Screw them. They're tactics in recent years have made sure I won't buy from them anytime soon. Best thing to do in such cases is to vote with your wallet.

They've been in this "Creative" position for over 15 years now, from dominating the ARB board to arrogant marketing deals to physx lockdowns to CUDA wrapper threats, etc. it's not a surprise this would be the next move

and frankly I still find modern card SLI to be a waste of money and power/resources (especially with gratuitously evolving APIs and even shorter driver deprecation life cycles). One single card is usually good enough to actually play games on, and SLI was only practical when it was a Voodoo2 doing it as it made up for a lack of performance tier variant of cards that weren't Quantum (just vram variants)

would still love to see their take on anti-piracy on the niche SLI market backfire though, they clearly didnt learn from Oculus 😁

AMD has already flopped.
Their R480 is around GTX980 performance levels max. So nowhere near the 1070/1080. You'd need two or more 480 cards to even get to the performance of a single 1070/1080.
Their high-end card is still many months away, and by the looks of the R480, it's not going to be competitive with nVidia's high-end (1070/1080 are midrange, they still have the P100 version in store, which they will probably call 1080Ti or Titan whatever. It will probably arrive at around the same time as AMD's Vega).
AMD is quite a ways behind currently, both on efficiency and on features (no conservative rasterization, no rasterizer ordered views, no SMP, slow tessellation etc). It's over.

Shouldn't DX12 Multi-GPU take care of these stuff? Why would you need to pay extra?

I think I'm going to push requests for OpenCL support to the devs of my (CUDA-exclusive) programs now. I really want to avoid having to buy NVIDIA at the moment.

@Scali
From what I see, AMD is trying to put a blow into the midrange market first. The rest is... wait why do I feel like I'm in one of those wccftech comment sections... ah nevermind... it's going nowhere.

Yes OTOH, often AMD would show up like a knight in shining armor for all these lofty ideals of being "open" but proper execution is what we all need. On the games side, I was hoping that DX12 would be AMD's stronghold, but looks like Async alone is far from being a wildcard. Godspeed Vega & Zen.

Anyway P100 goes to Tesla, 1080Ti won't be full version P100, probably not even the next Quadros/Titans, but who knows what they have in mind now, with all this "1080/1070 blows Titan" marketing.

Scali wrote:

AMD has already flopped. Their R480 is around GTX980 performance levels max. So nowhere near the 1070/1080. You'd need two or mo […]
Show full quote

AMD has already flopped.
Their R480 is around GTX980 performance levels max. So nowhere near the 1070/1080. You'd need two or more 480 cards to even get to the performance of a single 1070/1080.
Their high-end card is still many months away, and by the looks of the R480, it's not going to be competitive with nVidia's high-end (1070/1080 are midrange, they still have the P100 version in store, which they will probably call 1080Ti or Titan whatever. It will probably arrive at around the same time as AMD's Vega).
AMD is quite a ways behind currently, both on efficiency and on features (no conservative rasterization, no rasterizer ordered views, no SMP, slow tessellation etc). It's over.

But they aren't competing with the 1070/1080. They are aiming for the larger GPU market - the mid-range one. 1070/1080 cards are high-end, enthusiast level cards which belong to a smaller market, which they cant even properly supply because chip yields are still too small.

archsan wrote:

@Scali
From what I see, AMD is trying to put a blow into the midrange market first.

"Trying" being the operative word here. They are fighting from a disadvantage. Price is all they have. NVidia is already countering that by lowering the prices of their current products.
Once the 1050/1060 arrive, NVidia will have the upper hand again, since they have the technical advantage. They have the more advanced architecture, resulting in smaller chips, lower power consumption and cheaper-to-build cards. If AMD competes on price with those, it will be their downfall.

archsan wrote:

Yes OTOH, often AMD would show up like a knight in shining armor for all these lofty ideals of being "open" but proper execution is what we all need. On the games side, I was hoping that DX12 would be AMD's stronghold, but looks like Async alone is far from being a wildcard. Godspeed Vega & Zen.

AMD still doesn't implement various DX12 features, which both NVidia and Intel have supported for years now.
DX12 is anything but AMD's stronghold. It seems to be their downfall, because supporting the full API requires more of a GPU redesign than they have budget for.
To top it off, NVidia has also made a pre-emptive strike in the VR-market with their SMP, greatly improving performance. That's another thing AMD won't be able to answer (this is built as an extension of NVidia's PolyMorph technology, which is what has given them a considerable edge over AMD in tessellation for the past 5+ years. NVidia's architecture can handle geometry very efficiently, more specifically the 'inflation' of geometry that results from tessellation. SMP is very similar, where you input one set of geometry, and they expand it for up to 16 viewports simultaneously, distributed over two eyes for stereo rendering).

archsan wrote:

Anyway P100 goes to Tesla, 1080Ti won't be full version P100

What do you base this on?
In previous generations they always had a 'full' graphics card as well (okay, they artificially limited things like DP-performance as to avoid competition with Quadro/Tesla series, but for gaming that was irrelevant).
Bottom line is: there will be a larger/faster chip. Even a 'non full' P100 (or whatever they're going to call it) is considerably more powerful than the current GP104.

