AMD Radeon VII benchmarks(?)

[Jardeath]

Thanks Jardeath. Unfortunately the benchmark is not quite complete, but the results are very mixed - as expected.

I did note there were hash types that did not get tested, but not sure why that was the case, as I ran the all benchmark flag twice and got the same results. I won't have time to do further testing as the Radeon VII is getting sent back tomorrow morning (just heading off to bed), but if there is a better more complete command that I should run to perform the full benchmark let me know for future reference

[undeath]

The problem is not with your benchmark command but with the post I think. It's cut off in the middle during the Truecrypt benchmarks. Or did hashcat crash at that point?

[Jardeath]

The problem is not with your benchmark command but with the post I think. It's cut off in the middle during the Truecrypt benchmarks. Or did hashcat crash at that point?

That's where Hashcat stopped both times.

[joe2gaan]

Just messing around with Hashcat. Figured I would check out the forum. 

  the_experiment_continues_small.jpg (Size: 179.29 KB / Downloads: 31)

[rarecoil]

Bumping this thread with multi-GPU benchmarks on ROCm 2.10, because a new Radeon VII is now $500 on sale as they have been EOL instead of $700. I have not experienced heat issues; the cards stay at about 65-ish C under automatic fan control on stock clocks. These cards cost about $2070 US; I use them mainly for DL/ML work where the 16GB HBM2 per card is helpful.

This benchmark was run after getting the cards up to a stable temperature by running a brute force attempt on a single NTLM hash of all zeros for about an hour.

https://gist.github.com/rarecoil/5434028...5df2535c86

[Dino]

Hello,

I'm not understanding something on the benchmarks. It seems AMD kills Nvidia on 7-zip? Can the difference be this much? (snippets from two different tests people have posted)

System specifications:

* Intel Xeon E5-2680v2
* 128 GB RAM
* 4 XFX Radeon VII
* hashcat v5.1.0-1474-gd315f614 (git)
* rocm-dkms 2.10.14
* Ubuntu 18.04

Stock clocks, stock fan, sclk/mclk variable.

hashcat (v5.1.0-1474-gd315f614) starting in benchmark mode...

OpenCL API (OpenCL 2.1 AMD-APP (3019.0)) - Platform #1 [Advanced Micro Devices, Inc.]
=====================================================================================
* Device #1: gfx906+sram-ecc, 16256/16368 MB (13912 MB allocatable), 60MCU
* Device #2: gfx906+sram-ecc, 16256/16368 MB (13912 MB allocatable), 60MCU
* Device #3: gfx906+sram-ecc, 16256/16368 MB (13912 MB allocatable), 60MCU
* Device #4: gfx906+sram-ecc, 16256/16368 MB (13912 MB allocatable), 60MCU

Benchmark relevant options:
===========================
* --optimized-kernel-enable
* --workload-profile=3

Hashmode: 11600 - 7-Zip (Iterations: 16384)

Speed.#1.........: 426.1 kH/s (62.20ms) @ Accel:32 Loops:1024 Thr:256 Vec:1
Speed.#2.........: 417.6 kH/s (63.33ms) @ Accel:32 Loops:1024 Thr:256 Vec:1
Speed.#3.........: 416.4 kH/s (63.93ms) @ Accel:32 Loops:1024 Thr:256 Vec:1
Speed.#4.........: 418.9 kH/s (63.50ms) @ Accel:32 Loops:1024 Thr:256 Vec:1
Speed.#*.........: 1679.0 kH/s

--- AND ----


Benchmark Hashcat v5.1.0 on 10 * GTX 2080 Ti

Hashcat version: 5.1.0 (2018.12.02)
Hashcat options: -b -O -w 4
Nvidia GPUs: 10 * RTX 2080 Ti

Hashmode: 11600 - 7-Zip (Iterations: 524288)

Speed.#1.........: 23003 H/s (47.20ms) @ Accel:32 Loops:1024 Thr:256 Vec:1
Speed.#2.........: 22867 H/s (47.48ms) @ Accel:32 Loops:1024 Thr:256 Vec:1
Speed.#3.........: 22979 H/s (47.25ms) @ Accel:32 Loops:1024 Thr:256 Vec:1
Speed.#4.........: 22479 H/s (48.31ms) @ Accel:32 Loops:1024 Thr:256 Vec:1
Speed.#5.........: 22757 H/s (47.71ms) @ Accel:32 Loops:1024 Thr:256 Vec:1
Speed.#6.........: 22602 H/s (48.06ms) @ Accel:32 Loops:1024 Thr:256 Vec:1
Speed.#7.........: 22781 H/s (47.68ms) @ Accel:32 Loops:1024 Thr:256 Vec:1
Speed.#8.........: 22709 H/s (47.84ms) @ Accel:32 Loops:1024 Thr:256 Vec:1
Speed.#9.........: 22596 H/s (48.07ms) @ Accel:32 Loops:1024 Thr:256 Vec:1
Speed.#10.........: 22916 H/s (47.40ms) @ Accel:32 Loops:1024 Thr:256 Vec:1
Speed.#*.........: 227.7 kH/s

These are wildly different (H/s vice kH/s)?

Go easy on me. I'm new at this.

[Chick3nman]

Hashmode: 11600 - 7-Zip (Iterations: 16384)

Hashmode: 11600 - 7-Zip (Iterations: 524288)

The iterations in your Nvidia test are 32x higher. If you multiple the speeds achieved on the 2080Ti's by 32x to bring it up to the same easy difficulty, they would be running ~700kh/s each theoretically. Well out classing the Radeon VII.

[rarecoil]

are you aware of a way to fix the iteration counts hashcat decides to use for benchmarking? the variability makes easy comparison difficult on tests like this.