nVIDIA Maxwell

[boxleitnerb]

I should add that I read somewhere that Nvidia will NOT use those ARM CPUs in every Maxwell product. I believe they might use it in GM110 or not even there but only in Parker (Tegra 6). If only I knew where I read that...I'll go look and report back.

Edit:
Heise.de learned this at GTC 2013. ARM cores in desktop GPUs only from Volta onward.
http://translate.google.de/translat...e-auf-Desktop-GPUs-erst-ab-Volta-1826893.html

 

[tviceman]

It was 2.5yrs from 580 to Titan so does that mean Maxwell will be ready Q3/Q4 of 2015?

Both, the 680 and 770 use more power than a 560 Ti, though.

And Tahiti uses more power than both cayman and cypress, but I wasn't looking at perf/w or even straight up power use. I was only comparing performance of stock cards at 28nm and 40nm.

 

[tviceman]

My gut feeling is telling, that nVIDIA is going to follow the market and optimize Maxwell solely for power-efficiency and thus, making low-power designs their number 1 priority. I would like to be wrong, though.

To the same effect, Nvidia stated they were doing this very thing with Kepler and still managed extraordinary performance gains vs. Fermi chips of similar size. GK106 is more than twice as fast as GF116, and both GK104 and GK110 are up to twice as fast as their counterparts (GF114 and GF110). Even GK107 is 2x GF118 in performance and consumes less power as well.

Maximizing for efficiency is what allowed Kepler to perform so well, IMO.

 

[boxleitnerb]

And Tahiti uses more power than both cayman and cypress, but I wasn't looking at perf/w or even straight up power use. I was only comparing performance of stock cards at 28nm and 40nm.

Okay, but this comparison tells you things about future cards only so long as you can "buy" more performance with more power. Well of course you always can, but Nvidia (or AMD) may not want to.

 

[Erenhardt]

To the same effect, Nvidia stated they were doing this very thing with Kepler and still managed extraordinary performance gains vs. Fermi chips of similar size. GK106 is more than twice as fast as GF116, and both GK104 and GK110 are up to twice as fast as their counterparts (GF114 and GF110). Even GK107 is 2x GF118 in performance and consumes less power as well.

Maximizing for efficiency is what allowed Kepler to perform so well, IMO.

Ohh this compute performance increase. GK104 is FP64 king!

It should be noted that some gain were made by sacrificing some performance in other branches

 

[Subyman]

Quite kidding yourselves. The $499 maxwell will be equal to a Titan. Look at what nvidia pulled with the GK104 launch. They will release just enough performance to push sales, nothing more.

 

[brandon888]

Quite kidding yourselves. The $499 maxwell will be equal to a Titan. Look at what nvidia pulled with the GK104 launch. They will release just enough performance to push sales, nothing more.

so do you think only 15-20% performance leap over GTX 780 ? even worse scenario then from Fermi to kepler ? ... well not sure .... 600$ Maxwell with 30-35% better performance then Titan is more realistic i guess

 

[Erenhardt]

so do you think only 15-20% performance leap over GTX 780 ? even worse scenario then from Fermi to kepler ? ... well not sure .... 600$ Maxwell with 30-35% better performance then Titan is more realistic i guess

for $100 premium over $500 expect <10% performance increase.

 

[rgallant]

Quite kidding yourselves. The $499 maxwell will be equal to a Titan. Look at what nvidia pulled with the GK104 launch. They will release just enough performance to push sales, nothing more.

that's what I'm thinking also
-will nv save the big leap for the following year when the bigger core titan/gtx780 like models are released for the big bucks.

 

[BallaTheFeared]

The only way Nvidia will a) hold back and b) not use their big die for their flag ship is if c) AMD doesn't deliver again.

Optimistic as it sounds, I'm hoping AMD pulls through and comes to 20nm with a better product at better prices, out of the gate.

 

[tnt118]

My memory is failing a little, but will the entire series of 800 cards definitely be Maxwell-based? Or do I remember that sometimes they just rebrand something from a previous line (like a 770 becomes an 860 and is still Keplar)?

 

[Grooveriding]

Quite kidding yourselves. The $499 maxwell will be equal to a Titan. Look at what nvidia pulled with the GK104 launch. They will release just enough performance to push sales, nothing more.

Pretty much this. The first 20nm flagship will be a small die with a 20-30% increase over 780 and an even smaller one over Titan at around $500. It worked once, why on earth would they not do it again ?

 

[tviceman]

Okay, but this comparison tells you things about future cards only so long as you can "buy" more performance with more power. Well of course you always can, but Nvidia (or AMD) may not want to.

Well to that end I agree with what you are saying. I think insofar as TDP is concerned, GK104 is probably about as power hungry as either company would want their second-tier chip to be. I do not foresee GM104 (or Pitcairn's successor) to be significantly more power hungry than GK104.

 

[Siberian]

It was 2.5yrs from 580 to Titan so does that mean Maxwell will be ready Q3/Q4 of 2015?

Titan is only 1 generation ahead of the 580. The GK110 came out last year. Maxwell will be here sooner than you think.

