Report suggests AMD has pulled 'Vega' GPU launch forward to October

[Yarrow]

I have to say I love both AMD and Nvidia, both companies seem to be providing or planning to provide exactly what the market currently wants low power and high performance. I have Cards from both companies and am really happy with both of them... I look forward to trying the new 1080 on my laptop to see if I can get 4K performance, but really look forward to Second Gen VR and what these advances will provide. Its an exciting time in PC Hardware.

 

[mateau]

AMD did "move up" anything. They simply revealed a conservative road map then waited to see what NVidia does. Then they acted.

Since 16nm GTX 1080 has to be overclocked to beat 28nm R9 Fury X that has given AMD the entry point for 14nm Vega.

GTX 1080 is actually a disappointment. A new process and clock for clock GTX 1080 is no better than the old process. Yes 16nm GTX 1080 can be clocked to 2.1 gHz. So can 14nm Vega WITH HBM2.

AMD stole NVidia's Christmas AGAIN.

How is Nvidia's new product a "dissapointment" when it hasn't even been released yet. You base your assumptions on an Nvidia press event. Until the real product comes out, and we see real world testing, anything anyone says at this point is speculation.

I have to say I love both AMD and Nvidia, both companies seem to be providing or planning to provide exactly what the market currently wants low power and high performance. I have Cards from both companies and am really happy with both of them... I look forward to trying the new 1080 on my laptop to see if I can get 4K performance, but really look forward to Second Gen VR and what these advances will provide. Its an exciting time in PC Hardware.

There are reports of GTX 1080 dominating Titan X and Fury X. Evidently it is only 13% faster despite being 16nm and clocked at 2.1gHz. I do not call that any victory at all. Both Titan X and Fury X are 28nm processes and are clocked at 1050 MHz or so.

13% is a trifling disappointment

 

[mateau]

Panicking over geforce 1080 perhaps?

The general feeling is that AMD were pretty shocked over the clock speeds Nvidia were able to get out of their silicon. The architectural tweaks of Pascal aren't anything wondrous, but overall performance gets a big lift from the sheer speed the silicon runs at. Seems AMD were banking on more conservative clocks and allied with their own purported failed clock speed validation for Polaris at modest clocks. it might have them in a nervous state.

Having said that, the Vega in October rumour originates from someone posting at SemiAccurate's forums. I'd suggest the rumour be taken with more than a pinch of salt

Wonder if they will manage to fit on HBM2 or go with GDDR5X

Vega was slated to use HBM2. Having said that, Synopys (the company that lays out AMD's chips) shouldn't have any difficulty swapping out the HBM I/O and controller logic blocks for GDDR5X if required.

I originally thought that the 1080 would be HBM# based and the 1070 would be based on GDDR5X, but was of course wrong.

It was an unrealistic expectation. HBM2 hasn't started volume production, and even if it had, the assembly costs (GPU+HBM2+CoWoS interposer) would make it economically inviable at the prices the cards need to sell at. AMD effectively lost money on Fiji - I doubt Nvidia was planning on following suit. At the moment, I suspect a huge portion of HBM2 production is going towards getting the 4,500 Tesla P100 board contract the Piz Daint's upgrade fulfilled. At $10,000 per GPU module, it's a no-brainer as to where the HBM2 production should be going.
GDDR5X is very likely the same proposition. There is probably enough initial production to keep the GTX 1080 assembly line happy, but that will be a relatively small volume of cards. If the GTX 1070 mirrors the GTX 970 and sells 250,000 cards a month, that would be a tough ask for a newly ramping technology. I wouldn't be surprised to see a GTX 1070 refresh feature GDDR5X though - the memory controllers are obviously the same as for the 1080.

AMD did "move up" anything. They simply revealed a conservative road map

Basically the opposite of what AMD usually do then.
Vega was originally slated for 2017, but bringing a roadmap in is a whole lot better than letting it lapse...or forgetting about it completely

AMD is going to have a very good year. They will be profitable in the July earnings and if they launch Vega just before October earnings then they will be seen as firing on all 8 cylinders. There should be console announcements too.

