SAPPHIRE Technology has announced the launch of the much-anticipated SAPPHIRE Radeon Vega 64 enthusiast graphics cards. The Vega architecture boasts significant improvements focused on maximizing the performance. Vega cards are designed for enthusiasts seeking top-of-the-chart framerates in games of today and tomorrow – in Ultra details and VR.
Looks like HBM2 won't budge at all, it is maxed out already.
https://sapphirenation.net/sapphire-radeon-vega/?utm_source=newsletter
Likely it is maxed out but using those Sapphire details to prove it is irrelevant as those are the reference cards re-branded, not custom AIB versions.Looks like HBM2 won't budge at all, it is maxed out already.
Well, here's to hoping that just like 290 we'll be able to pick them up for $200 6 months to a year after releaseNot surprising given the voltage that has been reported. Vega is the 290 all over again. Pushed to it's limit, hot, loud, guzzling power. Over a year late to market and still rushed out the door... I've said it before and I'll say it again, AMD better have used some secret sauce on Vega to make it mine like a champ because mining is it's only hope to sell anywhere near MSRP.
Anyone know/educated guess if the memory on Vega 56 can be overclocked to 1890mhz?
True, but I was hoping they would tweak it a bit, but, these are just vanilla reference cards.Likely it is maxed out but using those Sapphire details to prove it is irrelevant as those are the reference cards re-branded, not custom AIB versions.
Do you believe the (lack of) popularity of DX12 and Vulcan supports this argument?
It's impossible to say how much because that will vary case by case (or should I say frame by frame). Proper lods and middlewares like Umbra already help with reducing useless geometry in games so I don't think the gains are going to be huge in actual games.I have asked simple question, how much do you guys think GPU will be faster in situation when software implements Vega Features. I did not implied it will be 2 times faster.
Yes, AMD have said that in CURRENT slack of software 150W Vega has similar performance per watt that GTX 1080.
Does any of current slack of software use any of Vega features that have most meaningful impact on performance?
Doubling the Geometry registered each clock by GCN is solving biggest problem it had in geometry throughput. And then on top of it, you have a culling technique that supposedly can cull at least 50% of geometry not used in scenes.
About the last paragraph. What if Vega already beats GTX 1080 by some margin? What if you will add on top of it hardware features, and Primitive Shaders to software? Is it impossible for Vega to gain 40% of performance with it?
What do you think defines performance of GPUs, per clock? Its a serious question. Why do you think they had to lose performance per clock to gain higher clock speeds?It's impossible to say how much because that will vary case by case (or should I say frame by frame). Also, who knows how much perf per clock they lost in order to gain higher clocks. Maybe the new stuff just help to gain the lost speed. To me it looks like they tried to achieve higher clocks but hit the wall (combination of GCN and manufacturing process problems).
I'm pretty certain that AMD has done the predictions already. They would have been way more bullish if they except Vega to beat big Pascal. If that ever happens, Volta will be out and it all is irrelevant. Also how come that there has never been talks about "wait for GPU feature implementations" if the GPU performance is lacking at launch? Massive majority of devs are not going bother. Hardware that requires extra work is just a bad idea.
Vega has a lot of compute power but that doesn't translate to massive gaming performance. It feels like they are making computing cards for miners and not graphics cards for gamers. NVIDIA made correct corrections after Fermi. AMD should have done the same after Hawaii but it looks like they have problems doing that.
Looks like HBM2 won't budge at all, it is maxed out already.
Well, it was already mentioned by this guy...They might be able to get better results next year if they have access to HBM2 that's actually running at the speeds it was originally supposed to. That, combined with a respin (process tweaks, like on 2nd gen Polaris), might provide marginally better performance and perf/watt, but it still will not be particularly competitive with Pascal for gaming.
Industry sources...
HBM2 is at version 3 currently, "Perfected Yields", version 4 allows for half-channel => 512-bit, and 2.4 GHz op which is 307.2 gigabyte per sec or 153.6 gigabyte per sec,"Low Cost".
Vega 12 w/ 512-bit HBM2... yum.
If losing tiny bit performance pre clock means much higher clocks then it's a worthwhile trade-off. Especially if you can gain back that with new features (that would mean no PPC decrease in practice). Why I think that might be the case is at least so far Vega has performed pretty much like higher clocked Fiji. Though, performance per clock doesn't matter but the actual outcome. There are numerous ways to do things.What do you think defines performance of GPUs, per clock? Its a serious question. Why do you think they had to lose performance per clock to gain higher clock speeds?
You have not answered my question.If losing tiny bit performance pre clock means much higher clocks then it's a worthwhile trade-off. Especially if you can gain back that with new features (that would mean no PPC decrease in practice). Why I think that might be the case is at least so far Vega has performed pretty much like higher clocked Fiji. Though, performance per clock doesn't matter but the actual outcome. There are numerous ways to do things.
In general it tends to be low PPC -> high frequency. High PPC -> low frequency. It's usually pretty darn hard to get both up. With Pascal NVIDIA managed to get the clock speed up a lot. AMD didn't succeed as well as NVIDIA. Also since the core config is same between Vega and Fiji (4096:256:64) it means that the main performance increase should come from higher clock speed and the modifications they made since Fiji. They hit the power and frequency wall, thus the results are what they are.
I did answer why I though they could have done it. In order to get the clocks higher. If losing a bit of performance per clock means much higher clock speeds and higher performance then that's one design reason.You have not answered my question.
What do you think affects performance per clock in GPUs? Why do you think they had to lose performance per clock in order to gain clock speeds?
You have answered why, but you have not answered what is defining performance per clock for GPUs.I did answer why I though they could have done it. In order to get the clocks higher. If losing a bit of performance per clock means much higher clock speeds and higher performance then that's one design reason.
Besides, what kind of questions is that? If we take GTX 1070 and GTX 1080 then the answer would be: GTX 1080 has more hardware so that it can do more stuff at the same clock speed (on high level).
Shorter, wider... Longer, narrower... Pretty much the same reason why Athlon 64 smoked Pentium 4.
It's gonna be even more than that, performance (specifically in DX11) is going to be an Epyc Vega FE disaster. I won't say more than that.Vega is a unmitigated disaster in terms of perf/watt and perf/sq mm.
Do we know what time the Vegas will be available for sale on Monday?
Nice!
IIRC, Monday is review embargo lift. Sale is on the 24th?
Nope.... 14th everything is let loose.IIRC, Monday is review embargo lift. Sale is on the 24th?
Check bottom of page.... https://sapphirenation.net/sapphire-radeon-vega/?utm_source=newsletterAvailability
The new Vega 64 based SAPPHIRE graphics cards will be available to buy from August 14th.
For full product information, head to the SAPPHIRE Technology website.
Nope.... 14th everything is let loose.
Check bottom of page.... https://sapphirenation.net/sapphire-radeon-vega/?utm_source=newsletter
Well not everything. Vega 64 goes on sale on the 14th. Another leak I read somewhere else said the 28th for Vega 56 to go on sale.