 

[Spjut]

Anyone else expecting DX12 with Maxwell?
In 2014, DX11 will have been out for five years, and Nvidia's earlier major changes in architectures coincided with the new DirectX versions (G80 with DX10, Fermi with DX11).
A new DirectX could also close the efficiency gap between PCs and the PS4/Xbox(there are still complaints about DX11 overhead)

My gut feeling tells me the time is right

 

[Keysplayr]

It will be worse than going from fermi to kepler.

I don't know man, going from a GTX580 to a Titan is a pretty huge honking jump.

 

[Elfear]

Titan is only 1 generation ahead of the 580.

Exactly and that one generation took 2.5yrs to make it to consumers (gaming consumers that is).

Maxwell will be here sooner than you think.

I'm sure we'll see Maxwell before 2015 but I doubt it will be GM110. Probably be a repeat of last year and we'll see GM104 first with 10-20% more performance than Titan and will sell for $550-650.

Going by Nvidia's recent timeline, we won't see GM110 until Q3/Q4 of 2015.

 

[psoomah]

Unless it's HSA compliant, I think Maxwell will be DOA with game developers, they'll want nothing to do with it.

With AMD is bed with game engine developers and publishers providing optimization for AMD hardware (read Kaveri/8xxx) and ease of porting between the various platforms. It's all AMD HSA APUs all the time. The publishers and developers are living the dream.

Upcoming AMD HSA based APUs and GPUs will fit seamlessly into that development ecosystem.

Upcoming Nvidia GPUs won't. Developers are likely to leave Nvidia to it's own devices more and more as time passes.

Crystal Dynamics just gave Nvidia the middle finger. It had to wait until Lara Croft came out to write drivers for it. I doubt is is the only developer to shut Nvidia out. I'm thinking Nvidia is going to face that more and more into the future.

It's going to become the skunk at the garden party.

 

[Siberian]

Exactly and that one generation took 2.5yrs to make it to consumers (gaming consumers that is).



I'm sure we'll see Maxwell before 2015 but I doubt it will be GM110. Probably be a repeat of last year and we'll see GM104 first with 10-20% more performance than Titan and will sell for $550-650.

Going by Nvidia's recent timeline, we won't see GM110 until Q3/Q4 of 2015.

NVIDIA is at least a generation ahead of AMD. With out competition NVIDIA will focus it's high end on HPC and use its midrange on the gamer market. AMD has slipped too much this generation. it may end up being their worst release yet.

 

[BallaTheFeared]

Unless it's HSA compliant, I think Maxwell will be DOA with game developers, they'll want nothing to do with it.

With AMD is bed with game engine developers and publishers providing optimization for AMD hardware (read Kaveri/8xxx) and ease of porting between the various platforms. It's all AMD HSA APUs all the time. The publishers and developers are living the dream.

Upcoming AMD HSA based APUs and GPUs will fit seamlessly into that development ecosystem.

Upcoming Nvidia GPUs won't. Developers are likely to leave Nvidia to it's own devices more and more as time passes.

Crystal Dynamics just gave Nvidia the middle finger. It had to wait until Lara Croft came out to write drivers for it. I doubt is is the only developer to shut Nvidia out. I'm thinking Nvidia is going to face that more and more into the future.

It's going to become the skunk at the garden party.

Do you even believe what you're saying, or is it just for the lulz?

Should I welcome my new x86, ARM, graphics, and GPGPU overlords, or should I prepare the lube aka FX $1000 processors?

 

[RussianSensation]

Optimistic as it sounds, I'm hoping AMD pulls through and comes to 20nm with a better product at better prices, out of the gate.

Why would that happen? AMD just saw NV pricing GTX690 at $1000, then Titan at $1000, then GTX780 at $650; and those cards sold without problems. If their 20nm card beats the Titan, I don't see how AMD will price it below $650.

If Maxwell lives up to the hype, NV should have no problem releasing a 300-350mm2 die chip with 10-20% more performance than the 780. What do you think NV will price such a chip at $350?

You can forget about cheap GPUs going forward. That ship has sailed. I bet NV will do everything it can to position a mid-range GM104 at $500-550 like it did with the GK104. Given the performance lead GTX780 has over the 7970GE and assuming Maxwell is a brand new compute architecture, while AMD will need to reuse GCN, NV should have an even greater lead next generation unless AMD pulls a rabbit out of a hat and goes 450-480mm2 die and GCN 2.0 is way more efficient.

 

[BallaTheFeared]

Optimistic as it sounds

 

[Will Robinson]

Kaveri APU is likely to up the ante on low to medium powered discrete cards and may leave NVDA with the high end market only which means $400 barnstormer cards maybe headed the way of the dodo.:\

 

[3DVagabond]

Kaveri APU is likely to up the ante on low to medium powered discrete cards and may leave NVDA with the high end market only which means $400 barnstormer cards maybe headed the way of the dodo.:\

It's only a matter of time until all dGPU's go the way of the dodo. When AMD APU's get to 20nm production, it will really start to hurt discrete graphics.

 

[JimmiG]

Well it's a new architecture and a new manufacturing process. Should be similar to the jump from G70 to G80 or from GT100 to GF100. Just because Nvidia has been stuck on 28nm for too long doesn't mean they've given up...