 

[mateau]

Panicking over geforce 1080 perhaps?

That's probably a portion of it. Either that or they want to be hold the performance crown for a good period this time around.

1080 is a disappointment.

2.1gHz, 16nm and ONLY 13% faster than last years 28nm silicon?

FinFet process will deliver faster clocks at 16nm and even faster yet at 14nm like Polaris.

 

[dividebyzero]

There are reports of GTX 1080 dominating Titan X and Fury X. Evidently it is only 13% faster despite being 16nm and clocked at 2.1gHz. I do not call that any victory at all. Both Titan X and Fury X are 28nm processes and are clocked at 1050 MHz or so.
13% is a trifling disappointment

You'll have to spin harder than that, young AMD padawan. Since mathematics is obviously not your strong suit, at that 2.1GHz the GTX 1080 is already showing a 61% increase over Titan X and a 57% over Fury X in Firestrike Ultra.

AMD is going to have a very good year. They will be profitable in the July earnings

AMD are always about to be become profitable...then the quarterly earnings are announced and the revolution is postponed for another quarter. Everyone's been hearing the same tune since 2007 - Saying something will happen doesn't mean it will - Just ask Derrick Meyer or any AMD feelgood marketeer who had the job before you got it.

1080 is a disappointment.......people will believe it when they see it. Too many have had to listen to the moronic prattle only to see the company slam the door on its nuts year after year.
2.1gHz, 16nm and ONLY 13% faster than last years 28nm silicon?

More of the same bad math was worth a third consecutive post?

 

[jauffins]

AMD did "move up" anything. They simply revealed a conservative road map then waited to see what NVidia does. Then they acted.

Since 16nm GTX 1080 has to be overclocked to beat 28nm R9 Fury X that has given AMD the entry point for 14nm Vega.

GTX 1080 is actually a disappointment. A new process and clock for clock GTX 1080 is no better than the old process. Yes 16nm GTX 1080 can be clocked to 2.1 gHz. So can 14nm Vega WITH HBM2.

AMD stole NVidia's Christmas AGAIN.

AMD fanboy, detected.

Also, I haven't been as up (apparently) as I could have been, RE: AMD's Polaris. I don't know if it was all the confusion of the 290/390 series (some are rebadged, some aren't, etc) but I thought Polaris was their Pascal. I didn't know Polaris was considered "mid-range." I guess that's why the AMD fans have been clamoring for it: many of them seem to be stuck firmly in the mid-range; CPUs, GPUs, etc.

 

[dividebyzero]

AMD fanboy, detected.

What gave him away, the wall of consecutive shilling posts, or the "return to profitability" punchline?

I thought Polaris was their Pascal. I didn't know Polaris was considered "mid-range." I guess that's why the AMD fans have been clamoring for it: many of them seem to be stuck firmly in the mid-range; CPUs, GPUs, etc.

Polaris is AMD's Pascal - just not a GTX 1080/1070 challenger by the sounds of it. Probably closer to Nvidia's GP 106 chip, which like Polaris should be a mid-range offering concentrating on more budget-minded gaming and mainstream laptop performance. The specifications tend to indicate 390/290X/290 (and below) performance using lower power.

 

[hay fizzy]

Panicking over geforce 1080 perhaps?

The general feeling is that AMD were pretty shocked over the clock speeds Nvidia were able to get out of their silicon. The architectural tweaks of Pascal aren't anything wondrous, but overall performance gets a big lift from the sheer speed the silicon runs at. Seems AMD were banking on more conservative clocks and allied with their own purported failed clock speed validation for Polaris at modest clocks. it might have them in a nervous state.


Qualcomm who are using the same process are getting over 2.2GHz clock speed so Nvidia getting 2.1 is not a surprise. AMD should be able to increase performance in a very short period of time by just binning at a higher clock speed and disabling less units, clocking memory and also producing a card with an extra lead for the higher wattage required.

 

[lripplinger]

There are reports of GTX 1080 dominating Titan X and Fury X. Evidently it is only 13% faster despite being 16nm and clocked at 2.1gHz. I do not call that any victory at all. Both Titan X and Fury X are 28nm processes and are clocked at 1050 MHz or so.

