hashcat not using multiple GPUs on MacOS #2083

singe · 2019-07-03T13:50:37Z

After commit e3500ff hashcat won't make use of the Intel and AMD GPU on a MBP 2016 running MacOS Mojave 10.14.5.

At the last working commit, running gave:
hashcat (v5.1.0-968-ge3500ff4) starting in benchmark mode...

[snip CUDA warnings ...]

OpenCL Platform #1: Apple
=========================
* Device #1: Intel(R) Core(TM) i7-6820HQ CPU @ 2.70GHz, skipped.
* Device #2: Intel(R) HD Graphics 530, 384/1536 MB allocatable, 24MCU
* Device #3: AMD Radeon Pro 455 Compute Engine, 512/2048 MB allocatable, 12MCU

The first commit that introduces the bad behaviour is 6fd936b), where running gives:

hashcat (v5.1.0-969-g6fd936b4) starting in benchmark mode...

OpenCL API (OpenCL 1.2 (Apr  7 2019 18:38:19)) - Platform #1 [Apple]
====================================================================
* Device #1: Intel(R) Core(TM) i7-6820HQ CPU @ 2.70GHz, skipped
* Device #2: Intel(R) HD Graphics 530, 384/1536 MB allocatable, 24MCU
* Device #3: AMD Radeon Pro 455 Compute Engine, skipped

As you can see only device 2 is used, and the second GPU device 3 is skipped. Device 3 can be forced with -d3, but then device 2 will be skipped. Specifying -d2,3 will just use device 2.

The text was updated successfully, but these errors were encountered:

matrix · 2019-07-08T15:46:27Z

@singe I have the same issue with a previous version of OSX.
I'll let you know as soon as I understand the cause.

matrix · 2019-07-08T16:48:04Z

@singe I found the "bug". Try manually patch src/backend.c and retry (using -D1,2 --force) :

-  //if (user_options->force == false)
-  //{
-  backend_ctx_find_alias_devices (hashcat_ctx);
-  //}
+  if (user_options->force == false)
+  {
+    backend_ctx_find_alias_devices (hashcat_ctx);
+  }

Let me known :)

matrix · 2019-07-11T11:25:56Z

@singe could you try to apply PR #2086 patch to check if the bug is out ?
Thanks!

singe · 2019-07-14T19:29:57Z

Confirmed that fixes the bug, thanks!

jsteube · 2019-08-06T10:45:58Z

Commit 57a1492 should fix this issue. Please check and reopen if it's not working.

singe · 2020-06-17T11:04:44Z

Looks like this issue is back with the latest 6.0.0 release and commit. I did a quick check, and applying the same patch @matrix did earlier to allow --force fixes it, so there's no technical reason why it shouldn't work, just looks like an error in the selection logic I haven't had time to dig into yet.

The diff:

diff --git a/src/backend.c b/src/backend.c
index 80fdbb38..e0c9246d 100644
--- a/src/backend.c
+++ b/src/backend.c
@@ -6606,7 +6606,7 @@ int backend_ctx_devices_init (hashcat_ctx_t *hashcat_ctx, const int comptime)
     // using force here enables both devices, which is the worst possible outcome
     // many users force by default, so this is not a good idea
 
-    //if (user_options->force == false)
+    if (user_options->force == false)
     //{
     backend_ctx_find_alias_devices (hashcat_ctx);
     //{

For completeness sake, here's the output from hashcat 5.1.0, 6.0.0 and 6.0.0 with my patch and --force:

5.1.0 (homebrew)

/usr/local/Cellar/hashcat/5.1.0/bin/hashcat -m1000 --benchmark
hashcat (v5.1.0) starting in benchmark mode...

Benchmarking uses hand-optimized kernel code by default.
You can use it in your cracking session by setting the -O option.
Note: Using optimized kernel code limits the maximum supported password length.
To disable the optimized kernel code in benchmark mode, use the -w option.

OpenCL Platform #1: Apple
=========================
* Device #1: Intel(R) Core(TM) i7-6820HQ CPU @ 2.70GHz, skipped.
* Device #2: Intel(R) HD Graphics 530, 384/1536 MB allocatable, 24MCU
* Device #3: AMD Radeon Pro 455 Compute Engine, 512/2048 MB allocatable, 12MCU

Benchmark relevant options:
===========================
* --optimized-kernel-enable

Hashmode: 1000 - NTLM

Speed.#2.........:   739.5 MH/s (32.47ms) @ Accel:128 Loops:32 Thr:256 Vec:1
Speed.#3.........:  4081.5 MH/s (95.39ms) @ Accel:512 Loops:256 Thr:256 Vec:1
Speed.#*.........:  4821.0 MH/s

Started: Wed Jun 17 13:01:44 2020
Stopped: Wed Jun 17 13:01:55 2020

6.0.0

./hashcat -m1000 --benchmark
hashcat (v6.0.0) starting in benchmark mode...

