Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
AMD introduces Radeon VII "7" new GPU @ $700
AmazingAvery
Age of Conan AdvocateMember UncommonPosts: 7,188
What is nice is that it support PCI-E 4.0 and the memory is clocked at 2ghz.
It really looks like a $700 card that barely matches an overclocked GTX 1080 TI while using more power despite being on 7nm, HBM and DX 12 advantages.
Not sure if the $/perf is competitive with Nvidia on the face of it would be no for me.
EDIT - No PCI 4 or Infinity Link support
Past that, as this is still a Vega architecture product, it’s the Vega we all know and love. There are no new graphical features here, so even if AMD has opted to shy away from putting Vega in the name of the product, it’s going to be comparable to those parts as far as gaming is concerned. The Vega 20 GPU does bring new compute features – particularly much higher FP64 compute throughput and new low-precision modes well-suited for neural network inferencing – but these features aren’t something consumers are likely to use. Past that, AMD will be employing some mild product segmentation here to avoid having the Radeon VII cannibalize the MI50 – the Radeon VII does not get PCIe 4.0 support, nor does it get Infinity Link support –
The other wildcard for the moment is TDP. The MI50 is rated for 300W, and while AMD’s event did not announce a TDP for the card, I fully expect AMD is running the Radeon VII just as hard here, if not a bit harder. Make no mistake: AMD is still having to go well outside the sweet spot on their voltage/frequency curve to hit these high clockspeeds, so AMD isn’t even trying to win the efficiency race.
AMD’s own proposition is actually fairly modest; with a $699 price tag they’re launching at the same price as the RTX 2080, over four months after the RTX 2080. They are presumably not going to be able to match NVIDIA’s energy efficiency, and they won’t have feature parity since AMD doesn’t (yet) have its own DirectX Raytracing (DXR) implementation.
But what AMD does have, besides an at least competitive price and presumably competitive performance in today’s games, is a VRAM advantage. Whereas NVIDIA didn’t increase their VRAM amounts between generations, AMD is for this half-generation card, giving them 16GB of VRAM to RTX 2080’s 8GB.
Comments
AMD believes that they can make more money selling it as a 300 W card that is competitive with a stock RTX 2080 than as a 150 W card that is competitive with a stock RTX 2070. And they're probably right about that. It's likely best to think of it as somewhere between an RTX 2070 and an RTX 2080, but merely with a more aggressive stock clock speed. If you overclocked an RTX 2070 far enough to match the stock Radeon VII performance, it would probably seem like quite a power hog, too.
That said, it probably isn't going to be that compelling of a product at $700 unless you really value 16 GB of memory. I think this is more of a stopgap until Navi, and a way to make them look more competitive on high end GPUs in the meantime, not a product that AMD really wants to sell a ton of.
If you had $600+ to spend on a video card - why would you chose the VII over 2080? I can't think of a single compelling real-world reason right now (especially with nVidia/Freesync), except to say FU nVidia.
More VRAM is better than less VRAM, but it's not like the 2080 routinely runs out of VRAM under normal conditions (sure, if you try to turn on RT at 4K, you pinch it, but good luck playing at that anyway).
AMD can't win the performance crown (they still don't appear to be close to a Ti/Titan), but they are going to chase a halo product price point. I'm sure the card costs enough that they needed to get a high price point, with that much HBM2 and all - but that isn't the same thing as the card making sense for the market.
I don't know if you could couple Vega with GDDR - it may be something with the architecture that is highly attuned to HBM requirements. HBM is cool, and it has some very good characteristics, but Vega in general doesn't exactly highlight any of them except the added cost.
I agree with JH - it's underwhelming. It may have made more sense if there were product tiers that also fell under here at lower price points to flesh out the gaps between this and RX590... but I didn't see any of that. Maybe some of the lower tiers could be more competitive, but if they are still on HBM, I don't know that we will see that.
The A12 is still the most impressive thing on 7nm, and likely will be until we finally see Ryzen 3 finally ship.
For compute in the Radeon Instinct MI60, it could be a very interesting card if priced properly. If it can offer 80% of the performance of a Tesla V100 for 20% of the price tag, that could make for a very interesting proposition. And that "20% of the price tag" would still mean about $2000 per card, which is a lot of money if you can sell very many of them. For applications where the bottleneck is global memory bandwidth, it's likely to beat a Tesla V100 outright.
