AMD VEGA confirmed for 2017 H1

[Boomstick777]

But they keep selling cards well after launch and how many times have we seen Nvidia cannot do no wrong yet Amd are getting slammed even before the cards are launched it was different when Amd was in the lead and gets their cards out first.

At least let them release the card before slating it, after it launch then slate it if you don't think it is good enough.

^^

 

[Lokken86]

People do realise the Pascal Titan X is actually not using a full chip?? There is the leeway Nvidia has - they release a fully enabled GP102 based graphics card.

Nope. To add, the TitanXP should be the gtx1080 and not this abomination selling for a frankly outrageous £1200. Price gouging at it's finest tbh.

 

[drunkenmaster]

Nope you are still making assumptions.

The pic I linked definitely states 3584 FP32 cores.

I should not have to point out to you that FP64 is no good for gaming.

I also notice that your addition is a bit out if you are adding the two together.

Just for the record both the Pascal Titan and the Quadro above have exactly the same number of active cores (3584).

Nope, you didn't state FP32 cores, you may have meant that, but you did not state it you said ACTIVE cores, nothing more or nothing less. As for FP64 being no good for gaming, in reality that isn't strictly true, there is, afaik, absolutely nothing stopping the software pushing a 32bit number through a FP64 shader, the question is can Nvidia bundle two 32bit instructions into one FP64 bit shader... the answer is probably no. The complexity is the power usage, the car has a similar power limit to a card without the FP64 shaders so using them all at the same time likely causes the card to throttle.

The point of my first post was to be pedantic, because that is what you were doing with your response, you were technically wrong so there you go.

Regardless you managed to miss that, no cores besides GP100 contain NVlink, so regardless of your intent for the people who know only GP100 contains Nvlink, posting a picture of GP100 and talking about a potential gaming card due to the NVlink connector.... could only be talking about Nvidia releasing a Gp100 for gaming, which isn't happening as it has no way to be faster than GP102 as a gaming card. GP102 has an extra couple of SM's available to be turned on for FP32 performance and has none of the efficiency issues of carrying an extra 1700+ FP64 cores that leak power and reduce the available TDP for Fp32 cores and preventing higher clock speeds.

 

[Kaapstad]

Nope, you didn't state FP32 cores, you may have meant that, but you did not state it you said ACTIVE cores, nothing more or nothing less. As for FP64 being no good for gaming, in reality that isn't strictly true, there is, afaik, absolutely nothing stopping the software pushing a 32bit number through a FP64 shader, the question is can Nvidia bundle two 32bit instructions into one FP64 bit shader... the answer is probably no. The complexity is the power usage, the car has a similar power limit to a card without the FP64 shaders so using them all at the same time likely causes the card to throttle.

The point of my first post was to be pedantic, because that is what you were doing with your response, you were technically wrong so there you go.

Regardless you managed to miss that, no cores besides GP100 contain NVlink, so regardless of your intent for the people who know only GP100 contains Nvlink, posting a picture of GP100 and talking about a potential gaming card due to the NVlink connector.... could only be talking about Nvidia releasing a Gp100 for gaming.

More assumptions and you have not mentioned that you got your addition wrong.

 

[Gerard]

Handbags at dawn.

 

[drunkenmaster]

More assumptions and you have not mentioned that you got your addition wrong.

Rather than posting, state which assumptions I'm making, state how cores other than GP100 have NVlink, I provided the quote and the link from arstechnica and other sources showing that GP102 has only SLI links, because it doesn't have NVlink, and that you can't find NVlink mentioned around a quadro P6000, Titan, any GP102 through 106 card, it's only mentioned in relation solely to GP100 and again arstechnica literally states that it has a unique NVlink connector differentiating it in the market from P6000.

Rather than posting pointless replies, why not highlight what assumptions I've made and post any links proving me wrong about NVlink, or is it because you can't?

If you're referring to 5379 instead of 5376... look at a numpad, it's a typo, not an incorrect addition, you managed to state GP100 has 3584 active cores... I was 3 out because of a typo, you were 1792 out because you don't know what you're talking about... three times you've failed to mention how you implied NVlink connector could be used on cards that don't contain NVlink.... for some kind of gaming boost.... when even if they had NVlink, it would offer zero boost to gaming performance anyway.

