. +500 on my card takes it to 3265MHz and it is not stable at that other than for short benchmarks.

 

[dono89]

I don't think they have. Like was mentioned above it depends on the base clocks so +450 or +500 mean different things to different people. Most people seem to get somewhere between 3-3.2GHz with few cards going as high as 3.3 GHz, it is somewhat unusual to get 3.3 GHz stable with the available voltage with stock cooling. +500 on my card takes it to 3265MHz and it is not stable at that other than for short benchmarks.

The issue is that the default voltage varies from card to card. For example, with mine, the default is 0.98V (1V with the voltage slider set to 100%). Running at 3.2GHz with 1V is one thing, but running the same frequency with 1.08V is another. The problem right now is that cards with lower default voltages are limited in their overclocking potential. Capping the frequency at around 3.2 in the majority of the cases although a few cards could potentially reach an additional 200MHz.

 

[Anaconda2430]

How did you figure out a way to overclock your vram so high? I'm using MSI Afterburner and it stops at 2Ghz. Igor's lab suggested to use evga precision tool at the same time so we can go beyond 2Ghz while still using Afterburner's GPU curve. I tried and it's not working well. When I enter +450 (similar to Afterburner curve) in EVGA the card runs at 2.3Ghz. I have Afterburner latest beta version. I tried with Afterburner 4.6.5 and it won't open with my 5080 but it worked with my 4070ti super. I'm pushing my Gigabyte Gaming OC to the edge of stability and can't get passed 9300 on Steel Nomad @ 3.3Ghz GPU/2Ghz VRAM.

 

[dono89]

How did you figure out a way to overclock your vram so high? I'm using MSI Afterburner and it stops at 2Ghz. Igor's lab suggested to use evga precision tool at the same time so we can go beyond 2Ghz while still using Afterburner's GPU curve. I tried and it's not working well. When I enter +450 (similar to Afterburner curve) in EVGA the card runs at 2.3Ghz. I have Afterburner latest beta version. I tried with Afterburner 4.6.5 and it won't open with my 5080 but it worked with my 4070ti super. I'm pushing my Gigabyte Gaming OC to the edge of stability and can't get passed 9300 on Steel Nomad @ 3.3Ghz GPU/2Ghz VRAM.

end of the thread

MSI Afterburner 4.6.6 Beta 5 for NVIDIA GeForce RTX 5000 series cards

MSI Afterburner beta for NVIDIA GeForce RTX 5000 series cards will be published in the next few days. It will be possible to download it here, at Guru3D, and on MSI.com (a few days later after CNY 2025 holidays). It contains the following changes (NVIDIA RTX 5000 specific items are highlighted...

forums.guru3d.com

 

[Anaconda2430]

No offense, mate, but that result you've got is pretty low considering you're pulling over 460W and hitting a max clock of 3405MHz.

With my 5080 Suprim @1.000V, I manage to get over 96 FPS with pretty low power consumption, i have to try with memos @18Ghz

I scored 9 654 in Steel Nomad View attachment 2695062

How did you figure out a way to overclock your vram so high? I'm using MSI Afterburner and it stops at 2Ghz. Igor's lab suggested to use evga precision tool at the same time so we can go beyond 2Ghz while still using Afterburner's GPU curve. I tried and it's not working well. When I enter +450 (similar to Afterburner curve) in EVGA the card runs at 2.3Ghz. I have Afterburner latest beta version. I tried with Afterburner 4.6.5 and it won't open with my 5080 but it worked with my 4070ti super. I'm pushing my Gigabyte Gaming OC to the edge of stability and can't get passed 9300 on Steel Nomad @ 3.3Ghz GPU/2Ghz VRAM.

 

[Mystic1978]

How did you figure out a way to overclock your vram so high? I'm using MSI Afterburner and it stops at 2Ghz. Igor's lab suggested to use evga precision tool at the same time so we can go beyond 2Ghz while still using Afterburner's GPU curve. I tried and it's not working well. When I enter +450 (similar to Afterburner curve) in EVGA the card runs at 2.3Ghz. I have Afterburner latest beta version. I tried with Afterburner 4.6.5 and it won't open with my 5080 but it worked with my 4070ti super. I'm pushing my Gigabyte Gaming OC to the edge of stability and can't get passed 9300 on Steel Nomad @ 3.3Ghz GPU/2Ghz VRAM.

