hashcat Forum
A few comments/questions regarding v3.00 - Printable Version

+- hashcat Forum (https://hashcat.net/forum)
+-- Forum: Support (https://hashcat.net/forum/forum-3.html)
+--- Forum: hashcat (https://hashcat.net/forum/forum-45.html)
+--- Thread: A few comments/questions regarding v3.00 (/thread-5623.html)



A few comments/questions regarding v3.00 - NikosD - 07-09-2016

Hello.
The latest hashcat v3.00 is a fantastic work enabling for the first time all OpenCL devices and the simultaneous use of them.

A few comments/ questions:

1) On my ancient Radeon 5750 (Juniper) I saw a huge drop in performance compared to v2.01 using the exact same system (Win 10 x64 - Catalyst 15.11.1 - Radeon 5750) on various popular algorithms like:

LM v2.01 : 1238 MH/s -> 190 MH/s (!) v3.00
WPA v2.01: 36000 H/s -> 19750 v3.00
7-zip v2.01:  639 H/s -> 250 H/s v3.00

Are you going to restore previous performance of old ATI cards or not interested in supporting such old devices ?

2) If I understand correctly the HUGE performance increase in my CPU (Haswell Core i7-4790) is the result of rebuilding from scratch the SIMD engine.

But is CPU supported via OpenCL only ?

I mean the previous hashcat (hashcat legacy) could use the CPU directly without OpenCL, although it was obviously not optimized.

Wouldn't hashcat be a lot faster if you supported the same new optimized SIMD engine directly without OpenCL's overhead ?

For the record I get ~ 1050 MH/s (!) on MD4 and NTLM but only 110 MH/s on LM

3) Is the NVIDIA's OpenCL implementation/support on par with AMD ?
I think they only care about CUDA and I'm not sure they have a proper OpenCL implementation for their cards.

4) Does NVIDIA have an OpenCL 2.x driver ? 
    Does it matter ? 

    I mean is OpenCL v2.x any faster than v1.x ?

Thanks!


RE: A few comments/questions regarding v3.00 - atom - 07-09-2016

(07-09-2016, 06:59 PM)NikosD Wrote: Are you going to restore previous performance of old ATI cards or not interested in supporting such old devices ?

I would, but I can't, because AMD dropped support for OpenCL in their driver for those old cards.

(07-09-2016, 06:59 PM)NikosD Wrote: But is CPU supported via OpenCL only ?

Yes

(07-09-2016, 06:59 PM)NikosD Wrote: Wouldn't hashcat be a lot faster if you supported the same new optimized SIMD engine directly without OpenCL's overhead ?


Not much but a bit, yes

(07-09-2016, 06:59 PM)NikosD Wrote: 3) Is the NVIDIA's OpenCL implementation/support on par with AMD ?
I think they only care about CUDA and I'm not sure they have a proper OpenCL implementation for their cards.


It changed with maxwell, they show very good OpenCL performance

(07-09-2016, 06:59 PM)NikosD Wrote: 4) Does NVIDIA have an OpenCL 2.x driver ? 


No

(07-09-2016, 06:59 PM)NikosD Wrote:
    Does it matter ? 


No

(07-09-2016, 06:59 PM)NikosD Wrote:
    I mean is OpenCL v2.x any faster than v1.x ?


No


RE: A few comments/questions regarding v3.00 - NikosD - 07-09-2016

Thank you for your prompt and detailed reply.

Well, Crimson drivers have indeed dropped support of legacy ATI/ AMD cards regarding OpenCL, but you could always use the good old Catalyst drivers that still support OpenCL on those.

Regarding Nvidia, the situation is like the new hashcat is "pushing" users to change their pre-Maxwell HW.

I hope they go to AMD though, for many many reasons.

I'm waiting for RX 470 in custom version (probably from Sapphire) that could be ~5-6 times faster than 5750, especially if it's a little factory overclocked.

I mean, it's almost unbelievable that you can buy 5 TFLOPS (and the integer speed for hashcat) with 150$ (!)

Best VFM for sure.


RE: A few comments/questions regarding v3.00 - atom - 07-11-2016

Quote:Well, Crimson drivers have indeed dropped support of legacy ATI/ AMD cards regarding OpenCL, but you could always use the good old Catalyst drivers that still support OpenCL on those.

Not really. With new drivers there's new features and some of them require to use special instructions. Those new instruction would be missing in older drivers, causing the entire process to fail.