NVIDIA Maxwell Speculation Thread

[Ninjaprime]

Hmm, interesting data from TPU lol

http://www.techpowerup.com/gpudb/2035/NVIDIA_GeForce_GTX_880.html

They even have release schedule and specs for Q4 2013 AMD cards. lol

http://www.techpowerup.com/gpudb/

Hmm, still sticked with GDDR5, any ideas for some type of progress with memory type?

5120MB of RAM ehh? Sounds legit!

 

[Blazkowicz]

It's misleading, it's really 4096MB, but using ten bit bytes, not bytes made of eight bits.
This makes it much more optimized for 10 bit per color displays, also FP precision is extended to 80bits. This is undebatable. I have JHH confirming it to me personnally in an encrypted IRC chat.

 

[DavidGraham]

Though it is merely a joke , the idea is not bad at all!

 

[DSC]

Maxwell references in CUDA 5.0 drivers.

http://www.geeks3d.com/forums/index.php/topic,3011.0.html

_GetMaxSharedMemory_maxwell _GetMaxLocalMemory_maxwell _GetConstantBankSize_maxwell _GetMaxTexturesPerEntry_maxwell _GetMaxSamplersPerEntry_maxwell _GetMaxSurfacesPerEntry_maxwell _GetParamConstBank_maxwell _GetParamConstBase_maxwell _GetMaxParamConstSize_maxwell _GetParamSharedBase_maxwell _GetMaxParamSharedSize_maxwell _GetEnvregConstBank_maxwell _GetOCGConstBank_maxwell _GetDriverConstantBank_maxwell _GetGlobalRelocConstBank_maxwell _GetConstBackingStoreBank_maxwell _GetIsUserConstBank_maxwell _GetIsGlobalConstBank_maxwell _GetEnvregConstBase_maxwell _GetConstBackingStoreBase_maxwell _GetUsesBindlessFormat_maxwell _GetBindlessTextureBank_maxwell _GetBindlessSurfaceBank_maxwell _GetUsesLayeredSurfaceEmulation_maxwell _GetIsQueryDescBank_maxwell _GetTextureQueryDescriptorStorage_maxwell _GetGlobalQueryDescTableBank_maxwell _GetGlobalTextureQueryDescTableBase_maxwell _GetGlobalSamplerQueryDescTableBase_maxwell _GetBindlessTextureImmediateOffset_maxwell _GetBindlessSurfaceHeaderSize_maxwell _GetBindlessTextureHeaderIndexSize_maxwell _GetUsesIndirectionForGlobals_maxwell _GetGlobalMemorySegment_maxwell _GetConstPointerKind_maxwell _GetPTXCbankForHWCbank_maxwell _GetHWStorageForPtxCbank_maxwell _GetSurfaceQueryDescriptorStorage_maxwell

 

[A1xLLcqAgt0qc2RyMz0y]

Maxwell references in CUDA 5.0 drivers.

http://www.geeks3d.com/forums/index.php/topic,3011.0.html

Well Maxwell is a 2014 product and March 2014 is 2 years since Kepler.

Having references now in CUDA to test early silicon seems likely.

 

[tviceman]

Well Maxwell is a 2014 product and March 2014 is 2 years since Kepler.

Having references now in CUDA to test early silicon seems likely.

I would think that if they had 20nm Maxwell silicon now, they are way ahead of schedule, The current rumors (may have started with Charlie) is that Apple bought all the first run 20nm wafers, and other customers won't get access to 20nm for the first two months or so. I don't know if this affects tape outs (I doubt it), but it would surely affect production and release dates.

 

[Ailuros]

I could think of quite a few more advantages for both AMD and NVIDIA to make their first next generation steps on a mature 28nm process, than to wait WTF TSMC can have reasonable quantities at reasonable yields for 20HP.

I don't know if it has anything to do with Apple but I severely doubt it also; if it really should be the case that Apple has pre-booked large capacities then the disadvantage could go to someone like Qualcomm for example. After some point those Apple related fairy tales begin to get tiresome; how long are we reading now that Samsung has supposedly given them the boot or that they'll leave Samsung over night or whatever else nonsense? Of course will there be changes but it's always better to make those step by step for all interested parties, just because the volumes are too big and who guarantees that Apple doesn't want to dual source in the end when it comes to manufacturing?

