Speed on Radeon HD6970

[paul6990]

Hi guys,

Anybody can please share the speed of the latest oclHashcat running on a Radeon HD6970, for sha512crypt and md5crypt (-m 1800 and -m 500)?

I thinking of buying a few of them (found in good condition, second hand) for auditing sha512crypt and md5crypt hashes and would love to know the speeds.

I notice atom's 6990 does over 30K/s for sha512crypt, so I presume a 6970 would do at least 15K.

Thanks in advance.

[epixoip]

correct, you can essentially cut atom's 6990 speeds in half. it will be a little bit faster due to higher clocks, but that will give you a good estimate.

I believe d3ad0ne may also have some 6970 benchmarks posted at http://ob-security.info/ as he was using 6970s before he upgraded to 7970s.

6970 is currently the 4th best card you can buy at the moment; general order of preference being 6990, 7970, 5970, 6970, 5870 (although #1 and #2 are open for debate, i personally prefer 7970 over 6990.) so if you can get 6970s for a good price, go for it.

[paul6990]

Thank you! Unfortunately, when d3ad0ne did his stats, sha512crypt was not available in hashcat. But if it's indeed 6990's divided by 2, then it works quite well for me. That actually means the 6970 is faster than 7970 from this point of view.

[blandyuk]

I know 7970s are faster do to memory speeds but I found the 6000 series to be a bit power hungry. See below:

http://www.gat3way.eu/est.php

5870: MD5 speed: 5821 ~ Watts: 188
6970: MD5 speed: 5878 ~ Watts: 250

Not sure how accurate the wattage is but if it's right, the 6970 uses 60 more watts for no performance gain!

I ended up buying 5 x 5870s off ebay for about £100 ($160) each.

[paul6990]

Hi Blandyuk,

Thanks for the info! Especially the list of different power consumptions is very useful!!

BTW, what is the sha512crypt (-m 1800) speed on a 5870 please?

[KT819GM]

On high iterated hashes 5 series slower than 6 by ~10% at same clocks, maybe because of VLIV? You can check benchmarks thread, there are 5970 and 6990 all hashes.

[blandyuk]

Speeds on 5870s for you with 1 x SHA512(Unix) using -n 160:

Code:

1 x 5870, 850MHz core, 1200MHz mem: 15158 c/s
4 x 5870, 850MHz core, 1200MHz mem: 60422 c/s

1 x 5870, 900MHz core, 1250MHz mem: 16036 c/s
4 x 5870, 900MHz core, 1250MHz mem: 63970 c/s

Can't help but notice the stats on here: http://hashcat.net/oclhashcat-plus/

How can a stock-core HD 7970 be slower than my stock-core HD 5870???

[paul6990]

Hi Blandyuk!

Thanks for the stats, your 5870s are pretty badass! Indeed, a 7970 is slower. I have two 7970's myself and both do around 13K sha512crypt / sec.

Not sure if it's a bug, but this is indeed the speed on 7970.

I already purchased the 6970s, will let you know how they fare when they arrive.

[epixoip]

I know 7970s are faster do to memory speeds but I found the 6000 series to be a bit power hungry. See below:

7970s are faster due to higher clock rates and the new GCN architecture.

Not sure how accurate the wattage is but if it's right, the 6970 uses 60 more watts for no performance gain!

The power consumption of the 5000-series is phenomenal, yes. But 6970 does edge out the 5870 for a few algorithms that benefit from the 6970's VLIW4 architecture vs the 5870's VLIW5 architecture, particularly DES-based hashes like descrypt and LM. Other slow and highly-iterated hashes will benefit from VLIW4 as well, and there's potentially even more benefit from GCN.

As amazing as the 5870s and 5970s are -- especially in the perf-per-watt area -- the only problem with buying a 5000-series card right now is that AMD will likely deprecate them within the next 12-18 months. 6000-series should still be good for another 2-3 years.

Indeed, a 7970 is slower. Not sure if it's a bug, but this is indeed the speed on 7970.

Probably not a bug per se, but there's likely room for optimization.

I think what we're experiencing here is VLIW5 has been around for quite a while now, and most all of the ways to optimize for that architecture have already been discovered and implemented.

GCN is very much superior to VLIW, but it's a brand-new architecture. I think that over time, new optimizations will be discovered that will allow hashcat to take full advantage of the hardware. And this applies not only to hashcat's kernels, but to the driver development as well.