I don't need to read what I've read years ago, you want to believe Charlie Demerjian, a loser at life who spend his whole life and waking moment hating on successful companies and can't even differentiate between GM200 and GM204 GPUs, good for you.
- Status
Arachnotronic
Lifer
- Mar 10, 2006
- 11,715
- 2,012
- 126
sontin, excellent find. I think the AMD fans have gone overboard in proclaiming that NV is going to be significantly behind AMD here.
NV is bigger, richer, and more focused on dGPUs than AMD is or really will ever be at this point, so it's kind of ridiculous to think that the guys who have been pummeling AMD in the marketplace will be caught with their pants down at this point.
NV is bigger, richer, and more focused on dGPUs than AMD is or really will ever be at this point, so it's kind of ridiculous to think that the guys who have been pummeling AMD in the marketplace will be caught with their pants down at this point.
It's not nVidia's MO though to use leading tech. They leverage what is best currently. It's worked well for them financially, but not at "being first".
There's an awful amount of defensiveness going around here for what appears to be some innocent analysis of the current situation and it's not coming from the ADF ...
I sincerely hope AMD comes out on top in perf/mm2 and perf/w with their releases this year. I seriously doubt they will.
I guess when things are going really bad for someone's favorite company, and it causes them to lose sleep and/or spend less time enjoying their hobby and more time lobbying for said company, they need all the slightest bit of favorable conjecture to help pump up their constantly sagging outlook.
I guess when things are going really bad for someone's favorite company, and it causes them to lose sleep and/or spend less time enjoying their hobby and more time lobbying for said company, they need all the slightest bit of favorable conjecture to help pump up their constantly sagging outlook.
With luck, GP104 (assuming that's the chip that would have been showed off for the PX2 if it were ready) is delayed, while GP100 is more or less on track. If the stars align, then maybe Nvidia could be forced to release a maintstream GP100 card to combat Polaris, and not some mid-range thing masquerading as a flagship.
The only engineering reason that GP104 would be delayed over GP100 is because HBM2 is ready now while GDDR5X is ready in June/July. Given that Nvidia hasn't led with it's big die for the past two generations, it would be out of the ordinary for Nvidia to lead with the big die now, especially on a new node. HOWEVER, also given that Nvidia hasn't released a new compute flagship card since GK110, I guess anything is possible.
Azix
Golden Member
- Apr 18, 2014
- 1,438
- 67
- 91
Pascal could be another fermi. Fermi was their big compute arch wasn't it? before kepler when they started cutting things. They've been talking about going big on compute again, and to me they would lose their energy efficiency advantage and possibly run into problems that AMD met head-on years ago with GCN.
HBM issues is just more on top of building a good compute GPU.
Could be a disaster. Could be good, could be just ok.
HBM issues is just more on top of building a good compute GPU.
Could be a disaster. Could be good, could be just ok.
Cookie Monster
Diamond Member
- May 7, 2005
- 5,161
- 32
- 86
Would makes sense except GK110 was their next big step in compute arch (DP focused). The GK210 still holds up its own in the compute space and maxwell has so far been excellent in the 3d performance side. If pascal is an evolution of the previous mentioned kepler/maxwell then there is less risk of it failing. I guess because its somewhat new on top of the new process tech that there is more risk than for previous 28nm based products.. atleast it will get interesting once again!
If it was completely new, there is always that risk of it failing but Pascal isn't a completely new uarch by the look of things. Plus there is a good chance that once again DP might be only available for Tesla customers.
Also wasn't the big Pascal going to be used for the NOAA project this year? yields could be terrible but they would still rake in alot of $$$ because of their pricetag.
The PX2 should be using GP106, not GP104. The slides indicate it will have 8 TFLOPS of performance; there are two GPUs, so that's 4 TFLOPS per chip. A GM204 can already exceed that (probably why they were using them for the prototype board before Pascal was ready). At 16FF+, Nvidia should be able to get that same 4 TFLOPS of performance in a chip that is roughly half the size of GM204 - ~200mm^2 instead of ~400. And that's a '6'-class die size.