 

[Kaapstad]

Rather than posting, state which assumptions I'm making, state how cores other than GP100 have NVlink, I provided the quote and the link from arstechnica and other sources showing that GP102 has only SLI links, because it doesn't have NVlink, and that you can't find NVlink mentioned around a quadro P6000, Titan, any GP102 through 106 card, it's only mentioned in relation solely to GP100 and again arstechnica literally states that it has a unique NVlink connector differentiating it in the market from P6000.

Rather than posting pointless replies, why not highlight what assumptions I've made and post any links proving me wrong about NVlink, or is it because you can't?

As my original post was solely about the colour of the NVLink everything else you added is an assumption.

And you still have not conceded that your addition is terrible.

 

[Kaapstad]

Rather than posting, state which assumptions I'm making, state how cores other than GP100 have NVlink, I provided the quote and the link from arstechnica and other sources showing that GP102 has only SLI links, because it doesn't have NVlink, and that you can't find NVlink mentioned around a quadro P6000, Titan, any GP102 through 106 card, it's only mentioned in relation solely to GP100 and again arstechnica literally states that it has a unique NVlink connector differentiating it in the market from P6000.

Rather than posting pointless replies, why not highlight what assumptions I've made and post any links proving me wrong about NVlink, or is it because you can't?

If you're referring to 5379 instead of 5376... look at a numpad, it's a typo, not an incorrect addition, you managed to state GP100 has 3584 active cores... I was 3 out because of a typo, you were 1792 out because you don't know what you're talking about... three times you've failed to mention how you implied NVlink connector could be used on cards that don't contain NVlink.... for some kind of gaming boost.... when even if they had NVlink, it would offer zero boost to gaming performance anyway.

First thing you learn in IT is to proof read to avoid things like poor addition.

Until such time as you can use FP64 for gaming both cards have 3584 active FP32 cores for that purpose.

 

[Panos]

People do realise the Pascal Titan X is actually not using a full chip?? There is the leeway Nvidia has - they release a fully enabled GP102 based graphics card.

The full chip goes for £3000+ in the form of the Quadro. TXP is the "failed" Quadro chips.

 

[drunkenmaster]

First thing you learn in IT is to proof read to avoid things like poor addition.

Until such time as you can use FP64 for gaming both cards have 3584 active FP32 cores for that purpose.

We're on a forum so no one is spending half an hour proof reading every post, it's not a formal presentation. In fact the only time anyone cares about grammar/typos on a forum... is when they are completely and utterly wrong and can't respond to the content of someone's reply to them. Also, the first thing in IT isn't learning to proof read, I honestly can't even wrap my head around you stating that, it's not a good comeback, it's not true, it doesn't mean anything at all. If I wrote a technical manual and my boss found a mistake he might say hey, your job is writing tech manuals, typos aren't acceptable..... but then every manual has mistakes regardless so meh, just try not to make any more.

You won't let a typo go, but you're ignoring that you tried to claim that NVlink bridge maybe used on cards that don't support NVlink. It's pretty obvious you won't discuss any of my points so why do you bother coming on here.

Gaming, again, and I quoted you before for easy access, you said the quadro had the same number of active cores as a Titan, not the same number of FP32 cores OR the same gaming performance, there was no quantifier, you stated they have the same number of active cores, compared to either GP100 or P6000(because you said quadro, not a specific quadro) neither has the same number of active cores, so you are wrong on this, flat out.

Second, I pointed out that the same number of FP32 cores doesn't mean the same gaming performance, GP100 has the same number of FP32 cores, but an extra 1792 power leaking FP64 bit cores, meaning in any relatively similar power bracket, there is no way GP100 can run at the same clock speeds as a GP102, so it won't match it's performance, ever, so GP100 has zero relevance as a gaming card. GP102 also has a further what 200+ FP32 cores that can be used for a gaming card without the downside of the Fp64 cores to take up a large chunk of the TDP.

SO your post had no merit, Nvlink is only available on GP100, so any idea of NVlink helping gaming could only come on GP100 and GP100 has no relevance to gaming, less performance for gaming than GP102.... but significantly higher price... wooo.