I don't know how AMD will price their Radeon Instinct cards. But Nvidia charges so much for their Tesla cards that there's a huge range where AMD can make a ton of money per card and still offer a compelling price/performance ratio. That's the real target of the GPU, not Radeon cards.
The Radeon version might also exist as a way to sell GPUs where something needed only for compute was defective. You can cripple double-precision compute or PCI Express 4.0 or some other things in a Radeon card and it's fine, which might beat throwing the GPU in the garbage because they can't sell it as a compute card. You don't have to offer a compelling price/performance ratio to sell a relative handful of the parts to fanboys.
A new $700 card to compete with another $750 card. Ok. I wasn't excited about the first, I'm still not excited the second time around. And now there is this huge gap... you go from RX590, which is based on Polaris, which has been around for 2 and a half years already... through some first gen Vega parts which are also older and haven't had the best availability or penetration, until you get to this new card...
Now AMD has had high priced cards before - the water cooled Vega 64 wasn't exactly cheap - but people could kind of accept that: the AIO unit could command a premium, there was a lower cost air cooled card, there was a lower cost Vega 56, and there were price points all up and down the chart.
VII doesn't exactly present that. I think Quiz is correct when he states this is the stopgap until Navi - AMD didn't want to let RTX go entirely unchallenged, and let's face it, the 2060 just pushed all the first gen Vega out the window.
Great that AMD is catching up in performance. But fact is they still haven't caught up, and the VII feels half-assed.
And the worst part is, they wasted the VII name when they could have coupled that with the Re-Release of FFVII PC edition (in 24 years when S/E finally releases it).
Rumors say that Navi will launch around the middle of this year. AMD has said that Zen 2 will launch around the middle of this year, so it's certainly plausible that Navi will launch around then, too.
I think that part of the point of Vega 20 is to get a moderately big die out early on 7 nm to test out the process node. See what breaks or otherwise doesn't work properly when they get dies back around early to mid 2018, and then pour that knowledge into making Zen 2 and Navi work better. It's best to do that on an old, well-understood architecture so that anything that breaks isn't a problem with your architecture.
If Navi is the new architecture rather than yet another GCN derivative, then AMD really needs for Navi to do well. They don't need for Vega 20 to do well, or at least, not at graphics. (For compute, if global memory bandwidth or PCI Express is your bottleneck, Vega 20 is likely to be the fastest GPU ever made.) Similarly, AMD really needs Zen 2 to do well, especially in servers. So if you're going to make mistakes because of an early process node, it's better to make them on a chip that isn't that important.
If we start to see 7mm Polaris respins instead of Navi the the end of AMD graphics is near.
Below that, right now AMD is competitive below that. But that's only Pascal versus Vega/Polaris right now, and we know that nVidia is rapidly moving past Pascal.
Now, Turing doesn't offer a lot above and beyond Pascal aside from RTX, and there is every reason to believe that lower tier products won't include the Tensor cores that comprise Raytracing. There isn't much reason to believe they should exist on a 2060 or 2070, and some may argue their need to exist at all.
But apart from that, regardless of if we see Turing-based lower tier cards coming, or Pascal-refresh cards coming from nVidia, they are coming. And we know nVidia still has Volta in their back pocket and it exists and ships in the pro space - there hasn't been any reason to bring that to the consumer GPU market yet, but there is no reason they couldn't if they were pressed.
So for AMD to come back with a 600 series Polaris Re-Refresh (keep in mind the 500 series is already a refresh of the 400 series), it wouldn't bode well for their tech lineup. Polaris competes in today's environment, but even though they have a price/performance edge in this marketspace, nVidia still sells plenty of 1060s and 1050s on brand recognition alone.
AMD does need Navi, badly, because you can bet that nVidia isn't going to let up, and doesn't have to let up, and Intel is threatening to get into the race as well very soon. And this is the only space on the PC side where AMD is able to compete - if they can't maintain their advantage, they can't afford to erode marketshare in any market and remain a competitor in the PC space.
I do believe Navi is around the corner - there's every reason to believe it will be in the next gen console APU. But how far around the corner is the question. You know nVidia has two generation advances in their back pocket (Volta/Turing) they could pull from if they chose to, and who knows what else when nV does finally make the move to match process nodes.
AMD is clearly setting the pace here. nVidia is happy to put just enough tech to stay a step ahead, and not release anything too much farther forward. I believe nVidia could release something a lot faster that Turing if they chose to, but for right now it only makes business sense to just stay one step ahead, and milk it for all it's worth. It's what's worked for Intel for a long time now, after all.