MSI Afterburner 4.6.6 Beta 5 for NVIDIA GeForce RTX 5000 series cards | guru3D Forums


This score was @ 2250 MHz 12 days ago ( +3000 in msi ab ) I scored 26 049 in Port Royal

 

[Panchovix]

View attachment 2701246

MSI Afterburner 4.6.6 Beta 5 for NVIDIA GeForce RTX 5000 series cards | guru3D Forums


This score was @ 2250 MHz 12 days ago ( +3000 in msi ab ) I scored 26 049 in Port Royal

View attachment 2701249

Not original comment, but man I wish we on the 5090 could do +4000Mhz on VRAM or had faster default GDDR7 VRAM.

+3000Mhz does just 2175Mhz (14000+3000Mhz), I think the 5080 starts at 15000.

 

[MQB888R]

View attachment 2701246

MSI Afterburner 4.6.6 Beta 5 for NVIDIA GeForce RTX 5000 series cards | guru3D Forums


This score was @ 2250 MHz 12 days ago ( +3000 in msi ab ) I scored 26 049 in Port Royal

View attachment 2701249

Thank you. I appreciate your generosity.

Using 572.75 with my reference PNY 5080 OC, vanilla bios and the modified .dat file, I was able to hit 9200 in Steel Nomad with +440/+3000, max v and power limit, achieving clocks >3200 and memory rock solid at 18001/2250 MHz.

Spoiler: Steel Nomad, 9204

I started the day at ~8800 and had a goal to break 9000; this is very cool. I have the Alphacool CORE block on the way, stoked to push this further with refined drivers, mods and bios flashes.

 

[Anaconda2430]

View attachment 2701246

MSI Afterburner 4.6.6 Beta 5 for NVIDIA GeForce RTX 5000 series cards | guru3D Forums


This score was @ 2250 MHz 12 days ago ( +3000 in msi ab ) I scored 26 049 in Port Royal

View attachment 2701249

Amazing, thanks for providing this so quickly. I've been experimenting with vram at 3ghz but couldn't get passed my previous score of 9227 @ 2Ghz vram in SteelNomad. Seems like I'm hitting the ceiling here. Gpu core at +460Mhz was peaking close to 3.3Ghz. It's probably because I'm on a non X3D cpu (9900x) and my gpu is mounted vertically. There could be some micro-latency in the PCIe data transfer. I'll retry tomorrow with early morning fresh air!

 

[Anaconda2430]

Thank you. I appreciate your generosity.

Using 572.75 with my reference PNY 5080 OC, vanilla bios and the modified .dat file, I was able to hit 9200 in Steel Nomad with +440/+3000, max v and power limit, achieving clocks >3200 and memory rock solid at 18001/2250 MHz.

Spoiler: Steel Nomad, 9204

View attachment 2701254

I started the day at ~8800 and had a goal to break 9000; this is very cool. I have the Alphacool CORE block on the way, stoked to push this further with refined drivers, mods and bios flashes.

Great score! Did you have a warning in SteelNomad that 572.75 driver is not supported? I tried a benchmark with that version and received the warning. Can't post my score on 3dmark website. Downgraded to 572.70 and the warning is gone.

 

[MQB888R]

Great score! Did you have a warning in SteelNomad that 572.75 driver is not supported? I tried a benchmark with that version and received the warning. Can't post my score on 3dmark website. Downgraded to 572.70 and the warning is gone.

Yes, it was not an officially supported score, but I am not chasing the top100 right now, just trying to establish a good baseline

 

[Mystic1978]

Amazing, thanks for providing this so quickly. I've been experimenting with vram at 3ghz but couldn't get passed my previous score of 9227 @ 2Ghz vram in SteelNomad. Seems like I'm hitting the ceiling here. Gpu core at +460Mhz was peaking close to 3.3Ghz. It's probably because I'm on a non X3D cpu (9900x) and my gpu is mounted vertically. There could be some micro-latency in the PCIe data transfer. I'll retry tomorrow with early morning fresh air!

