All things Radeon VII / Vega 20

Jim1348
Jim1348
Joined: 19 Jan 06
Posts: 463
Credit: 257957147
RAC: 0

I wouldn't do it.  Unless you

I wouldn't do it.  Unless you need the double-precision (as on Milky Way), I would wait for Navi.

https://www.theverge.com/2019/2/11/18194190/amds-radeon-vii-is-a-hot-loud-powerful-answer-to-nvidia

ExtraTerrestrial Apes
ExtraTerrestria...
Joined: 10 Nov 04
Posts: 770
Credit: 534774336
RAC: 189278

With undervolting one can get

With undervolting one can get it down to ~250 W at a few % higher speed. That's at least the same power level as other flagship cards (which can also be eco-tuned), so should be manageable. And at Einstein cards are usually not running at their full gaming power consumption.

I would say: if the performance is there, it's power consumption may be justified. And you're right, Navi should appear somewhere in H2/2019. I would expect significant gaming improvements from it, but am not sure about computing. AMD has not revealed much about that, though, so I could be totally wrong.

MrS

Scanning for our furry friends since Jan 2002

Jim1348
Jim1348
Joined: 19 Jan 06
Posts: 463
Credit: 257957147
RAC: 0

Navi will use GDDR6 memory

Navi will use GDDR6 memory instead of HBM2, which will save a lot of power (and cost).  I don't know how much of a difference that will make on Einstein, but it may not be much of a loss.  But yes, under-volting the AMD chips helps.

PS - I find that I can overclock/undervolt my RX 570 a lot better on MilkyWay than on Einstein.  Along with the very good DP performance, that is where I would use the Vega VII, assuming they were willing to send out any work units.

 

mmonnin
mmonnin
Joined: 29 May 16
Posts: 290
Credit: 3212249019
RAC: 9371

Jim1348 wrote:Navi will use

Jim1348 wrote:

Navi will use GDDR6 memory instead of HBM2, which will save a lot of power (and cost).  I don't know how much of a difference that will make on Einstein, but it may not be much of a loss.  But yes, under-volting the AMD chips helps.

PS - I find that I can overclock/undervolt my RX 570 a lot better on MilkyWay than on Einstein.  Along with the very good DP performance, that is where I would use the Vega VII, assuming they were willing to send out any work units.

 

HBM is a hefty power reduction compared to GDDR5 with the trade off being its much more expensive. I'm not sure GDDR6 will save that much if any over HBM.

Millenium
Millenium
Joined: 8 Oct 14
Posts: 21
Credit: 33102476
RAC: 0

Jim1348 wrote:I wouldn't do

Jim1348 wrote:

I wouldn't do it.  Unless you need the double-precision (as on Milky Way), I would wait for Navi.

https://www.theverge.com/2019/2/11/18194190/amds-radeon-vii-is-a-hot-loud-powerful-answer-to-nvidia

Well I would not trust "the verge" on computer things... but hey, personal opinion.

Gavin
Gavin
Joined: 21 Sep 10
Posts: 191
Credit: 40635175321
RAC: 2822008

Jim1348 wrote:I wouldn't do

Jim1348 wrote:

I wouldn't do it.  Unless you need the double-precision (as on Milky Way), I would wait for Navi.

https://www.theverge.com/2019/2/11/18194190/amds-radeon-vii-is-a-hot-loud-powerful-answer-to-nvidia

Horses for courses.

Not sure what you are expecting from Navi but given that Radeon VII has just been released as the 'flagship' card from AMD it does rather suggest that when Navi does make an appearance (supposedly later this year) it will be in the form of a mid market set of gpu's, any Navi based high end kit will be at least 12 - 24 months away and will then probably still use HBM of some sort :-)

MarkJ
MarkJ
Joined: 28 Feb 08
Posts: 437
Credit: 137177862
RAC: 27154

I think it was the Linus Tech

I think it was the Linus Tech Tips review where he said the Vega VII came from their compute card which would explain the large memory and high FP64 numbers.

The wattage is a show stopper for me. Its way too high.

mmonnin
mmonnin
Joined: 29 May 16
Posts: 290
Credit: 3212249019
RAC: 9371

For me if I had one, Vega 7

For me if I had one, Vega 7 would be a MW/Collatz card to make use of its FP64. E@H has performed well with some older AMD high FP64 cards like 7970 and it's derivatives. Vega 7 should have ~3.5 the FP64 performance for less than 3.5x the power. If MW is a person's project then Vega 7 is a no brainer.

mesman21
mesman21
Joined: 30 Jan 07
Posts: 15
Credit: 3337606947
RAC: 0

The VII has arrived and it's

The VII has arrived and it's crunching away! Initial impressions are good, likely capable of 1.7M /day or better.

I installed it in a machine that previously had Vega 64's; so disregard all WU's dated prior to 17 Feb UTC.

https://einsteinathome.org/host/12661032

It's currently set to -20% power limit with otherwise default settings. I'm going to let it crunch as-is overnight, and I'll experiment with settings and power draw tomorrow.  Running 3 concurrent WU's on all my machines.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5840
Credit: 109034952427
RAC: 33940802

I had a quick look and could

I had a quick look and could easily identify the changeover.  The previous times immediately beforehand seemed to be around 870-900 seconds (they seemed a bit variable too) and with the new GPU that dropped to around 530 seconds.  On the basis of 530 secs and 3x concurrency, that translates to around 1.7M RAC so I agree with your assessment and therefore must be looking in the right place :-).  On the basis of 980 secs for a Vega 64 3x, the previous theoretical RAC would be around 910-920K.  I didn't add up and average a bunch of Vega results - I just took a quick guesstimate.

Even so, that's certainly a very healthy increase!! :-).  I wonder how quickly some more will start showing up here!  Thanks very much for making us all extremely jealous!!

 

Cheers,
Gary.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.