Benchmarking: Geekbench-6 GPU

[J7SC_Orion]

Here is Geekbench 6 / CPU as an example of why tuning a Ryzen with PBO, CO and DynamicOC (if available) can really make a difference. FYI, the highest score on the right of the 5950X is not 'max' frequency for this CPU can do (=upper right pic), just a 24/7 daily with about 5080 +- on the best core and low voltage. The mobo is the Asus X570 DarkHero; OS is Win 10 Pro, ambient was 24 C.

 

I will do a full max-everything run soon but for comparative purposes re. different Ryzen bios features, this works well, IMO 

 

 

 

edit:

kicking up the multi-score a bit (at the same v-core )

 

Edited March 10, 2023 by J7SC_Orion

[Mr. Fox]

So, got the block installed. The kit did not have screws for the backplate (had short screws for naked installation). Screws should be here from Amazon in the next hour or so. I am hoping having pads behind the core will help hotspot temps (still quite high, but better). Have not touched the firmware yet. Will dive into MPT later today or tomorrow, once I get the temps as good as I can get them without the chiller.  It will get better, as it is hitting the power limit and remaining there throughout all of the graphics tests. Totally maxed out with the stock firmware limits and can't do much more until I raise the limits.

 

https://hwbot.org/submission/5222705_ | https://www.3dmark.com/3dm/90656314#

Edited March 10, 2023 by Mr. Fox

[J7SC_Orion]

2 hours ago, Mr. Fox said:

So, got the block installed. The kit did not have screws for the backplate (had short screws for naked installation). Screws should be here from Amazon in the next hour or so. I am hoping having pads behind the core will help hotspot temps (still quite high, but better). Have not touched the firmware yet. Will dive into MPT later today or tomorrow, once I get the temps as good as I can get them without the chiller.  It will get better, as it is hitting the power limit and remaining there throughout all of the graphics tests. Totally maxed out with the stock firmware limits and can't do much more until I raise the limits.

 

https://hwbot.org/submission/5222705_ | https://www.3dmark.com/3dm/90656314#

 

...you are getting there !

FYI, my stock Giga-G-OC backplate actually had a soft thermal pad on the back of the GPU die. When I w-cooled it with a custom backplate that came with the kit, I used a wallop of thermal putty on the back of the GPU die, then also mounted a huge heatsink...seems to work well, judging by the temps, even at elevated PL via MPT.

 

 

 

 

Edited March 10, 2023 by J7SC_Orion

[Mr. Fox]

42 minutes ago, J7SC_Orion said:

 

...you are getting there !

FYI, my stock Giga-G-OC backplate actually had a soft thermal pad on the back of the GPU die. When I w-cooled it with a custom backplate that came with the kit, I used a wallop of thermal putty on the back of the GPU die, then also mounted a huge heatsink...seems to work well, judging by the temps, even at elevated PL via MPT.

 

 

 

 

Yeah that looks great. If the hotspot temps are still high I may try that. I added a thermal pad on the backplate side directly over the die area.

I am not even sure if the hotspot readings are accurate. Before installing the backplate, I used an infrared thermometer and with the GPU loaded full tile nothing measured over 50°C anywhere, yet the hotspot still showed 90°C. That's why I am thinking the reading may be wrong.

 

But, I got the screws and now the backplate is installed. Time to do a bit of quick studying and get cracking the firmware tweaking.

[J7SC_Orion]

...here is Geekbench 6 for CPUs on the 2950X Threadripper.

That older system now does duty as a render and ML setup since it might be slower on the processor but it has plenty of fast 4-channel RAM, and a RTX 3090 + 2x 2080 Ti (everything w-cooled). I sometimes even use it as a secondary gamer (w/ RTX 3090 only) -- at 4K, it actually works really well for gaming.

  

 

[bonami2]

3028

Single-Core Score

20872

Multi-Core Score

Micro-Star International Co., Ltd. MS-7E03 - Geekbench Browser

BROWSER.GEEKBENCH.COM

---

Benchmark results for a Micro-Star International Co., Ltd. MS-7E03 with an Intel Core i9-13900K processor.

 

 

[J7SC_Orion]

8 minutes ago, bonami2 said:

3028

Single-Core Score

20872

Multi-Core Score

Micro-Star International Co., Ltd. MS-7E03 - Geekbench Browser

BROWSER.GEEKBENCH.COM

---

Benchmark results for a Micro-Star International Co., Ltd. MS-7E03 with an Intel Core i9-13900K processor.

