Folding Forum

Sounds interesting to us folder:

AMD wrote:and up to 62 percent higher performance in ... OpenCL ...compared to the AMD Radeon RX Vega 64

https://www.amd.com/en/press-releases/2 ... xceptional

yes but vega 64 was a complete power hog so unless we get 2080 performance or better nvidia will still be more efficient

I guess due to AMD new 7nm process the Radeon Vega II matches efficency with current Nvidia 14nm process in turing RTX 2080 GPUs. Sounds good to have a real alternative to Nvidia. (That is why Nvidia will also do 7nm shrink release in 2020)

Technically, half the die size means either half the power consumption, or nearly double the performance. Usually, it's a mix of both.
At 60% better performance, we can expect 40% lower power consumption than the Vega cards
New 7nm AMD cards might be interesting for folders, if those cards will be stable and are proven to be reliable. 2 things that get harder and harder to do, with smaller lithography.

Does anyone have the AMD Radeon VII or knows the PPD?

If gaming performance is any indication, it's slightly faster than an RTX 2070, and slightly slower than an RTX2080, so probably around 1.3-1.4M PPD would be my guess.
Then again, it has a lot of memory helping out their performance in some games, which FAH doesn't need, so it might actually fold slower than that.
AMD themselves said that their 7nm has only few 7nm nodes. Most of them are 14nm and some transistors of the GPU even larger than that; which is why it's still less efficient than the 12nm Nvidia cards.

I think I saw that Radeon VII has quite a wide amplitude, from 500k PPD to 1.5M depending on the number of atoms in the WUs ...

foldy wrote:Does anyone have the AMD Radeon VII or knows the PPD?

Ananad Tech tested the Single and Double Precision using FAHBench and it falls short of the 2070FE but ahead of the Vega64 by about the same amount in SP but beats even the 2080FE in DP (not that a lot of that is used in FAH)

The Overclockers.net GPU Performance Spreadsheet shows 1,646,140 points on a sample size of 1 so on this unknown WU somewhere between a 2080 and a 2080 Ti's averages but the result could have been cherry-picked and we wont know until they either have a bigger sample set so it can show up on the Projects tab.

GPUs are generally divided into two categories: Home and Professional. Both manufacturers assume that businesses are willing to pay a higher price if then boost the Double Precision performance and Home users are more attentive to price (and they generally use less or no DP). FAH's design was aimed at the home user and originally avoided using Double Precision. Early tests were made with the compiler flag that converted all SP operations to DP \. That slowed completion of the WUs quite significantly. but also avoided some scientific limitations if only SP was used. This led to a re-write the FAHCore to use DP as little as possible -- now referred to as "Mixed Precision".

Seems like some conflicting data on the VII.
I don't understand why their gaming performance is between the 2070 and the 2080, but their specs are a lot higher than a 2080.

The link below indicates that the Radeon VII, has twice the memory bandwidth of an RTX 2080, and works on a 4096 bit memory bus, vs 256 on the 2080.
It also uses the same GPU boost frequency as the 2080.
I can only speculate that these benefits, won't affect folding by much.

But the VII has about 30% more cores.
I wonder if this means at least 30% HIGHER PPD than a 2080 (more due to the bonus points)?
If that's the case, the VII should get close to 2M PPD or more...

It also has 30% higher transistor count, 13,2 Bn vs 10.6 on the 2080.

Their gaming performance benchmarks still show the 2080 has better gaming performance, which can indicate that there's an internal slow down (that could use optimization) on the VII.

If you multiply the core count with the frequency of the VII vs the Vega 64,
the VII gets you: 3840 x 1711 = 6.570.240
the 64 gets you: 4096 x 1525 = 6.246.400

Vega 64 is reported to get about 1M PPD, so VII should get a little over 1.1-1.2M PPD, what is in line with said above.
The problem is that the card has a TDP of over 300W. How much of this will be used for folding isn't very clear.
But since an RTX 2080 can do well over 1,2M PPD at a mere 125 Watts, will make this market quite tough for the VII to get into.


https://www.pcmag.com/compare/366379/am ... -gaming-ca

AMD Radeon VII (Vega 2) architecture is not efficient enough and will get superseded by AMD Navi GPUs which get released this summer.