I don't think that 9900X is a limiting factor. I'm running a 14900KF at just 5GHz with DDR4 RAM on PCIe 4.0, with my RAM at 4000MHz CL15 in Gear 1, and I managed to break 26K in Port Royal. In Steel Nomad, I reach 97 FPS without any issues.

I scored 9 709 in Steel Nomad

Your limiting factor might actually be power limits if your stock voltages are high. Have you checked the average clocks you get during the benchmark? Temperatures could also be a factor if they are relatively high.


All my results were achieved using the factory BIOS with a 400W limit, and I still had plenty of power headroom because my voltages were very low, around 0.995 to 1.005V, while passing all synthetic tests. This is an advantage because I don't even need more than 350W, but I have an issue: for now, I can't manually push voltages above 1.020V, which is too low if I aim for 3400MHz. I would need at least 1.050V or more to reach that.


With this latest hotfix driver, at 1.020V fixed, I can get 3307MHz pretty solid. I imagine that at 1.1V, I could reach 3450 or 3500MHz, as long as I keep temps under control.


For now, on GPUs that run at very low voltage, there's no way to push more than 0.020V, which is barely anything. Even with a custom curve, I can't set fixed frequencies and voltages beyond 1.020V in my case. So for now, I'm kind of screwed if I want to aim for the top 1 spot on any 3DMark bench.


My goal was Port Royal since I'm only 600 points away from the #1 spot, and my limiting factor is voltage. I was considering doing a power mod, to unlock the voltage just to push it a bit further. However, I don’t really feel like losing my warranty, and I don’t have the time or motivation to go through the whole process. So for now, I’ll just wait and see if this gets resolved in some other way.


Regarding the benchmark not being validated, it's likely because driver 572.75 is not official it's just a hotfix over the previous 572.70 driver.

 

[Mystic1978]

The other day, I saw a video from a fellow user sugi0lover, who has an RTX 5090, where he tested Doom Eternal at 1080p Low with his 9800X3D delidded and pushed to its maximum OC, along with water-cooled RAM. His 5090 was also overclocked and water-cooled, using extreme radiators in a 10°C ambient temperature to achieve this:

On the other hand, a subscriber asked me to replicate the same benchmark with my more modest setup. As I mentioned before, I'm running a 14900KF at stock ( on this test ) with an AIO, and a 5080 on air cooling using DDR4.


To my surprise, I got very close to his results. Here’s the proof:

In the end, the X3D CPUs aren't that big of a deal 😅

 

[Anaconda2430]

I don't think that 9900X is a limiting factor. I'm running a 14900KF at just 5GHz with DDR4 RAM on PCIe 4.0, with my RAM at 4000MHz CL15 in Gear 1, and I managed to break 26K in Port Royal. In Steel Nomad, I reach 97 FPS without any issues.

I scored 9 709 in Steel Nomad

Your limiting factor might actually be power limits if your stock voltages are high. Have you checked the average clocks you get during the benchmark? Temperatures could also be a factor if they are relatively high.


All my results were achieved using the factory BIOS with a 400W limit, and I still had plenty of power headroom because my voltages were very low, around 0.995 to 1.005V, while passing all synthetic tests. This is an advantage because I don't even need more than 350W, but I have an issue: for now, I can't manually push voltages above 1.020V, which is too low if I aim for 3400MHz. I would need at least 1.050V or more to reach that.


With this latest hotfix driver, at 1.020V fixed, I can get 3307MHz pretty solid. I imagine that at 1.1V, I could reach 3450 or 3500MHz, as long as I keep temps under control.


For now, on GPUs that run at very low voltage, there's no way to push more than 0.020V, which is barely anything. Even with a custom curve, I can't set fixed frequencies and voltages beyond 1.020V in my case. So for now, I'm kind of screwed if I want to aim for the top 1 spot on any 3DMark bench.


