Will AMD release any high performance GPU's in Q2 or Q3?

[chronia]

Not sure they would care too much if the cards are selling, AIBs will want in on that action, if they are making a profit and a quick consistent one at that right now then doesn't matter, also AMD pretty much have contracts with MS, Sony and Apple when it comes to non data centre GPU tasks, I don't think they need to worry.

They hardly had cards though, pretty consistent rumours are that for example in europe they had <10k cards for launch and another <10k cards coming for the RX 56 launch, not really hard to sell out if you don't have product. Only countries that had any stock allotted in the EU seem to be UK/FR/GER/NO/ESP/PL/RU/GR with the highest named number being 800 for the whole of Germany.

Contracts can be nice, but you need to know the margins though, and usually margins on the products in such contracts aren't that great. surely the volume helps but generally those contracts aren't the biggest money makers profit wise.

[Deleted]

They hardly had cards though, pretty consistent rumours are that for example in europe they had <10k cards for launch and another <10k cards coming for the RX 56 launch, not really hard to sell out if you don't have product. Only countries that had any stock allotted seem to be UK/FR/GER/NO/ESP/PL/RU/GR with the highest named number being 800 for the whole of Germany.

Contracts can be nice, but you need to know the margins though, and usually margins on the products in such contracts aren't that great. surely the volume helps but generally those contracts aren't the biggest money makers profit wise.

Those are the consumer cards, most of the stock for the Vega chip has gone to the professional market such as the instinct line and the Imac pro, Apple can pay a pretty penny, I'm pretty sure AMD don't have to worry about overall Vega sales, if the enterprise and professional market has bought them already then AMD can just give the scraps to gamers which looking at the recent products they have been pushing, this makes sense.

Does mean the gaming space has become boring on the GPU front from AMD though, would of been nice to see something more fine tuned, mean Polaris is still a great card, just needs to be upscaled and be fine, Vega has a crap ton of compute features thats not found in Polaris.

[Deleted]

Its sold out, not sure if you can call that DOA.

Because of low supply...

Some fanboys will buy whatever AMD offers. It wont work for the mass market.

Also dont forget: Intel and Nvidia are making billions of profit. While AMD has to sell them too cheap. AMDs cards are good for professional use like mining - and sadly not gaming.

[Evildeffy]

If 800 for launch numbers for Germany is accurate... it's more stock than the GTX 1080/1070 had when they launched... a year ago.

I can see Vega 56 being a success when 3rd party designs come out but reference... meh.
The whole price thing is also pretty shit though... I mean I know nVidia did/does the same but still a dick move no-one expected from AMD.

It seems to do well for compute side of things but meh ... not exactly what I was hoping for in terms of performance competition.

Ah well.

[Sorshen]

Its sold out, not sure if you can call that DOA.

Probably cause miners pre-ordered it based on rumors.

Again, V56 might be fine with custom cooling and voltage tweaks.

[Zenny]

Difference in process execution, I specifically mentioned latency rather than throughput.
AMD has more parallel hardware and like I said before, which you seem to be confusing with gaming performance, in compute terms AMD is, and almost has been, stronger if not for the lack of eco system.

AMD generally has higher compute performance, but that's got dick all to do with how parallel the hardware is. AMD has more parallel hardware is also nonsense, it has Instruction Hardware schedulers which is nothing to do with parallel computing or execution. It certainly allows the GPU to be fed multiple instruction queues via hardware, but Nvidia can do the exact same thing with it's driver.

And the compute crown has been Nvidia's for the past 1.5 years, and if you count overclocking, it still is. Overclocked 1080ti is around 14.6 TFLOPs while a Vega 64 at 1712MHz (highest I've seen) is 13.9 TFLOP.

Again compute != gaming performance, even though in Vulkan AMD cards tend to outclass nVidia quite handily ... the only problem with this is we only have 1 Vulkan game out so it's hard to "compare".

If you compare equal price points then Nvidia is slightly behind in Doom. If you compare equivalent TFLOP rating, then Nvidia still crushes AMD in Doom. We also have several games that utilize Vulkan on the market, Dota, Talos Project and Doom is the three I can think of.

That said I never stated nVidia didn't optimize their software scheduling, they have to a very good point in fact, I've only mentioned the differences along with Pros and Cons... not "ZOMG IT'S SUPERIOR!!111oneoneone".

You originally stated that Nvidia is a serial GPU:

And yet Pascal can still only throw everything through it's pipelines serially.

I'm pointing out that Nvidia GPU's are not anymore serial then AMD. Even if Instruction scheduling is done in the driver.

First part:
It does when the entire GPU uArch is designed and built around this fact and their competition is not.

