AMD Radeon VII benchmarks(?)

does anybody here have got the benchmarks on AMD Radeon VII? I know it is a little bit early since today is the release date and end of performance embargo I'm just curious since the gaming and rendering benchmarks are already online and some OpenCL benchmarks show somewhat competitive values.

I saw a post regarding Ethereum performance and it goes like this:

Vega64 ~ 45 MH/s

Titan V ~ 70 MH/s

Radeon 7 ~ 90 MH/s (unoptimized - out of the box)

Read here:

It could have a great performance for cracking too and better power efficiency with undervolting.
Ether is just one hash algorithm and it prefers AMDs architecture. Plus fast memory, while hashcat does not. Not comparable at all.

I'd be surprised if the Radeon VII will be on par with the RTX2080. It would need to be 50% faster than the Vega 64, but with lesser cores and only a small bump in clock rate that might be difficult. But then AMD might *finally* got rid of the throttling problems, so who knows.

It anyway shows that there is still a long way to AMDs Navi. Otherwise they would not have launched the Radeon VII.
The Vega20 inside Radeon 7 is not just a die shrink to 7nm of Vega10.

It has A LOT of minor changes in the rendering and compute pipelines, adding 700M more transistors and e.g additional integer accumulators possibly useful for hashcat.

The result of those changes is a faster Radeon 7 GPU clock-for-clock than Vega64, according to Anandtech's tests.

Adding a faster clock, double memory bandwidth and VRAM to the equation give us a faster performance from 30% to 100% than Vega64.

I think we could see a lot of surprises running hashcat with this beast.
I'm afraid it is just a die shrink, since the architecture behind is still GCN5, although a bit tweaked. And again, faster and bigger memory is almost irrelevant for hashcat.

Take the Anandtech benches for GPU computing and you'll find roughly +30%, +40% and +20% over the Vega 64 in the three Compubenches and only +12% in the folding bench. The Geekbench , where the RTX2080 made a huge leap forward over the GTX1080, we see a merely +3% from the Rdeon VII over the Vega 64. The GP Processing bench shows a +20% and fits into the spectrum.

By taking the +3% to +40% in these OpenCL benches an overall +50% for hashcat would be a big surprise. But, of course, in certain hash modes there might be an advantage over the RTX2080. In any way it would only reach the latter with the RTX 2080Ti staying far ahead.

Considering the big potential of the 7nm technology over NVidias 12nm it shows very poor performance. Even if it could reach the RTX2080 in hashcat, it would still draw much more power with its 300Watts and all that at an equal price. No recommendation in sight.
Judging by real-world tests and not just specs, we can already see a 100% faster performance in Ethereum for Radeon VII vs Vega64, out of the box.

Also, TDP of 2080 is a lot higher than previous generation Pascals and again by seeing real-world performance figures the two cards - Radeon VII and 2080 - are very close regarding power consumption.

For AMD cards is very common nowadays to achieve great power efficiency using mild undervolting while on the same time you can actually overclock the card!

Before recommending or not the card, it will be wiser to wait - as always - real world performance numbers by running hashcat benchmarks and actual cracking using the GPU.
You're cherry picking. As mentioned, Etherum takes a big profit out of fast memory. Meaningless for hashcat.

The 2080 tops out at 225W if not tweaked. No idea how you manage to see it anywhere close to the >300W the Radeon VII uses under heavy load. Check out tomshardware, they measured the card very specific on all power rails.

<performance, >power, =price -> no reason to buy. I'd be surprised if it's going to be any different and you're allowed to sneer at me if so Wink
(02-07-2019, 07:02 PM)azaran Wrote: does anybody here have got the benchmarks on AMD Radeon VII?

Here you go:
Thank you very much.
Thanks Jardeath. Unfortunately the benchmark is not quite complete, but the results are very mixed - as expected.

Fisrt glance: The Radeon VII can most often not even gain a +20% over the Vega 64. In fast hashes like MD5 or NTLM it's even only 10% or lower. Even in hash modes where the Vega 64 already had been strong (NTLM, bcrypt) the Radeon VII is only a bit (~10%) faster.

It can sometimes match the RTX 2080 (e.g. NTLM, NetNTLMv1 , NetNTLMv1, TrueCrypt 512bit) and at the same time it falls (way) behind in most benchmarks.

Compared to the RTX 2080: slower, more energy, same price -> no recommendation

Compared to the Vega 64: a bit faster, same energy, double price -> no recommendation

It's not a bad chip, it's just way too expensive. This might work for the gamer market, where 16GB VRAM is a very nice to have for future titles. But for hashcat that's irrelevant.
Since the new chip is only 70% the size of the old one, AMD should be able to produce the Radeon VII at the same price like the Vega 64 (considering the additional HBM modules).