Nvidia needs to compete with KNL on 14nm this year. They will want a top notch Pascal large die GPU to compete in HPC. NV still has a huge software advantage, but Intel is building a solid soup to nuts hardware system to compete (high speed multi-level RAM systems, local and wide scale connectivity, etc. I'm not sure what IBM and NV have lined up for hardware this year (aside from the GPU). Intel is also being very "aggressive" in pricing (sound familiar?).
I think this is what Nvidia is focused on. Given limited foundry capacity at TSMC for 16FF+, they'd rather prioritize GP100 compute-oriented chips that they can sell for multiple thousands of dollars apiece. They figure that Maxwell is already fully amortized and cheap to manufacture on 28nm, so they can play that out in the consumer space for another year. There are a lot of low-information buyers who will only purchase Nvidia cards, regardless of price/performance or performance/watt. Even if AMD releases a Polaris card that offers better than GM204/Hawaii-class performance at under 150W, quite a few people will still buy the GTX 970 instead.
If they are going for a HPC Tesla SKU with GP100 (which I agree they will), low yielding, there's gonna be harvested parts and that may well be in a GTX.
What have you done in your life either?? You earned billions,got a nobel prize,etc?? You idolised Bill Gates for years?? Putin?? Some Saudi prince??
He is earning the money not you and idol worshipping somebody who really could not care about any of us does not really mean anything.
Unless he is offering you a job or helping you out in life,how much he earns or not earns has no effect on you or anybody else. You need to worry more about your own life,than worrying about what the next person earns in some faceless corporation.
Unless you are a billionaire yourself,you will always be down the food chain,so better worry about your situation first!:thumbsup:
Last edited:
The same was true of Kepler in 2012. The first publicly available GK110 card was the Tesla K20X in November 2012. That card, despite a launch price of $7,699, only had 2688 of the 2880 shaders active. The same was true of the first-generation Titan, released in February 2013. It wasn't until the Quadro K6000 ($5,265) in July 2013 that a fully-enabled GK110 chip was released on anything, and not until November that it reached a price level that ordinary mortals could afford. (And it only happened that soon because AMD released Hawaii and Nvidia needed something to beat it.)
If Nvidia does manufacture GP100 this year, yields will be incredibly low and even the salvage dies will be going into HPC cards costing thousands of dollars. I expect to see a GP100 Titan in Q1 2017 - that matches up with the past two Titan releases, in Q1 2013 and Q1 2015.
Even if we remove the net conclusion Charlie came to since we all know his bias, there's still some unsettling facts presented there. He's dead on about the tools, there's no disputing that. The faked proto at CES is also a blunder, though not quite wood screw levels this time. It doesn't mean that we should assume AMD is months ahead. That said, I'd be willing to wager that AMD will be first out the gate in the mid and high range sections of the market if only because they've demonstrated functioning hardware already. Will it be better faster, more efficient or more successful? We won't know until both sides have played their hands.
moonbogg
Lifer
- Jan 8, 2011
- 10,732
- 3,449
- 136
Know what Nvidia should do? They should release another 980ti ninja kick to the throat and they should do it first, then release mid range stuff later like they used to do. With such a ridiculous beast Pascal on the market for $650, only to be followed by more awesome mid range goodness later, not a single gamer would consider a single AMD GPU next round. And if they did, its because they are just fans of the company. Nvidia can afford to do this, but AMD simply can't afford another slaughter, especially not one from the start of a new round.
I think that would just about kill AMD for good. I really do.
I think that would just about kill AMD for good. I really do.
I'm sure if they can they will. You might want to read the article though to decide the likelihood of that happening.
^^^This!^^^
Zen will be the horse they hitch their wagon to. They really need for it to succeed. It doesn't have to be the out and out best, that would be expecting too much considering Intel's manufacturing lead and R&D budget, but they need to be able to compete.
AtenRa
Lifer
- Feb 2, 2009
- 14,003
- 3,362
- 136
Perhaps you should take a look at the numbers again because that was not the case.
Fermi was release Q2 2010.
And this is Desktop Discrete GPU market share at Q2 2010.
Almost 50% more in High-End $300+ market (HD5870 MSRP was $379)
Last edited:
ShintaiDK
Lifer
- Apr 22, 2012
- 20,378
- 146
- 106
- Status
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter