Post your Results for the New Solar Bay Benchmark from 3DMark

[J7SC_Orion]

@Bastiaan_NL @Avacado @Mr. Fox @Sir Beregond ...interesting new 3D Mark bench which challenges both CPU and GPU. Also, lots of ray tracing. Over 600 W GPU usage for my run which (for now) is at #1 at HOF. I am still learning about this bench and others will do the same so I don't expect to stay there for long, but my 5950X result could hold on a bit longer in class...

  

 

  

[Mr. Fox]

Thanks for posting this. I was not even aware it existed, and I just posted some of my best Time Spy results in the past week or so. I haven't seen any mention of it anywhere until now. 

 

FYI, it doesn't like RTSS. If I have RTSS/OSD running the benchmark fails at launch with the meaningless  message "an error has occurred" (typical of 3DMark errors that tell you nothing of value). If anyone else has the same issue, shut down RTSS and it should complete.

 

Edited August 15, 2023 by Mr. Fox

[UltraMega]

Context: 

 

UL Solutions Launches 3DMark Solar Bay, New Cross-Platform Ray Tracing Benchmark (guru3d.com) 

[Mr. Fox]

It seem very unpredictable and unreliable. Very wild extremes in scoring between runs. In less than a dozen runs on the same system I am seeing a spread of more than 10,000 3DMarks. Some people are posting illogically high benchmark scores without GPU overclocking and I think there are a lot of bugged submissions. 

 

Example: all over the board, erratic results (same computer)

Edited August 15, 2023 by Mr. Fox

[Mr. Fox]

1 hour ago, Mr. Fox said:

It seem very unpredictable and unreliable. Very wild extremes in scoring between runs. In less than a dozen runs on the same system I am seeing a spread of more than 10,000 3DMarks. Some people are posting illogically high benchmark scores without GPU overclocking and I think there are a lot of bugged submissions. 

 

Example: all over the board, erratic results (same computer)

Another sloppy mess from the clowns at UL.  It is designed for laptops and weaker computing devices, so maybe the thought process is that it is just a toy, not a tool, and accuracy doesn't matter. 
 

The same erratic scoring is visible looking at details in submissions from the 3DMark Hall of Fame.

 

It does pull some hefty watts with the ray tracing in spite of the erratic results. There is no CPU scoring and change to CPU clocks and voltage doesn't seem to affect much.

 

Edited August 15, 2023 by Mr. Fox

[J7SC_Orion]

...Solar Bay is supposed to be 'cross-platform'. I does have a certain amount of run-to-run variance but no worse than Port Royal and Speedway.  My VRAM cooling is working against me as VRAM would need +55 C or so to really oc into the +1650 range but at the same time Solar Bay GPU at 600 W+ also likes good cooling. From what I have read, artifacts in Solar Bay will actually reduce scores significantly, so that's an improvement over earlier benches.

 

I haven't posted this at 3DM yet but this run was at 27 C instead of 29 C ambient for the water-cooled 5950X and 4900. 

 

 

 

 

...I like that older AM4 / DDR4 setup. I am using Win 10 Pro in daily trim and w-cooling...hottest day of the year is not 'ideal' for any of this 

 

 

Edited August 16, 2023 by J7SC_Orion

[Mr. Fox]

I was able to get it to run with RTSS/OSD by manipulating settings in RTSS. I am not running it because it lowers benchmark scores, but I was at least able to get it to stop crashing with RTSS running.

 

My honest opinion is this benchmark just isn't ready for prime time and it is extremely unreliable and flawed. This result is not related to tuning so much as running the benchmark enough times that it eventually pukes up an abnormally high score. Looks like this is the best score I am going to get until I have time to run the benchmark another 20 or 30 times until it calculates a higher score. I am seeing 2500-3500 3DMark points variance between runs without changing any settings. I have also tried two versions of Windows 10 (LTSC 2019 and LTSC 2021) and three versions of Windows 11 (one totally stock with all of the cancer and two mods with Micro$lop trash exorcised) and it doesn't matter how many processes or services are running in terms of benchmark results. I am still seeing lots of illogically inflated results in the HOF with people running the same hardware (CPU, mobo, drivers and vBIOS) with lower overclocks. If I lower my overclock to match the illogically high results, my scores are like 3000-4000 points lower than their reported clocks depending on how many times I re-run it.

 

https://www.3dmark.com/sb/25014

Wild Variance (Same OC values) https://www.3dmark.com/compare/sb/25019/sb/25014#

Edited August 19, 2023 by Mr. Fox

[J7SC_Orion]

5 hours ago, Mr. Fox said:

I was able to get it to run with RTSS/OSD by manipulating settings in RTSS. I am not running it because it lowers benchmark scores, but I was at least able to get it to stop crashing with RTSS running.

 

My honest opinion is this benchmark just isn't ready for prime time and it is extremely unreliable and flawed. This result is not related to tuning so much as running the benchmark enough times that it eventually pukes up an abnormally high score. Looks like this is the best score I am going to get until I have time to run the benchmark another 20 or 30 times until it calculates a higher score. I am seeing 2500-3500 3DMark points variance between runs without changing any settings. I have also tried two versions of Windows 10 (LTSC 2019 and LTSC 2021) and three versions of Windows 11 (one totally stock with all of the cancer and two mods with Micro$lop trash exorcised) and it doesn't matter how many processes or services are running in terms of benchmark results. I am still seeing lots of illogically inflated results in the HOF with people running the same hardware (CPU, mobo, drivers and vBIOS) with lower overclocks. If I lower my overclock to match the illogically high results, my scores are like 3000-4000 points lower than their reported clocks depending on how many times I re-run it.