Based on what exactly? Nvidia can do parallel compute, execution, you name it. Can you actually give any shred of evidence that it does not? Actual sourced evidence from Nvidia or some other reliable source? Yes, as I've pointed out Nvidia uses software scheduling in part, but that does not prevent parallel queue submissions in DX12. Hell if you need evidence of this check the Draw Call API Overhead test by 3D Mark, Nvidia cards show remarkable boosts in DX12/Vulkan over DX11.

http://www.anandtech.com/show/11223/...k-api-overhead

Oh look, Nvidia getting a massive boost from multiple parallel threads in DX12/Vulkan. AMD gets a even larger boost, but that's due to the dismal DX11 performance they have.

Second part:
Not really as the Warp Schedulers still have to receive commands from the driver in order to work, calling it a hybrid approach is... a little too friendly.

Warp Schedulers schedule what work each SM cluster is doing, which is done in hardware so claiming Nvidia is using pure software scheduling when the bulk of the work is done by hardware scheduler is a falsehood.

[Deleted]

Probably cause miners pre-ordered it based on rumors.

Very likely though DoA would emphasise a card that won't sell, but it has regardless of who the buyers are, just not the performance people wanted from it in gaming.

[Life-Binder]

I mean I know nVidia did/does the same

when did Nvidia straight up lie about actual MSRP prices to its own retailers and deliberately mislead all the reviewers ? (assuming that wccf link is true)


if anything FE Pascal was the opposite, they straight up priced it higher, kept it there and it got them some flak in reviews .. ppl could get FE or wait for cheaper entry level aftermarket cards like ~EVGA ACX 3.0

[SLSAMG]

Still nothing to compete or beat the 1080ti as expected. Guess I won't be selling my 1080ti anytime soon.

[chronia]

Those are the consumer cards, most of the stock for the Vega chip has gone to the professional market such as the instinct line and the Imac pro, Apple can pay a pretty penny, I'm pretty sure AMD don't have to worry about overall Vega sales, if the enterprise and professional market has bought them already then AMD can just give the scraps to gamers which looking at the recent products they have been pushing, this makes sense.

Does mean the gaming space has become boring on the GPU front from AMD though, would of been nice to see something more fine tuned, mean Polaris is still a great card, just needs to be upscaled and be fine, Vega has a crap ton of compute features thats not found in Polaris.

Surely Apple can pay a pretty penny, question is will they. Like i said normally contracts like these are very good for volume, but not per se for margin. Now AMD won't loose on a contract like that (atleast i don't hope so) but i doubt the margins are super great seeing Apple is known to negotiate pretty hard on pricing so they don't hurt their own bottom line which is what is important for Apple.

And yes, AMD should focus on the profitable enterprise and professional markets as they need market share and big profits, they are in the race again or coming in the race again on alot of fronts with Vega based products and Ryzen / TR / Epyc but now they also need to make sure they can stay in the race, and for that they will probably need much higher R&D budgets than they have now.

[Evildeffy]

AMD generally has higher compute performance, but that's got dick all to do with how parallel the hardware is. AMD has more parallel hardware is also nonsense, it has Instruction Hardware schedulers which is nothing to do with parallel computing or execution. It certainly allows the GPU to be fed multiple instruction queues via hardware, but Nvidia can do the exact same thing with it's driver.

Really? Nothing to do with hardware? Well that's weird since it's always the case with such.
But regardless this has been said many times in the past on this forum, on terms of Hawaii and Maxwell it was 8 graphics and compute pipelines including 8 command queues (AMD) vs. 1 command queue and 1 compute pipeline with 31 graphics pipelines for nVidia.
Quite a big difference, dive a little into history as that'll tell you more than I can re-hash at the minute.

Sufficed to say that engineers also stated that AMD's hardware is "more parallel" than nVidia.

And the compute crown has been Nvidia's for the past 1.5 years, and if you count overclocking, it still is. Overclocked 1080ti is around 14.6 TFLOPs while a Vega 64 at 1712MHz (highest I've seen) is 13.9 TFLOP.

There are already some reviews out (mentioned prior) where the RX Vega is already beating the 1080Ti/Titan X at these compute side of things, I'd say they still have that lead... Steve Burke from Gamer's Nexus' Vega 56 review states the same thing in his review.

If you compare equal price points then Nvidia is slightly behind in Doom. If you compare equivalent TFLOP rating, then Nvidia still crushes AMD in Doom. We also have several games that utilize Vulkan on the market, Dota, Talos Project and Doom is the three I can think of.

