AMD Vega 20 3DMark11 benchmark surfaces

I bet miners will appreciate that memory speed with their wallets

 

There must be a new 7 nm Vega gaming card, because even without much improvements into the design, it would be cheaper for AMD to manufacture Vega chips.

 

Is that P score?

 

Hmm 3% combined increase. Is it really worth making this product? Isn't an overclocked Vega 10 card in this area of performance already? I'm confused why AMD is putting all this effort in to an blah Gpu instead of using their resources for a new series? Is this their attempt at a Vega TI?

 

You could say the same about Zen+. It's optimization, hopefully for same price. Maybe to safe money on production and give consumers more options.

 

Yes I get Zen+ 10 percent increase plus in performance. But you couldn't push Zen to the Zen+ level of performance along with memory performance increase. This Vega 20 seems really redundant with performance not better at base level than an overclocked Vega 10. And who knows what the price will be.

 

Well, it's good for marketing. Release a new GPU, give a good price for performance and you got new customers. GPU market could use some stimulation.

 

Clock speeds are set to 1Ghz hmm.

 

Depends on die size and power. If Vega 20 is only ~350mm2 and ~180w for similar performance then I think it's worth it. AMD doesn't have the money to continuously spit out massive architecture changes anymore - they are probably dumping the majority of their GPU R&D funds into Navi. I think this release will be more akin to polaris, trying to reach the larger market with a product that comes in at a good price/performance ratio and giving up some of the higher end stuff to Nvidia again.

 

Is it me or is that saying 32GB RAM? And 1GHz core clock, engineering sample?

 

Mid-range GPUs are where the money is made. I hope they have the volume.
NAVI should be interesting.

 

I thought it was for compute not gaming....well at least till they put rx in front of it
still not much to see here

 

If I'm reading that screenshot correctly it's actually a performing a combined 3% worse but at ~2/3 the clockspeed of the Vega 64.

 

It performs around the same as a overclocked Vega GPU yeah but it seems to be primarily intended as a work GPU for deep learning and compute work and the benchmark might not report the speeds correctly but hardware wise it appears to be very similar to Vega10 but with a die shrink and possibly tweaks to voltage and clock speeds as a result.

For a desktop card for gaming if that's all there is then 5 - 10% gains on average perhaps similar to overclocking the current models unless there's some other tweaks here to the architecture which we'll see I guess, on the positive side of things it might be able to use even less voltage which would be pretty excellent. *

* Seriously AMD often goes for higher voltages and clock speeds but Vega and the third party models in particular are on a whole new level though they are also very interesting under-volt and over-clock achievers.
(Fury doing -50mv to -96mv on some models was pretty good, Vega going for -200mv is really impressive though it varies from GPU to GPU how well it can clock or under-volt of course, same as always.)

Up to 1.2v core voltage for some models but it can be reduced as far as 1.0v or even 0.95v while still achieving it's core clock speeds or higher (1550+ Mhz under full load.) though Wattman doesn't allow for altering HBM2 voltages directly but bios flashing takes it from 1.2v to 1.35v I think it is (Compatible Vega64 air bios.) but then it's chance if you end up with the Samsung HBM2 modules which can reach 1200Mhz or if it's Hynix which vary between 1000 to 1100 I think it was for these at best, have to do more reading and there's not too many reviews for these custom Vega GPU's either and the new Nano PCB variants of them. (Which are doing really well but heat is a concern.)

 

Still won't compete with the 1080Ti for the performance crown of course if these numbers hold up, they're around 20 - 25% faster if not more I think (And then overclocks can take them even higher.) and if Vega 20 here improves memory performance even further and increase compute performance too then mining is going to a prime target for these so bye to any sane pricing for quite some time assuming mining doesn't somehow just crash which well I don't think that's happening any time soon.

EDIT: But we'll see I suppose, if AMD announces desktop variants of Vega20 for late 2018 or early 2019 before Navi those might have some tweaks taking them a bit closer to the competition.
(Still has pricing a core concern though, depending on how the situation unfolds.)

1600 to 1800Mhz core clock and 1200Mhz HBM2 clock wouldn't be bad and then well voltage is going to be AMD ++ as always I suppose, should put them a little bit above Vega 10 even for the better overclock results but we'll have to see what AMD announces I suppose, just speculating a bit here.

Higher quality binned HBM2 modules wouldn't be cheap either whether it's 16 GB or full on 32 for workstation which I guess might not see desktop card usage, I think AMD planned 1000 Mhz for Vega but then dropped down to 800 with overclocking now possible to push it a little bit higher so if they can get 1000 for Vega 20 that's a pretty good boost too.

Or 1200 depending on how accurate the results are for the above benchmark which might differ between engineer samples and the final product.
(4096 bus width instead of 2048 could do something too but I don't think this is a bottleneck area in itself.)

 

I don't think AMD needs to pump out a brand new architecture. There is plenty good about Vega, it just needs refinement. But, if this is really just nothing more than a die shrink with better memory, this is looking very bleak. AMD had plenty of time and money to fine-tune Vega. The miners were a saving grace to them, and so far it looks like they're not taking advantage of that. They can't afford making another product that's mediocre to the gaming community (I personally think Vega is great, for non-gaming purposes).

The way I see it, whatever Vega 20 comes out to be would still be less than what people were expecting Vega 10 should've been. I sincerely hope Vega 20 is more interesting than we think.

 

nice 2x vega 10 perf!

 

Nice, so a Vega 20 @1GHz is almost as fast as a Vega 10 @+1.6GHz and using the power of a Polaris GPU.
Bring it on!

 

Some progress , nonetheless good news !