My goal was Port Royal since I'm only 600 points away from the #1 spot, and my limiting factor is voltage. I was considering doing a power mod, to unlock the voltage just to push it a bit further. However, I don’t really feel like losing my warranty, and I don’t have the time or motivation to go through the whole process. So for now, I’ll just wait and see if this gets resolved in some other way.


Regarding the benchmark not being validated, it's likely because driver 572.75 is not official it's just a hotfix over the previous 572.70 driver.

Thanks for the elaborated answer! Very useful. Will experiment a little more with the power limits factor, as you suggested. That's what could be the main culprit throttling down my clock. I'm joining the links to my latest SteelNomad scores, at 2Ghz and 3Ghz vram. You're totally right about the X3D/non-X3D. Doesn't seem to be a factor since as you'll see in my score list, the top user achieved 9630 with a 9900x. At this test resolution, cpu cache shouldn't be a bottleneck. And you also mentioned that you're on PCIe 4.0 instead of 5.0. So that should answer my vertical mount factor. GPUz shows that I'm at 5.0. And my temps are in the 50s at full load, never reaching 60. The only thing remaining to investigate in more depth is power limits.

Highest score at vram 2Ghz: I scored 9 227 in Steel Nomad
Second highest score at vram 3Ghz: I scored 9 225 in Steel Nomad

I also came first on 561 with TimeSpy's overall score under 9900x category. My cpu performs really well on cpu gfx calculation test. But my graphic score places me 9th.

Here are some printscreens with more details:

 

[Anaconda2430]

Post deleted

 

[Anaconda2430]

No offense, mate, but that result you've got is pretty low considering you're pulling over 460W and hitting a max clock of 3405MHz.

With my 5080 Suprim @1.000V, I manage to get over 96 FPS with pretty low power consumption, i have to try with memos @18Ghz

I scored 9 654 in Steel Nomad View attachment 2695062

I'm also adding details of my gpu average clock on 2 separate runs of SteelNomad. One remained at 3.221 throughout the whole test while the other one dropped at 3.189 (when vram is at 3ghz). SN is sometimes acting weird. On one run, the clock could remain constant and stable throughout the whole test and on a second run, the test will make the clock fluctuate. I often see my clock peak beyond 3.3Ghz near the end of SN test, making the test crash. I'm still trying to figure out why is the gpu clock stable on one SN run and fluctuating from 3.2 to beyond 3.3 on a subsequent run. That makes testing gpu clock stability really hard to perform when clock fluctuations happen between runs.

I've only got the Gigabyte Gaming OC since March 6 and so far I must say that I'm really impressed with the built quality and oc potential. Temps are remaining low under heavy loads, the card is not too heavy since it has a plastic shroud instead of being all metal like the astral which at 4kg weighs over a kg more than the GB. That's hard on a pcie slot, even with stands or mounted vertically. And it has zero coil whine. This card is silent and it's one of the few that offers 125% power limit. And it has a 4y warranty. I've always been an Asus guy and their products never let me down, even after decades of heavy use. I wanted an astral or a msi suprim but couldn't get one. I just hope that it won't let me down since some users had issues with GB RMA that were not fulfilled as they expected.

 

[SamGreeman13]

I have a question;
So I flashed the GB 450W Waterforce Bios onto my MSI Inspire 3X OC. I have gotten my steel nomad score to 9295. My issue with overclocking is that the hw logs don’t match up. In gpu tweak, I have a VF curve tuned core offset of +364 (which on the waterforce should be 2805 +364 =3,169 which it says in GPU tweak). The curve flattens out at 3441 MHz @ 1090mV. HWinfo reports the max frequency during the benchmark as 3300 MHz and 1070mV. I have found that 1070mV will give me about 3300 MHz, and now that I have the voltage slider unlocked with the new driver, voltage has reached 1095mV while not in the benchmark. I’m trying to create a curve that makes it so that 1070 mv will give 3300 MHz, and the curve will continue to go up until 1100 mv (whatever the clock speed is with that voltage) at which point I will flatten it out, but because of the gap in GPU tweak’s number and the actual clock speed recorded, I have no clue how to match up the numbers. Any help?