13% is a trifling disappointment

That is still speculation. Once again, wait until the products are out for people to test. Then you can draw whatever conclusions you want.

 

[Darth Shiv]

Pretty sure AMD would be worried about the 1080. They have been knocked senseless for a few gens now. The market knows it - have a look at the desktop market share. It's a blood bath.

And you can't count your chickens - until AMD actually makes it back in the black, I ain't holding my breath.

 

[dividebyzero]

Qualcomm who are using the same process are getting over 2.2GHz clock speed so Nvidia getting 2.1 is not a surprise.

There's rather a lot of difference between a relative straighforward ARM-based chip like the Kryo and a monothithic GPU.

AMD should be able to increase performance in a very short period of time by just binning at a higher clock speed and disabling less units, clocking memory and also producing a card with an extra lead for the higher wattage required.

No doubt, but the initial clock speeds don't gel so far with the specs given for Polaris chips. The apparent FLOPS of Polaris 10 are supposed to be 5.5 TFLOPS ( a little better than Hawaii). Knowing the FLOPS and a range of possible ALU counts automatically provides the clock speed.
5.5 TFOPS / 2 ops/clock / 2048 ALUs = 1340-1345MHz. No doubt AMD have left some overclocking headroom, but if their mission was to stay within a 150W envelope that might not too great.
As for whether AMD could actual deduce Pascal's ability, Kyle Bennett (among others) have stated that Pascal's clocking range came as a surprise to the Nvidia engineers. I find it a little difficult to believe that AMD knew more about Pascal than the guys who designed it.

I would suggest that even NVIDIA is blown away by how overclockable the first cards have shown to be. Some good stories behind that, but I will have to discuss at a later date - Kyle Bennett HardOCP

If AMD indeed were well aware of Pascal's clocking and performance gained from such, why the hurried Polaris webinar, and the talk of Vega's arrival being pulled in. I don't buy that it is all coincidence.

 

[Roddy_K]

"Panicking over geforce 1080 perhaps?" - Not panicking. If you have been around from the beginning of the war between these two companies over the last 20+ years, you'd know that no one is panicking. These are all strategic moves made intentionally as either a trigger or a counter to the other's. AMD unveils Polaris at extremely low price points for low-mid range, providing high end performance at a $299 price point, stating that their high-end parts won't be available till 2017 (nice move). NV counters by revealing that they are not only releasing Pascal sooner than expected, but they are lowering the price to $379 for 1070 and $599 for 1080 in an attempt to smash AMD. AMD chuckles and responds by saying, "Oh, did we say 2017, just kidding, we'll give you the goods in October" to counter NVidia. NVidia will then counter with 1080Ti and Titan 2, and AMD will counter with Vega 11 and Pro Duo 2.

It's a continuous game of back and forth, as it has always been. Neither company is freaking out, neither company is the best. They both have their time in the spotlight. AMD has held the performance crown for the fastest single card GPU for 2 years+ now. Even when NV releases 1080, it still won't be as fast as a Pro Duo, but for a single GPU card, it will be incredible.

AMD is leaning toward utilizing multi-gpus more often than ever before rather than constantly trying to make one huge single die, especially considering their interest and investment in VR. They have learned that dedicating a GPU per eye will provide the best results possible. I like the idea too, but $1500 is stupid. If they brought that card down to $799, I'd consider it, because you can practically purchase two Fury Nano's for that much anyway.

Regardless of who you go with for a GPU, it won't really matter. Both companies are making incredible GPU's and you can't go wrong with either of them. I give credit to AMD for being the small little underdog company that forces the huge companies to reduce their prices, so that all of we consumers can benefit.

I find it amazing that everything AMD does is subpar in quality. CPU's and video cards are subpar in quality and it shows. They don't spend anything close to what Nvidia or Intel spend in R&D. They have a fraction of what the other company's have in market share. If you want the most bang for you buck you will buy Intel and Nvidia for a awesome and trouble free gaming experience.