[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20200805182702.GA12722@openwall.com>
Date: Wed, 5 Aug 2020 20:27:02 +0200
From: Solar Designer <solar@...nwall.com>
To: john-users@...ts.openwall.com
Subject: Re: sha512crypt-opencl / Self test failed (cmp_all(1))
On Tue, Aug 04, 2020 at 03:04:28PM +0200, Solar Designer wrote:
> sha512crypt-opencl works for me on Vega 64 with:
>
> Device #0 (1) name: gfx900
> Board name: Radeon RX Vega
> Device vendor: Advanced Micro Devices, Inc.
> Device type: GPU (LE)
> Device version: OpenCL 2.0 AMD-APP (2766.4)
> Driver version: 2766.4 (PAL,HSAIL) - AMDGPU-Pro
>
> but the speed is very poor (for that hardware):
>
> LWS=64 GWS=262144 (4096 blocks) DONE
> Speed for cost 1 (iteration count) of 5000
> Raw: 65372 c/s real, 5242K c/s virtual
As Claudio pointed out to me, that must have been a temporary glitch -
maybe the GPU was in other use at that time. Here's what I am getting
now that I rerun the test:
$ ./john -test -form=sha512crypt-opencl
Device 1: gfx900 [Radeon RX Vega]
Benchmarking: sha512crypt-opencl, crypt(3) $6$ (rounds=5000) [SHA512 OpenCL]... LWS=256 GWS=262144 (1024 blocks) DONE
Speed for cost 1 (iteration count) of 5000
Raw: 129453 c/s real, 5242K c/s virtual
Even forcing it to LWS=64 (which it had autotuned to before), I get
almost the same better speed:
$ LWS=64 ./john -test -form=sha512crypt-opencl
Device 1: gfx900 [Radeon RX Vega]
Benchmarking: sha512crypt-opencl, crypt(3) $6$ (rounds=5000) [SHA512 OpenCL]... LWS=64 GWS=262144 (4096 blocks) DONE
Speed for cost 1 (iteration count) of 5000
Raw: 127875 c/s real, 4032K c/s virtual
which is reasonable scaling from:
> It also works for me on Tahiti with this ancient driver:
>
> Device #0 (3) name: Tahiti
> Board name: AMD Radeon HD 7900 Series
> Device vendor: Advanced Micro Devices, Inc.
> Device type: GPU (LE)
> Device version: OpenCL 1.2 AMD-APP (1642.5)
> Driver version: 1642.5 (VM) - Catalyst 14.12(Omega) [supported]
>
> delivering a saner speed (for that hardware):
>
> LWS=64 GWS=262144 (4096 blocks) DONE
> Speed for cost 1 (iteration count) of 5000
> Raw: 51150 c/s real, 2621K c/s virtual
Vega 64 is a twice larger GPU at 1.5x higher clock rate than Tahiti, so
a ~3x difference is expected.
Alexander
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