DotA and Talos Project have Vulkan tacked on as an afterthought ... DOOM (2016) had it developed from that engine, we know the difference for those engines.. DX12 is a prime example.
Don't pick stuff that is tacked on after, it's useless as they can't really drive it, hence why I picked DOOM only.
Otherwise I should adhere to this list: https://en.wikipedia.org/wiki/List_o...Vulkan_support

You originally stated that Nvidia is a serial GPU:

Correct, by that I meant the aforementioned single pipeline for commands and processing for execution, that hasn't changed... that is nVidia's hardware.
AMD has 8 in Hawaii (R9 290(X) and R9 390(X), unsure of Polaris and deffo unsure of RX Vega) that are deffo controlled and done in parallel where it can ONLY be done serially in Maxwell/Pascal.

I'm pointing out that Nvidia GPU's are not anymore serial then AMD. Even if Instruction scheduling is done in the driver.

They deffo are, they mightn't suck at their driver scheduling, but this stands.

Based on what exactly? Nvidia can do parallel compute, execution, you name it. Can you actually give any shred of evidence that it does not? Actual sourced evidence from Nvidia or some other reliable source? Yes, as I've pointed out Nvidia uses software scheduling in part, but that does not prevent parallel queue submissions in DX12. Hell if you need evidence of this check the Draw Call API Overhead test by 3D Mark, Nvidia cards show remarkable boosts in DX12/Vulkan over DX11.

http://www.anandtech.com/show/11223/...k-api-overhead

Oh look, Nvidia getting a massive boost from multiple parallel threads in DX12/Vulkan. AMD gets a even larger boost, but that's due to the dismal DX11 performance they have.

You think this has nothing to do with the fact of how their hardware is designed?
Why is it you think Maxwell showed regression in DX12/VULKAN? If they hadn't input the mid process interrupt they showcased in Pascal launch event they'd still have regression, because of their serial pipeline.
This example was showcased here: Clicky for Picture!

Your statement is saying that AMD's DX11 performance is purely based on drivers at this rate where their own Architecture for parallelization is getting in the way.

Your evidence is all in the history of this forum... perhaps @Artorius can give you a link to it since he's paying attention, otherwise dive into the history of my posts and find it somewhere, it's all in there.

Warp Schedulers schedule what work each SM cluster is doing, which is done in hardware so claiming Nvidia is using pure software scheduling when the bulk of the work is done by hardware scheduler is a falsehood.

See this is where we disagree, you claim bulk is fully parallel, I claim having a driver through a single pipeline telling the card what to do isn't really parallel but at the same time is.

Let me ask it of you this way:
Are the Warp Schedulers as nVidia refers to it fully independent? Or are they to be driven by the drivers?
Are the similar features in AMD cards driven by drivers or are they driven by the HWS?

There's a big difference and I didn't state nVidia sucked for it... far from it, they've done well, but having a driver middleman is not 100% parallelization.

[Lathais]

Oh look, Nvidia getting a massive boost from multiple parallel threads in DX12/Vulkan. AMD gets a even larger boost, but that's due to the dismal DX11 performance they have.

and you don't think that the reason that AMD gets a bigger boost is because doing it with the hardware scheduler is more efficient? So that does not prove that it IS the hardware that is making the difference.

No one is saying nvidias GPUs can not do this, just that being done by software means that the hardware is less parallel. Yes, the SOFTWARE can handle the parallel instructions, but the hardware can not. With AMD, the hardware CAN handle it because it's hardware is parallel and nvidia's is not. nvidia emulates parallel hardware with software that is less efficient.

[Evildeffy]

when did Nvidia straight up lie about actual MSRP prices to its own retailers and deliberately mislead all the reviewers ? (assuming that wccf link is true)

if anything FE Pascal was the opposite, they straight up priced it higher, kept it there and it got them some flak in reviews .. ppl could get FE or wait for cheaper entry level aftermarket cards like ~EVGA ACX 3.0

nVidia has done so by claiming MSRP to be cheaper vs. the MSRP but it never has been by any manufacturer unless they rebuilt the full reference design with an even crappier cooler, in fact the tech community and reviewers were in an uproar simply because they know the FE prices would be more held to "MSRP" prices than nVidia's own MSRP rating.

That's what I mean with did/does the same, not change the prices later as a hidden trick... that's a dick move that is AMD's own.

Also will you learn to quote a person properly already ... Goddamnit man.

[Thunderball]

and you don't think that the reason that AMD gets a bigger boost is because doing it with the hardware scheduler is more efficient? So that does not prove that it IS the hardware that is making the difference.

No one is saying nvidias GPUs can not do this, just that being done by software means that the hardware is less parallel. Yes, the SOFTWARE can handle the parallel instructions, but the hardware can not. With AMD, the hardware CAN handle it because it's hardware is parallel and nvidia's is not. nvidia emulates parallel hardware with software that is less efficient.

It is, but 1) you need software designed specifically for it; 2) It's not exactly free, those schedulers will likely drive you maximum clock speeds down, increase die size>production costs, heat dissipation and power consumption. Again, Nvidia's hardware is as parallel as AMD's, AMD just makes "wider" chips in general, which is again, not free at all.

