Benchmarking top results question

[lostsole]

Hi,

I just got a new workstation with all the bells and whistles. I have two Xeon gold 2.2 - 4.0 turbo GHZ at 48 cores total. I have a RTX 3090 GPU. I have 400+Gb memory. And, I bought enterprise U.2 drives at 3K read and 2K write. My temp folder is on the fast drive. I'm running the parallel setting. I just ran the benchmark and only came in at 300+. Others had double or greater the result over me. Where do you suppose I'm bottlenecking compared to top results?

Thanks!

Deven

[lostsole]

As a comparison, with my previous machine, I had dual Xeon 3.0- 4.0 Ghz at 16 total cores, a GTX 1070, 300+ GB memory, and standard mechanical drives only. I was coming in at 270 on the benchmark with that machine. So, the improvement with the new machine is not near what I would have liked to see. Weird.

[Catherine]

First of all, are the benchmarks scores you've mentioned both with OpenCL enabled (300+ on the new machine and 270 on the old one)?

For a CPU score this could be about right, but if it's with OpenCL enabled then both are way too low, for which one reason I could think of is PCI-Express bus issues (e. g. the graphics card not getting the full x16 link).

[lostsole]

Thanks for your reply. Yes, I made sure OpenCL was on for testing. That's why I'm so blown away by the low results especially with a 3090. I do have the U.2 drive taking up a PCIe slot, but that's about it. The old machine had nothing in the PCIe slots. It's strange.

If anyone reads this that has a 600 or greater score, what is the difference between your machine and what I've listed? I'm open to ideas.

Thank you!

[Catherine]

We're getting 460-470 on a slightly overclocked GTX 1080 Ti.
Use the free GPU-Z utility for troubleshooting. watch the GPU utilization on the "Sensors" tab during the benchmark, does it ever hit 100%? And then run it again and watch the PCI-E bus mode on the main tab, does it say @ x16 3.0 (which it should be)?

[lostsole]

Thanks for the advice. With GPU-Z I didn't see a GPU utilization section, but there's a GPU Load. The load said I was getting an avg. of 35ish and a high of 40ish each time I tried it. The PCIe did say 3.0 after I ran the test again.

What do you think?

[lostsole]

If it matters any, I have the Founder's edition of the 3090.

In another thread about building a fast machine, you had mentioned that you guys had not tested the 3090 yet for compatibility. Might that be my issue?

[Catherine]

In another thread about building a fast machine, you had mentioned that you guys had not tested the 3090 yet for compatibility. Might that be my issue?

For all we know, it's not likely (we expect it to either work fine or not work at all), but you could be right. Perhaps it is best to postpone further investigation and let us test these cards first.
Also, I now realize that 100% GPU load is rarely achievable, and certainly not in the benchmark, and your 40% could be just fine. The load percentage is not a good indication of whether or not everything is working properly, only the score is.

[lostsole]

Okay, well thank you for helping me and let us know if you learn anything about the new GPU's. Given the machine I have, the new GPU's seem to be the big question mark for my low benchmarks.