 

 

 

...clicked on your link  for details - says it is invalid ?  ?

 

 

[bonami2]

7 minutes ago, J7SC_Orion said:

 

...clicked on your link  for details - says it is invalid ?  ?

 

 

 

 

No idea what wrong with it.

 

Am reinstalling windows right now because of some issue. Will retry.

[J7SC_Orion]

23 minutes ago, bonami2 said:

 

 

No idea what wrong with it.

 

Am reinstalling windows right now because of some issue. Will retry.

 

...haven't seen that before in Geekbench 6, but in other benchies, a RAM  or cache OC on the edge (or just a bit over) can result in 'timer' error messages. If it still persist, I would  test-run Geekbench-6 with HWInfo open re. WHEA errors (bottom entry)

 

 

 

Edited March 11, 2023 by J7SC_Orion

[bonami2]

3 hours ago, J7SC_Orion said:

 

...haven't seen that before in Geekbench 6, but in other benchies, a RAM  or cache OC on the edge (or just a bit over) can result in 'timer' error messages. If it still persist, I would  test-run Geekbench-6 with HWInfo open re. WHEA errors (bottom entry)

 

 

 

Rerun it without issue here a valid one  

Micro-Star International Co., Ltd. MS-7E03 - Geekbench Browser

BROWSER.GEEKBENCH.COM

---

Benchmark results for a Micro-Star International Co., Ltd. MS-7E03 with an Intel Core i9-13900K processor.

 

[J7SC_Orion]

14 minutes ago, bonami2 said:

Rerun it without issue here a valid one  

Micro-Star International Co., Ltd. MS-7E03 - Geekbench Browser

BROWSER.GEEKBENCH.COM

---

Benchmark results for a Micro-Star International Co., Ltd. MS-7E03 with an Intel Core i9-13900K processor.

 

 

[J7SC_Orion]

...quick ' fyi ' on power_hogs / pigs on gas watts...

 

As discussed before with @Bastiaan_NL and others, Geekbench 6 GPU OpenCL is light on power usage, but I accidentally ran my RTX 4090 on 'max PL / Galax' vbios all day . Near the end I tried a quick TimeSpy Extreme GT2 to check the new driver via MSI AB sliders (not the max bench curve profile shown at the bottom) . To my horror, that turned out to be 625 W ; good thing I had HWInfo minimized in the system tray and checked afterwards. 

 

...I have exceeded 671 W before with ROG Furmark on this Galax HOF vbios but that was a brief test and on purpose when I was aware of the full PL. This on the other hand had been used for various apps for a longer time, though TimeSpy Extreme GT2 is a known power hog. Fortunately, the system has strong (but silent) w-cooling & also a good thing I did not yet replace the Galax HOF 'regular' vbios with the Asus XOC (shows 900W+ on my GPU model) I downloaded. Still, the RTX 4090 is perfectly happy and delivers great fps at ~400 W; 600W+ for longer periods makes me queasy.

 

Moral of the story: If you have a switchable vbios on your GPU, better remember which one you're on 

 

[tictoc]

Here's my 6900xt on Linux running the Vulkan compute benchmark.

 

 

https://browser.geekbench.com/v6/compute/187133

 

[Nikado7]

All i got is cpu for now.  GPU's are uh, its complicated lol.  

 

ASRock B650E Taichi - Geekbench Browser

BROWSER.GEEKBENCH.COM

---

Benchmark results for an ASRock B650E Taichi with an AMD Ryzen 9 7950X processor.

 

[J7SC_Orion]

1 hour ago, tictoc said:

Here's my 6900xt on Linux running the Vulkan compute benchmark.

 

 

https://browser.geekbench.com/v6/compute/187133

 

  

Nice ! I also noticed that Vulkan is a bit faster than Open CL with RDNA2 and some older RTX...While on the topic of Vulkan, too bad Microsoft won't allow Vulkan for titles such as FS 2020 for obvious reasons of their own...I actually prefer Vulkan over DX12 (and Open CL) in apps where choice is available.

[tictoc]

Vulkan compute on this bench is overall substantially better than Geekbench's OpenCL on Linux.  The Linux OpenCL score is quite a bit slower than Windows, and I imagine my 6900xt running 2950core/2150mem is clocked higher than  the vast majority of 6900xt submissions.

 

Here's an OpenCL compute run for comparison.

 

   

Edited March 13, 2023 by tictoc