AMD Pulls Radeon "Vega" Launch to October

[TheinsanegamerN]

What is the point of GDDR5X 256 on gtx 1080, if 356 bus of GDDR5 is still faster ... ? maybe i don't understand something....

It comes down to several things.

A.) a 256 bit bus is much easier to wire up physically on a PCB than a 384 bit bus, to say nothing of a 512 bit bus. This reduces the number of faulty PCBs during manufacturing and decreased PCB complexity, increasing yields.

B.) a 256 bit bus does not require as many memory controllers (4 64 bit buses compared to 6 for 384 bit or 8 for 512 bit) which cuts down on not just power use, but chip complexity as well. Again, cheaper and easier to make with better yields than a 384 bit chip. This was also part of the reason hawaii was such a power hog, a 512 bit bus is hard to make power efficient.

C.) a 256 bit bus can outdo a 384 bit bus, depending on the memory tech behind it. We dont know how much bandwidth GDDR5X pushes on the 1080. It could very well push more data than a 384 bit GDDR5 bus. GDDR5X clocks much higher, and the final bandwidth number is what matters, not the number of bits. GDDR5X also has larger memory chips available than ddr5 does IIRC, making it cheaper to offer 8GB VRAM capacities. Fewer memory chips also means lower power draw.

 

[EarthDog]

Guess that means NVidia won't respond with the 1080ti until after October. Gotta wait even longer

I didn't think October was the date due to HBM2 yields....

 

[ensabrenoir]

Amazing how much information is here

As of now, NOBODY had neither 1070, 1080 or any Polaris for an independent test, Nvidia silicon wasn't shown last week (somebody smart will quickly claim opposite, yes CEO was *holding* one), we had selective benchmarks by manufacturer claiming *something* - and final 'proof' was some guy supposedly running unreleased game on youtube at, supposedly, insane speed.

Meanwhile, in the opposite corner, several months ago we saw unnamed card beating another 'comparative' one (at least it was manufacturer silicon) in... power-usage. And now the release date of another future product suddenly become CLOSER for 6 months, despite the opposite being true in most cases.

But thanks to these informative posts, now I know so much more! Who-was-punched-and-where and what other cards beat TitanX and so on...

Personally, I consider card released when I am able to buy it in local store. At that time, there are usually whole bunch of benchmarks proving this-and-that, and prices are often adjusted from initial recommendations. If I'm buying something (usually mid-range), I decide *then*. But that's just me...

.....Most are going by past amd/nvidia release vs marketing experience. Which has been consistent the past few years.

 

[efikkan]

Is it just me, or is the naming scheme retarded. Polaris has 10 faster than 11, and Vega has 11 faster than 10 ?!?!?!

That's a mistake, and another indication that it might be BS. The number 10 chips will be faster in both cases.

 

[v12dock]

Blah blah blah speculation... We all know AMD is more concerned with bigger clients Sony, Microsoft, Nintendo (Polaris) than pushing out a high end card.

 

[cdawall]

Blah blah blah speculation... We all know AMD is more concerned with bigger clients Sony, Microsoft, Nintendo than pushing out a high end card.

Sony will no longer be a client if these PS4K's don't sell well. That company is a sinking ship.

 

[Eroticus]

Sony will no longer be a client if these PS4K's don't sell well. That company is a sinking ship.

Sony won't make cash from Console sales.

Money is coming from PSN+ and Game sales.

"PlayStation 4 Build Cost Is $381- Only $18 Below The Retail Price"

 

[v12dock]

Sony will no longer be a client if these PS4K's don't sell well. That company is a sinking ship.

Sony's only viable brand is Playstation they will have been killing everything else off to keep afloat. Not like any of this is AMDs problem as long as they can pay. Besides they all have deeper pockets than your average Joe.

 

[ZoneDymo]

Sony won't make cash from Console sales.

Money is coming from PSN+ and Game sales.