 

https://www.3dmark.com/sb/25014

Wild Variance (Same OC values) https://www.3dmark.com/compare/sb/25019/sb/25014#

 

I like Solar Bay, also because it is using the Vulkan 1.1 API - the more apps use it, the better, for a variety of reasons. Being on ambient water and using an older 5950X does put some upper limits on for my setup though. FYI, a lot of folks at 3DM Solar Bay HOF have started using sub-ambient cooling (ie. chiller) on their Intel 13900s in combination with the Asus 4090 XOC 1000 W vbios (no safeties). I have that vbios but have not flashed it so far as I am not on sub-ambient or sub-zero; plus I'm quite happy with my latest score, using the 'regular' Galax HOF vbios on my G-G-OC card.

 

In any event, I expect the LN2 crowd to make their entry soon (LN2 on both CPU and GPU) and dominate HOF. Perhaps Splave or so will use it to set some Solar Bay records coinciding with the upcoming release of the Asus 4090 Matrix Platinum...then folks with the Galax 4090 HOF OCL will play their 2x 12VHPWR strengths, and so it goes...off to the races, get the popcorn   

[Avacado]

This is manual 5.8/4.6Ghz. GPU PL @133%, everything else is stock. No custom vBIOS

 

 

Edited August 19, 2023 by Avacado

[Mr. Fox]

34 minutes ago, Avacado said:

This is manual 5.8/4.6Ghz. GPU PL @133%, everything else is stock. No custom vBIOS

 

 

If you run it 15 or 20 times back-to-back you will probably get a score similar to mine. It helped reduce the variance to download and install the latest Vulkan runtime. That kept my low scores from being as low and tightened the range of erratic scores.

8 hours ago, J7SC_Orion said:

 

I like Solar Bay, also because it is using the Vulkan 1.1 API - the more apps use it, the better, for a variety of reasons. Being on ambient water and using an older 5950X does put some upper limits on for my setup though. FYI, a lot of folks at 3DM Solar Bay HOF have started using sub-ambient cooling (ie. chiller) on their Intel 13900s in combination with the Asus 4090 XOC 1000 W vbios (no safeties). I have that vbios but have not flashed it so far as I am not on sub-ambient or sub-zero; plus I'm quite happy with my latest score, using the 'regular' Galax HOF vbios on my G-G-OC card.

 

In any event, I expect the LN2 crowd to make their entry soon (LN2 on both CPU and GPU) and dominate HOF. Perhaps Splave or so will use it to set some Solar Bay records coinciding with the upcoming release of the Asus 4090 Matrix Platinum...then folks with the Galax 4090 HOF OCL will play their 2x 12VHPWR strengths, and so it goes...off to the races, get the popcorn   

I use the 1000W vBIOS exclusively. It produces better results than the 666W Galax vBIOS, which isn't really any better than my stock Suprim firmware in terms of benching.

 

My best results included use of the chiller, but still took a dozen runs to get one bugged enough to make the top 10. Most of the scores higher than mine have higher temperatures and lower overclocks, which is one of the telltale signs of erroneous results. That is why HWBOT is requiring ECC to be enabled for 3DMark. UL is not excluding erroneous results from validation or HOF ranking.

 

I agree about Vulkan. I am all for helping it be successful. I do not like DX12 or Micro$lop, so an alternative is important.

 

Edited August 19, 2023 by Mr. Fox

[J7SC_Orion]

4 hours ago, Mr. Fox said:

If you run it 15 or 20 times back-to-back you will probably get a score similar to mine. It helped reduce the variance to download and install the latest Vulkan runtime. That kept my low scores from being as low and tightened the range of erratic scores.

I use the 1000W vBIOS exclusively. It produces better results than the 666W Galax vBIOS, which isn't really any better than my stock Suprim firmware in terms of benching.

 

My best results included use of the chiller, but still took a dozen runs to get one bugged enough to make the top 10. Most of the scores higher than mine have higher temperatures and lower overclocks, which is one of the telltale signs of erroneous results. That is why HWBOT is requiring ECC to be enabled for 3DMark. UL is not excluding erroneous results from validation or HOF ranking.

 

I agree about Vulkan. I am all for helping it be successful. I do not like DX12 or Micro$lop, so an alternative is important.

 

 

I like to emphasize that bugged scores via VRAM don't really work with Solar Bay, unlike they did with Speedway and PR starting those whole ECC arguments. Several long-term OCer folks on another OC thread mentioned this about Solar Bay, and the two or so bugged runs I saw when the VRAM was cold confirmed it. There is also a related warning that pops up at 3DM. My biggest hold-back is cold VRAM as I encased it in thermal putty, thinking back to the 3090 GDDR6X ovens...turned out to be a mistake I will have to correct; my VRAM oc's get much better when warmer.

 

In any case, VRAM-bugged runs in Solar Bay seem to produce a lower score for that bench as 3DMark staff worked on that issue after 4090 release and Port Royal and Speedway.

 

CPU & RAM are also important and there are a few secrets about Solar Bay I haven't quite got to yet...my best run so far with my trusty old 5950 on ambient w-cooling is below, but I do know that I will fall further behind the more up-to-date CPUs - still my score is about 50+ spots ahead of the next fastest 5950X 4090 combo in HOF.

 

I am waiting re. upgrading this system until the fall when there is more info about new CPUs for AM5 and TR, as well as Intel's new offerings 

 

 
 

Regarding Vulkan, below is a relatively short run with ROG Vulkan Furmark when I let temps come up a bit (and also helping VRAM) with artifact scanner option engaged. As already mentioned, I would love to see Vulkan spread its wings across a lot of major game engines...

 

Edited August 19, 2023 by J7SC_Orion