- - - Updated - - -

nVidia has done so by claiming MSRP to be cheaper vs. the MSRP but it never has been by any manufacturer unless they rebuilt the full reference design with an even crappier cooler, in fact the tech community and reviewers were in an uproar simply because they know the FE prices would be more held to "MSRP" prices than nVidia's own MSRP rating.

That's what I mean with did/does the same, not change the prices later as a hidden trick... that's a dick move that is AMD's own.

Also will you learn to quote a person properly already ... Goddamnit man.

So you're saying that there havent been any Pascal cards cheaper than FE? If that's the case that's a blatant lie.

[Evildeffy]

So you're saying that there havent been any Pascal cards cheaper than FE? If that's the case that's a blatant lie.

I should have phrased that differently so everyone can understand:

nVidia has done so by claiming MSRP to be cheaper vs. the MSRP but it never has been by any manufacturer unless they rebuilt the full reference design with an even crappier cooler, in fact the tech community and reviewers were in an uproar simply because they knew the FE prices would be held closer to "MSRP" prices than nVidia's own MSRP value of AIBs, in essence from official AIB MSRP to nVidia MSRP for FE.

That's what I mean with did/does the same, not change the prices later as a hidden trick... that's a dick move that is AMD's own.

Also will you learn to quote a person properly already ... Goddamnit man.

Happy now?

[Thunderball]

Happy now?

Not really, FE is FE, AIB cards are AIB cards, FE inherently retails for more, AIB cards retail depending on their R&D and production costs (they dont have to stick to GPU model MSRP). That absolutely still leaves board partners some room to make their cards cheaper than FE.

With Vega I'm not seeing that at all.

[Lathais]

Not really, FE is FE, AIB cards are AIB cards, FE inherently retails for more, AIB cards retail depending on their R&D and production costs (they dont have to stick to GPU model MSRP). That absolutely still leaves board partners some room to make their cards cheaper than FE.

With Vega I'm not seeing that at all.

The problem is, the FE was nothing more than a glorified reference cooler, which historically have been less expensive than AIBs. Of course, we won't know for quite some time, but perhaps that's what AMD was attempting to do here, but then still give early adopters a better price? I dunno, it kinda sounds like I am making excuses for them saying that though. I guess we'll see when AIBs come out if they are less expensive or not.

[chronia]

Not really, FE is FE, AIB cards are AIB cards, FE inherently retails for more, AIB cards retail depending on their R&D and production costs (they dont have to stick to GPU model MSRP). That absolutely still leaves board partners some room to make their cards cheaper than FE.

With Vega I'm not seeing that at all.

I wouldn't expect most AIB cards to be cheaper, esp not when compared to day 1 MSRP, see my earlier post on page 40 with some quotes from the MSI marketing director for the Benelux. He expects prices to go up by quite a amount due to shortage, Vega being a hothead so AIB coolers need to be top notch and AMD lowballing day 1 MSRP.

Now ofc his predictions are made on the day 1 MSRP, seeing that cards have gone up (atleast in EU) by +- 100 euro today this might mean that AIB cards might get close to the new prices.

[Gaidax]

The problem is, the FE was nothing more than a glorified reference cooler, which historically have been less expensive than AIBs. Of course, we won't know for quite some time, but perhaps that's what AMD was attempting to do here, but then still give early adopters a better price? I dunno, it kinda sounds like I am making excuses for them saying that though. I guess we'll see when AIBs come out if they are less expensive or not.

But they officially hyped it as $499 solution???

[Evildeffy]

Not really, FE is FE, AIB cards are AIB cards, FE inherently retails for more, AIB cards retail depending on their R&D and production costs (they dont have to stick to GPU model MSRP). That absolutely still leaves board partners some room to make their cards cheaper than FE.

With Vega I'm not seeing that at all.

Failing to see the point I see...

Most of the time, as prior history has shown, MSRP was actually pretty close before the GTX 10 series and this series Vega launch (mining crap not included), the MSRP was also upheld with Polaris AIBs for the most part even.

The point is that FE is nothing but reference design and "garbage" compared to 3rd party designs... and the AIB partners should have been cheaper and the point was that they weren't always doing that because nVidia was a big enough dick to invoke FE tax and board partners had to do the same.

It's the same type of trickery with playing with prices, that was my point, nVidia asked more for FE to have AIBs justify asking more for their designs.
AMD asked less with their reference designs to jack up the prices later, either which way MSRP isn't adhered to at all and we suffer for it.

Clearly I have to elaborate to the detailed points to get it across.

- - - Updated - - -

But they officially hyped it as $499 solution???

They did ... which is why it is a dick move.