"PlayStation 4 Build Cost Is $381- Only $18 Below The Retail Price"

Ermm both Sony and MS have not made money of the console sales ever....
Only Nintendo did that but with the WiiU that changed as well.
Its about the game sales, not the console.

 

[cdawall]

Sony won't make cash from Console sales.

Money is coming from PSN+ and Game sales.

"PlayStation 4 Build Cost Is $381- Only $18 Below The Retail Price"

Right and if you don't buy the console I would imagine you probably wont buy any games for it.

 

[TRWOV]

HBM2 will be scarce throughout Q1 2017, but why not use GDDR5X? It's more than fast enough for gaming anyway.

Exactly. vega should have been built around gddr5x, with another super chip built around HBM2.

Maybe that's what we'll see... a stop-gag Vega GPU with GDDR5X while the HBM2 version launches.

If Polaris is ready for production I suppose that Vega is close too and the only reason for the Q1 2017 launch was the dependency on HBM2.

 

[$ReaPeR$]

well, this is good news..

 

[neatfeatguy]

......Maybe Amd was focusing on the consoles and figured nvidia would take it easy this go around and release something with very minor performance increases.

I find it hard to believe that AMD would venture a guess that Nvidia would be taking it easy. Myself (and hopefully most other people), I figured Nvidia's Pascal was going to give a great performance jump based on the available information floating around the internet. Granted you can't believe everything out there, but enough gave you an idea of what could be expected. AMD should have been able to draw a similar conclusion.

If the 1070 actually gives 980Ti performance, I'd say that's a pretty good jump from Nvidia. That means they still have the 1080, 1080Ti and possibly Titan that just widens the gap even more from Maxwell's high end performers.

AMD can't keep playing the "wait and see what our competition brings to the table" game. If they can get Vega out before Nvidia's top tier cards (1080Ti and/or a Titan model) and can prove they outclass the 1080 with ease, then they can hold the market hostage for a bit until the 1080Ti or new Titan comes out to compete.

 

[ensabrenoir]

I find it hard to believe that AMD would venture a guess that Nvidia would be taking it easy. Myself (and hopefully most other people), I figured Nvidia's Pascal was going to give a great performance jump based on the available information floating around the internet. Granted you can't believe everything out there, but enough gave you an idea of what could be expected. AMD should have been able to draw a similar conclusion.

If the 1070 actually gives 980Ti performance, I'd say that's a pretty good jump from Nvidia. That means they still have the 1080, 1080Ti and possibly Titan that just widens the gap even more from Maxwell's high end performers.

AMD can't keep playing the "wait and see what our competition brings to the table" game. If they can get Vega out before Nvidia's top tier cards (1080Ti and/or a Titan model) and can prove they outclass the 1080 with ease, then they can hold the market hostage for a bit until the 1080Ti or new Titan comes out to compete.

......never said it was a smart plan....they're playing the only hand they got.

 

[Casecutter]

huh. doesn't sound good to me.

Why now that GP104 seemed to gotten pushed-up form most original speculation (more end of July) is good, but this isn't?

Then a gain, could be FUD or stealth PR.

More likely it is... as most of what we get here is anymore.

does this mean they think Polaris will be outmatched

I think only you, thought Polaris was supposed to be a GP104 competitor.

hitting on the back to school times

As a parent you buy your kid some awesome "play toy" when he/she's supposed to be hitting the books? The out-of-school summer, H-S Graduation, crowd that's when aspiring gamers or a parent might buy Jonny/Jane something to keep them entertained. Figure most 1070 AIB custom's will price out like $400-430, especially as they could see limited channel ingress. If Polaris has a true volume release, I think there's a lot more gamers/parents having $300 (≈25% lower) to spend. I think AMD could be in a good place.

1070 prices at $380

See above ^. I think the chance of a "decent" 1070 for $380 will be perhaps more like the end of summer.

Oh, a rushed product.

And we don't think Nvidia has bumped up the 1070/1080. Whom is more rushed?

now willing to launch with very limited supply?