 

[Kaapstad]

Snip

I refuse to take this thread anymore off topic by arguing with you. You have totally failed to understand my original post about NVLink.

Back on topic, will you be getting a Vega card?

 

[Deleted member 651465]

^ good move. I think it best if you chaps agree to disagree on this one.

 

[drunkenmaster]

I refuse to take this thread anymore off topic by arguing with you. You have totally failed to understand my original post about NVLink.

Back on topic, will you be getting a Vega card?

Last post, won't respond again, you implied NVlink could come to other cards, I didn't get that ridiculous idea.... because Nvlink is not available on the other cards you apparently thought it might come on. You accused me of making assumptions, but those things were directly implied because only GP100 supports Nvlink. You not knowing this decided I was assuming things not evident, but if you realise only GP100 came with Nvlink, to anyone who knows that, your original post can ONLY mean you think Nvidia will release a GP100 based gaming card, there is literally no other possibility once you realise that no core but GP100 supports Nvlink.

 

[meldarthx]

Let's look at some facts and rumors.

Rumors - what was running at CES was Big Vega (Despite what some say its not confirmed)
Rumors - what was running CES was Little Vega (Again Despite what some say its not confirmed)
Rumors - Polaris respin in the works (this is still in the rumorville)
Rumors (Little Vega to replace Polaris) This has some weight to it - but from what Raja has said in his interviews; Vega 10 is to replace Hawaii/Grenada series; and Vega 11 replaces Fury Series. If that is the case and I believe it is.......then very well possible Vega 10 - was what was running Doom and ashes and easily be around 1080; while Vega 11 - will be a lot faster.....

Facts Vega is not GCN; but NCU - IP9 vs IP8 architecture in Polaris and Fiji. It's a bigger change than Polaris. (Rumor) This along with HBM2 may explain the bold claims from AMD that Vega will be much more power efficient than Polaris.

Honestly I don't think we've seen Vega 11 - AMD doesn't need to drop the Big Gun just yet specially if Vega 10 beats 1080 which I believe it does.

Makes not sense for Vega 10 to replace Polaris 10 which has been out 6ish months to 8 when its release. Polaris is doing its job well - Vega 10 will take on 1070/1080 - IMHO and Vega 11 - will much faster

 

[Rroff]

There was definitely a comment from an AMD person at some point saying that the small Vega process was behind the big one on timeline and that they didn't have working samples of small Vega at the point the comment was made. I can't off the top of my head find it now.

EDIT: According to reddit its comments Raja said in a video which is why I can't find it googling.

 

[Besty]

i would expect that the chip that well held up and shown to the press was big vega and the card running the doom demo was the same chip. I expect that, more than anything it was to re-assure investors rather than gamers that AMD had a working prototype and not just a chip

 

[Sargatanas2511]

I maintain that even a Polaris with 4096 shaders and no other optimisations should be pushing a 1080 quite close.

With all the other optimisations that Vega will have, I can't imagine why it would only compete with the 1080.

 

[Diagro]

People!

This thread is for VEGA... NOT NVIDIA poop cards.

Please get back on topic, my eyes hurt from all the poo im reading

 

[Rroff]

I maintain that even a Polaris with 4096 shaders and no other optimisations should be pushing a 1080 quite close.

With all the other optimisations that Vega will have, I can't imagine why it would only compete with the 1080.

Just incrementally increasing shader count doesn't necessarily equate to anything like a linear increase in performance - infact you can end up with quite a lot of utilisation problems depending on the architecture.

People!

This thread is for VEGA... NOT NVIDIA poop cards.

Please get back on topic, my eyes hurt from all the poo im reading

Use it - if it isn't about Vega there probably isn't anything new :s

 

[JediFragger]

i would expect that the chip that well held up and shown to the press was big vega and the card running the doom demo was the same chip. I expect that, more than anything it was to re-assure investors rather than gamers that AMD had a working prototype and not just a chip

Big Vega will have 6144 shaders yes? I'm expecting it to be a bit of a beast tbh and eclipse the TXP (at least in DX12 titles)

If it has 16GB of HBM2 I can see it being approx £1k. I'd be happy-ish with that, esp considering it'd last me a goooooooooooood long while (3840x1600 res)