It's far more easier to believe that 20HP just stinks and will be far too expensive and risky to use at its kickstart than any other fairy tales some would want us to believe.

 

[Alexko]

Yeah, and NVIDIA did give an entire presentation to complain about 20nm, remember?

http://www.extremetech.com/computin...y-with-tsmc-claims-22nm-essentially-worthless

 

[A1xLLcqAgt0qc2RyMz0y]

I would think that if they had 20nm Maxwell silicon now, they are way ahead of schedule

Who says prototypes for functional testing have to be built on 20nm.

The current rumors (may have started with Charlie) is that Apple bought all the first run 20nm wafers, and other customers won't get access to 20nm for the first two months or so.

And you believe Charlie hook, line and sinker?

 

[A1xLLcqAgt0qc2RyMz0y]

Yeah, and NVIDIA did give an entire presentation to complain about 20nm, remember?

http://www.extremetech.com/computin...y-with-tsmc-claims-22nm-essentially-worthless

Which AMD agreed with.

Further evidence for the accuracy of NV’s presentation comes, ironically, from the company’s primary GPU competitor. At AMD’s Financial Analyst Day, CEO Rory Read made a point of saying that the company no longer intends to aggressively transition to new process nodes given the diminishing marginal returns from doing so.

 

[tviceman]

Who says prototypes for functional testing have to be built on 20nm.

And you believe Charlie hook, line and sinker?

True on the first point, and heck no on the second point. I wad just stating where I read the original rumor.

 

[A1xLLcqAgt0qc2RyMz0y]

Interesting quote on energy efficiency

the energy efficiency of Maxwell, it is going to crush Kepler. […] We know exactly how to measure it now and we know what it means to be good

http://www.xbitlabs.com/news/graphi...well_Architecture_Will_Break_New_Grounds.html

 

[FUDie]

http://www.xbitlabs.com/news/graphi...well_Architecture_Will_Break_New_Grounds.html

That's a quote from Jen-sun. What else do you expect him to say? Maxwell is going to suck relative to Kepler?

-FUDie

 

[jlippo]

That's a quote from Jen-sun. What else do you expect him to say? Maxwell is going to suck relative to Kepler?

It certainly should be more power efficient, it's pretty much all they have been talking about for past few years.
What interests me most is the programmability part, which I believe is about project Denver and the ARM cores it bring to table.

 

[jimbo75]

http://www.xbitlabs.com/news/graphi...well_Architecture_Will_Break_New_Grounds.html

Two things basically - Maxwell is 20nm and it really better crush Kepler in energy efficiency.

And typically when Nvidia has nothing much to brag about in the present, they talk up their future stuff even more than usual. I'm sure Maxwell and Tegra 5 will be improvements but to be frank they'd better be.

 

[Novum]

How exactly is Kepler not competitive at the moment?

 

[trinibwoy]

And typically when Nvidia has nothing much to brag about in the present, they talk up their future stuff even more than usual.

Tegra is floundering but Geforce and Tesla are currently at the top of their respective games.

 

[jlippo]

And typically when Nvidia has nothing much to brag about in the present, they talk up their future stuff even more than usual.

More than usual?
They have been giving presentations on their future projects for years, including Maxwell and beyond.

 

[pjbliverpool]

Unified virtual memory support and the integrated ARM cores both sound very interesting. Could this be a potential answer to the new consoles HSA abilities?

 

[lanek]

Unified virtual memory support and the integrated ARM cores both sound very interesting. Could this be a potential answer to the new consoles HSA abilities?

In reality, unified Virtual memory is allready part of AMD Sea Island, GCN is in reality allready capable of it but it will be for the next generation of GCN cards. It have been moved as part of HSA functionnalities then.
(Im pretty sure if i can get old roadmamps, you was seen unified virtual memory allready part of actual GCN SouthIsland and coherency with system added on Sea Island )

As for the ARM cores, it is a bit unclear now if all maxwell cards will get it, or only first on specific chips. But anyway somewhere using ARM permit to Nvidia to have access to HSA features without been inside HSA fundation.
This is typically the strange type of thing we see from Nvidia time to time, all mobiles SOC, hardware conceptors ( ARM, Qualcomm, Samsung, etc ) are part of HSA; but Nvidia who want make his way on the mobile market, stay outside of it.