I think that's the same for both. Nvidia got GDDRX and Vega is moving up even though it means no having a full channel... Seem as though "what was good for the goose, is good for the gander".

they need to compete now

No they need to sell cards in the most lucrative and profitable segment.

vega should have been built around gddr5x, with another super chip built around HBM2.

I don't think you could want/do two dissimilar memory controllers on one die.

 

[laopa]

Wait.......Are you saying "Grenada"? But that is the 390/390x family name. So....what will Polaris do? I mean now everthing is confusing haha.

 

[rav]

Looks like AMD just stole NVidia's Christmas, AGAIN!!!

GTX 1080 is ONLY 13% faster than 28nm R9 Fury X and to get there it must be clocked to 1.6-2.1 gHz despite being a 16nm process. NOT very impressive since R9 clocks at 1.05gHz!!!!

Vega with HBM2 AND a 14nm process AND Asynch Compute in HARDWARE not SOFTWARE like NVidia will romp all over GTX right at Christmas!!!

Very SMART move by AMD.

The 14nm Polaris release already shows a huge advantage over GTX 980 and that will likely price around $350 or so.

 

[Eroticus]

Right and if you don't buy the console I would imagine you probably wont buy any games for it.

Looks like AMD just stole NVidia's Christmas, AGAIN!!!

GTX 1080 is ONLY 13% faster than 28nm R9 Fury X and to get there it must be clocked to 1.6-2.1 gHz despite being a 16nm process. NOT very impressive since R9 clocks at 1.05gHz!!!!

Vega with HBM2 AND a 14nm process AND Asynch Compute in HARDWARE not SOFTWARE like NVidia will romp all over GTX right at Christmas!!!

Very SMART move by AMD.

The 14nm Polaris release already shows a huge advantage over GTX 980 and that will likely price around $350 or so.

DirectX12 Performance ?

cuz non x 380 is better than 980....

 

[MLScrow]

Guys, this is how these two GPU makers have been competing from the beginning. Why people are getting their panties in such a bunch is unreasonable. AMD announces Polaris for low-mid rage up to $299 with better performance and efficiency than anything NVidia has to offer, all while stating that their high end won't come out till 2017. NVidia see's this and says, "Yes, if we release now, we can own the market for a while before AMD can respond." And NVidia counters with 1070/1080. AMD laughs, says, they moved the way we thought they'd move, now we respond with Vega not in 2017, but October 2016, and we won't release the full fat versions, we'll release a 4096 and a 6xxx sp version. We'll then wait for NVidia to respond with 1080Ti and Titan 2, and then we will counter again with an 8192 sp full fat GPU. The fact that AMD is targeting the VR market so much, makes me believe that we'll be seeing a lot more of the dual GPU-type cards as AMD seems to like the idea of dedicating one GPU per eye as they did with the Pro Duo.

It's a back and forth game that has been played for eternity and is very obvious. People should relax and enjoy the bout as usual and purchase whatever you want in the end, because it'll all be good.

 

[Assimilator]

Assuming that necessary HBM2 stockpiles don't materialise by October, AMD has 3 options, none of them good:

* Release Vega 10 with HBM2 and almost no availability due to a lack of HBM2 supply. It'll be like Fury X2's "launch", only worse - if that's even possible.
* Rework Vega 10's memory controller to interface with HBM1. Probably not feasible due to the 4GB limitation of HBM1 which would likely cripple Vega, and I'm not sure how much - if any - HBM1 will continue to be produced now the industry is going to HBM2.
* Rework Vega 10's memory controller to interface with GDDR5/X. Supply won't be a problem and the characteristics of GDDR5 are a known factor, but since we can safely assume Vega has been designed around HBM2, it's almost certain that using GDDR5 will result in sub-optimal performance. That's not to mention the fact that the chip design will have to be respun and retested, and the PCB will have to be redesigned around a small GPU and separate memory chips as opposed to a single monolithic MCM...