Benchmarking uses hand-optimized kernel code by default.
You can use it in your cracking session by setting the -O option.
Note: Using optimized kernel code limits the maximum supported password length.
To disable the optimized kernel code in benchmark mode, use the -w option.

OpenCL API (OpenCL 1.2 (May  3 2020 20:15:19)) - Platform #1 [Apple]
====================================================================
* Device #1: Intel(R) Core(TM) i7-6820HQ CPU @ 2.70GHz, skipped
* Device #2: Intel(R) HD Graphics 530, 1472/1536 MB (384 MB allocatable), 24MCU
* Device #3: AMD Radeon Pro 455 Compute Engine, skipped

Benchmark relevant options:
===========================
* --optimized-kernel-enable

Hashmode: 1000 - NTLM

Speed.#2.........:   669.5 MH/s (73.83ms) @ Accel:1024 Loops:256 Thr:8 Vec:1

Started: Wed Jun 17 13:00:49 2020
Stopped: Wed Jun 17 13:00:54 2020

6.0.0 patched with --force

./hashcat -m1000 --benchmark --force
hashcat (v6.0.0) starting in benchmark mode...

Benchmarking uses hand-optimized kernel code by default.
You can use it in your cracking session by setting the -O option.
Note: Using optimized kernel code limits the maximum supported password length.
To disable the optimized kernel code in benchmark mode, use the -w option.

You have enabled --force to bypass dangerous warnings and errors!
This can hide serious problems and should only be done when debugging.
Do not report hashcat issues encountered when using --force.
OpenCL API (OpenCL 1.2 (May  3 2020 20:15:19)) - Platform #1 [Apple]
====================================================================
* Device #1: Intel(R) Core(TM) i7-6820HQ CPU @ 2.70GHz, skipped
* Device #2: Intel(R) HD Graphics 530, 1472/1536 MB (384 MB allocatable), 24MCU
* Device #3: AMD Radeon Pro 455 Compute Engine, 1984/2048 MB (512 MB allocatable), 12MCU

Benchmark relevant options:
===========================
* --force
* --optimized-kernel-enable

Hashmode: 1000 - NTLM

Speed.#2.........:   674.1 MH/s (73.28ms) @ Accel:1024 Loops:256 Thr:8 Vec:1
Speed.#3.........:  4171.1 MH/s (95.53ms) @ Accel:1024 Loops:512 Thr:64 Vec:1
Speed.#*.........:  4845.2 MH/s

Started: Wed Jun 17 13:01:06 2020
Stopped: Wed Jun 17 13:01:11 2020

jsteube · 2020-06-17T11:57:33Z

I have no way to reproduce this because I do not have access to an OSX system with AMD GPU. Therefore I can't patch the sources, because I can't reproduce if the patch is actually doing what it should do.

However, the original issue is very different from the problem you have now, because the device is not being ignored but not being selected. IOW, the issue is a very different one, even if it seems the same.

singe · 2020-06-17T15:21:02Z

Ok, I'll open a new issue. Happy to do some testing when I get some time and try submit a patch.

This is a repeat of hashcat@57a1492 which was undone in this commit 34f71aa#diff-7af547b52769371de74af0bfca268a62R6599 The original issue was hashcat#2083 and discussion of a fix was hashcat#2086. The problem is with Apple devices with an Intel and AMD GPU and using both simultaneously and the alias check.

matrix added a commit to matrix/hashcat that referenced this issue Jul 8, 2019

Fix hashcat#2083: hashcat not using multiple GPUs on MacOS

6d5c24e

matrix mentioned this issue Jul 8, 2019

Fix #2083: hashcat not using multiple GPUs on MacOS #2086

Closed

jsteube closed this as completed Aug 6, 2019

singe mentioned this issue Jun 17, 2020

macOS AMD can't use multiple GPUs #2460

Closed

singe mentioned this issue Jun 22, 2020

Re-implement alias check #2466

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

hashcat not using multiple GPUs on MacOS #2083

hashcat not using multiple GPUs on MacOS #2083

singe commented Jul 3, 2019 •

edited

Loading

matrix commented Jul 8, 2019

matrix commented Jul 8, 2019 •

edited

Loading

matrix commented Jul 11, 2019

singe commented Jul 14, 2019

jsteube commented Aug 6, 2019

singe commented Jun 17, 2020

jsteube commented Jun 17, 2020

singe commented Jun 17, 2020 •

edited

Loading

hashcat not using multiple GPUs on MacOS #2083

hashcat not using multiple GPUs on MacOS #2083

Comments

singe commented Jul 3, 2019 • edited Loading

matrix commented Jul 8, 2019

matrix commented Jul 8, 2019 • edited Loading

matrix commented Jul 11, 2019

singe commented Jul 14, 2019

jsteube commented Aug 6, 2019

singe commented Jun 17, 2020

jsteube commented Jun 17, 2020

singe commented Jun 17, 2020 • edited Loading

singe commented Jul 3, 2019 •

edited

Loading

matrix commented Jul 8, 2019 •

edited

Loading

singe commented Jun 17, 2020 •

edited

Loading