hashcat Forum

Hello Folks,

I've searched the net and this forum and couldn't find anything about R9 nano yet.

Shouldn't R9 Nano this be a nice choice because a) offering 4096 shaders and b) low power consumption? 
Did use hd7990 before but the power usage is insane and then the temperature problems ... oh boy, ... looking for a fast but less power consuming alternative, currently seriously considering R9 nano but due to lacking benchmarks and experiences i am somewhat hesitating to actually buy one. ...

No, it's not a nice choice at all. The R9 Nano is a piece of shit, stay far away from it. Don't buy into AMD's lies about low power consumption.

You said you had a HD7990 and it had insane power usage and temperature issues -- well the R9 Nano and Fury X are essentially just 2x 7970s packed onto one chip (in other words, a 7990 in a smaller form factor.) It will still draw the same insane amount of power and generate the same amount of heat, except the firmware throttles the hell out of it to keep power and heat down, meaning it runs very very slowly.

I've written a lot about this on the forums, search harder (Hint, search for 390X or Fury instead of Nano. We were referring to the Fury X as the 390X before AMD introduced the "new" GPU as the Fury X, and just rebranded the 290X as the 390X. And the Nano and the Fury X are the same GPU.)

If you're looking for high speeds and low power consumption, consider the GTX 970 or GTX 980 instead. Although compared to AMD GPUs, even the 980Ti and Titan X have low power consumption. These are currently the best four GPUs to use with oclHashcat, there's absolutely no reason to buy an AMD GPU anymore.

Incredible to read this. - Thanks so much for the warning, will look the nvidia direction then.

p.s. what about "gtx 970" vs. "gtx 970 mini"? any drawbacks on the "mini" version, except the fans being axial?

You'll likely be fine with the mini if you want to go that route. Normally I'm vehemently opposed to OEM design cards, but the GTX 970 is kind of the exception to the rule for two reasons,

1. The reference GTX 970 is difficult to get (only available at Best Buy in the US)

2. The card draws at most 155W (but usually less than 140W) which is about how much power a high-end CPU draws, so it's pretty easy to cool

the other thing in mind is, q2 2016 will bring pascal which is rumored to be a LOT faster than maxwell ...

There's no official release date for Pascal. Everything that is out there is all rumors and speculation. And "a lot faster" at pushing pixels rarely translates into faster for password cracking -- we will have to wait and see actual specs to see just how much faster Pascal might be for hashcat (I'm not doubting Pascal will be faster, but historically it's rare to see more than 25-35% performance increase between iterations.)

well, "a lot faster" was mentioned to be "10x faster than maxwell" according to some nvidia presentations i have seen on the net (e.g. https://wccftech.com/nvidia-pascal-gpu-gtc-2015/) ...

of course this won't directly translate into hashcat. agree on the 25-35% historically. they say this time the gap will be bigger though. i've read it's supposed to have 4 TFlops DP and 12 TFlops SP. That should - theoretically - mean approx. 100% performance increase compared to titan x, which would be solid. Theoretically.

The assumption of +100% is based on doing this calculus: number_of_shaders times mhz_core_frequency times 2 = flops -> for titan x that would be 1000*2*3072 = 6 tflops, which is half what pascal is supposed to offer. Since they cannot increase frequency to 2000 MHz they have to add cores to achieve the 12 TFlops SP. Which is just as good for Hashcat. Please correct me if i got this wrong.

Not necessarily. Floating point operations are handled by the FPUs, not the ALUs. Each CUDA core currently has one FPU and one ALU (referred to in the diagram below as the "FP unit and "INT unit.")



So, they could double the FLOPS by simply doubling the FPU count, e.g. each CUDA core could have two FP units and one INT unit, and the scheduler could dispatch four FP32 operations per clock cycle instead of two.

Also with the die shrink, increasing the clock rate is indeed a possibility. 

We won't really know more until actual details are released. Until then this is all just rumors and speculation. Go buy a Maxwell card already

+1 on maxwell. will probably also go the 980 direction, as it's seems to be the tradeoff between performance, power consumption and initial investment.