Of course there's always the possibility that HBM2 production is progressing a lot better than has been revealed, and Samsung is just keeping their cards close to their chest - although I don't see why they would, since any positive news about HBM2 would result in a press release to boost their stock price. Then there's the fact that NVIDIA will be looking to grab HBM2 for GP100, and if they're able and willing to pay more for HBM2 supplies than AMD, we all know who Samsung will sell to.

It may be that AMD are betting on SK Hynix's HBM2 in Q3 rather than Samsung's - since SK Hynix was responsible for HBM1, and thus may have more success with HBM2 - but if that's so then they're cutting things very, very close.

Any way you look at it, the next 6 months are gonna be interesting. I will reserve judgement until I've seen hard benchmarks of GP104, but if it's as fast as NVIDIA claims, and GP100 is significantly faster... I fear that AMD is going to have a painful year.

 

[thewhitestigvk]

You all know what this means, right? AMD realized Polaris couldn't compete with GP104 so they pulled their next gen cards to this year in order to battle the 1080. But in essence this must mean they have absolutely nothing against the fat Pascal. This isn't good.

 

[medi01]

Is it just me, or is the naming scheme retarded. Polaris has 10 faster than 11, and Vega has 11 faster than 10 ?!?!?!

Numbers are time based.
So with Polaris they started with bigger chip first, so 10 > 11.
With Vega other way round (it seems)

 

[Casecutter]

It may be that AMD are betting on SK Hynix's HBM2 in Q3 rather than Samsung's - since SK Hynix was responsible for HBM1, and thus may have more success with HBM2

I would consider RTG has SK Hynix HBM2 as the prominent supplier, and was reported back March 2016, "mass production of its 4GB and 8GB HBM2 stacks in Q3 and Q4 2016 respectively". So if there's even a trickle in Q2 of good stuff they'll use it. I think there work with the interposer and such manufacture partner on Fiji, gave them a firm foundation to be in front.

While I thought Nvidia since it patched thing up with Samsung will have their production, that was to commence mass production of HBM stacked memory starting early 2016. At issue, is all that is for Nvidias' HPC contracts and Tesla production first. They will fill the high margin Professional first, so I would look for a Titan/GTX till into the Q1 2017.

 

[Kanan]

The source of this news is a german news site, let me translate it to you:

"From our forums the rumour is spread that AMD changed to launch date of Vega to October."

It's a 99,9% rumour. But have fun discussing it and geting emotional about something which is probably not more than hot air.

It's very unlikely AMD is this fast with their HBM2 cards, whereas Nvidia is not. Also they would go for GDDR5X on an performance card, not HBM2, because it's senseless there. Only the fastest chips will need very expensive HBM2, it's a waste on performance cards.

Also, Vega is not a performance chip, it's enthusiast segment with HBM2 (that's why it has HBM2!).

hahaha.. so many things are wrong here.

 

[G33k2Fr34k]

The source of this news is a german news site, let me translate it to you:

"From our forums the rumour is spread that AMD changed to launch date of Vega to October."

It's a 99,9% rumour. But have fun discussing it and geting emotional about something which is probably not more than hot air.

It's very unlikely AMD is this fast with their HBM2 cards, whereas Nvidia is not. Also they would go for GDDR5X on an performance card, not HBM2, because it's senseless there. Only the fastest chips will need very expensive HBM2, it's a waste on performance cards.

Also, Vega is not a performance chip, it's enthusiast segment with HBM2 (that's why it has HBM2!).

hahaha.. so many things are wrong here.

There is some truth to this rumor. Vega is rumored to be around 380mm^2. AMD's approach to high performance is in the case of Vega is 60% more shaders than the fully-enabled Polaris chip. Vega will definitely be able to at least compete with the GTX1080, which is what AMD need sto do real soon to start selling high profit GFX cards.


What Nvidia did with the GP104 is just unprincipled. If the performance rumors are true, then TitanX performance for $380 when a 390/GTX970 can play all your games at decent frame rates on max settings is is just unnecessary. Nvidia is trying to strangle AMD and squeeze them out of the high performance high profit GFX market.