Munx wrote:

But they aren't competing with the 1070/1080.

Does that matter?
1070/1080 are the midrange tier of NVidia's Pascal architecture (nope, it's not high-end, 1080Ti will be high-end).
The low-end tiers will arrive soon, and will share the performance characteristics of 1070/1080. We can easily extrapolate how that will compare to Polaris.
That is what AMD will be competing with eventually. Likewise, we can extrapolate how Vega will perform, based on Polaris.
Anyway, they have 'flopped' in that they aren't going to release a chip that competes with the 1070/1080, given the topic of this discussion... If you are considering a 1070/1080 SLI setup, as the OP does, then RX480 is not going to be an alternative, since it's in a lower performance class.

What does all of this have to do with the topic we're discussing?

1070 is not quite a "high end", just like 970 wasn't. It might've been closer to the price/perf winner IF obtainable at the promised $379 price point. Definitely not at $500+.

@Scali
GP102 is what's widely reported (not officially confirmed of course) to be on 1080 Ti and Titan.
GP100 is already being used in Tesla variations (external proprietary, and internal PCIe version incoming), but Quadro lineup might use it later.

@F2bnp
such is the nature of complaint threads...

Anyway back to topic, why would you need SLI/CF if DX12 Multi-GPU works as intended? (okay, that's an IF)

archsan wrote:

1070 is not quite a "high end", just like 970 wasn't. It might've been closer to the price/perf winner IF obtainable at the promised $379 price point. Definitely not at $500+.

Well, given the fact that it matches/beats the performance of the Titan X that was $1000, it's still a great deal.
I suppose eventually the prices will drop to 970-ish levels, since it probably costs about as much to manufacture. But given that there is no competition, why would they?

archsan wrote:

@Scali
GP102 is what's widely reported (not officially confirmed of course) to be on 1080 Ti and Titan.
GP100 is already being used in Tesla variations (external proprietary, and internal PCIe version incoming), but Quadro lineup might use it later.

That's my point: we don't know exactly what the chip is going to be called yet, nor what its exact specs are, since NVidia has yet to make an official statement on that.
What we DO know however, is that GP104 is a 'small' chip, and they already have the GP100 in Tesla, so they can already make a 'high-end' Pascal.
This WILL be used in a graphics card in the not-so-distant future, in one form or another. So there will be a card that is a class faster/more high-end than the 1080. And it most probably will also use HBM2, like the Tesla GP100 does.
Whether that chip is a true GP100, a slightly limited GP100, or whether they call it GP102 or whatever, isn't relevant.

archsan wrote:

Anyway back to topic, why would you need SLI/CF if DX12 Multi-GPU works as intended? (okay, that's an IF)

That's a HUGE if atm, because afaik no DX12 game supports multi GPU at all yet. The fact that developers have to implement it themselves now, will probably mean that support and performance will be all over the place, so DX12 is probably the worst time ever to go for multi-GPU.

I'm just curious, Scali, have you bought/will you buy the 1080/1070?

Because I agree with most of what you have said about the NVIDIA/AMD situation, but still I can't (don't want to) justify the current inflated prices (especially for something that's not even that "high end" -- really it's just like seeing 680s or 980s and 970s in hindsight, that much is clear), and I don't want to support some of their close-natured practices (Gameworks etc).

Thing is I still need CUDA and will have to get one card for it soon. A bit of a dilemma thrown in.

archsan wrote:

I'm just curious, Scali, have you bought/will you buy the 1080/1070?

I already have a GTX970, so it's not like I'm really looking for a new videocard yet. I already have quite a capable DX12-card and a 4k monitor.
If I didn't have the GTX970, I probably would have bought the 1070. In fact, I was planning to sell the GTX970 to my brother, and get a 1070, because he wanted a new card.
But that idiot bought a new GTX970 (before the pricedrops).
I don't feel like bothering to put my card up on Ebay or whatever for some random buyer, so I'll just stick with the GTX970 for now. It does the job for me.

Right now it depends on how long my GTX970 still suits me, and how much the 1070 will drop in price in the coming months. Perhaps at some point it will arrive at a 'sweet spot' where I want to upgrade. Or perhaps I will just wait for the 1170 (or whatever they will call it). Difficult to say.

I never believed in SLI/CrossFire though.