NVIDIA Maxwell Speculation Thread

Notebookcheck has listed quite some Notebooks with it (GT840M specifically). For a chip which is available for quite a while it is indeed still quite a mystery. Die size, transistor count? Make your guess (mine would be very slightly above gk208 for size). Would it even support gddr5? No idea.
I estimate 83 mm^2. (I made an estimate of 74-77 mm^2 a while ago but I don't think that's accurate after looking at other chip packages.) Do the other small GPUs like GF117 support GDDR5? Since GM108 seems similar to those, I don't think we'll see a part with GDDR5 actually come to market even if the chip itself supported it.
 
I estimate 83 mm^2. (I made an estimate of 74-77 mm^2 a while ago but I don't think that's accurate after looking at other chip packages.) Do the other small GPUs like GF117 support GDDR5? Since GM108 seems similar to those, I don't think we'll see a part with GDDR5 actually come to market even if the chip itself supported it.
Well my preferred point of comparison would be gk208 (die size 87mm² according to some sources, though others claim 79mm²). Which definitely supports gddr5 though afaik not in a mobile configuration. Well if gm108 and gk208 are the same size that's quite impressive for gm108.
I think gf117 (just like gf108) was ddr3 only, but didn't need gddr5 because it had a 128bit memory interface. Don't quote me on that though...
 
Which begs the question. Why do GK208 and GM108 co-exist? They're basically identical.
Well even GF117 is still in the lineup, though I have no idea if the die size is the same for that too (the number in tpu's database seems obviously wrong). 820m - gf117, 825m - gk208, 830m/840m - 2/3 smm gm108.
I have no idea neither why nvidia still sells that old junk if gm108 should be no more expensive to produce.
 
Is there even 100% confirmation that GM108 exists and is not just a code name for GM107 with fused off parts? I think that if there were multiple maxwell chips, we'd actually know that for sure and tech sites like Anandtech would have reported on it. Further, Nvidia continues to release Kepler based 700 series desktop parts (last month and also this week), but no GM108??? Makes no sense.

I don't think GM108 exists as a different chip than what GM107 is.
 
Is there even 100% confirmation that GM108 exists and is not just a code name for GM107 with fused off parts? I think that if there were multiple maxwell chips, we'd actually know that for sure and tech sites like Anandtech would have reported on it. Further, Nvidia continues to release Kepler based 700 series desktop parts (last month and also this week), but no GM108??? Makes no sense.

I don't think GM108 exists as a different chip than what GM107 is.


You're wrong on this one.
 
CUDA and die measuring proves the existence.

Which begs the question. Why do GK208 and GM108 co-exist? They're basically identical.
Maybe GM108 has no diplay pipelines like GF117 and can only be used in platforms with IGP?


The future-nvidia-gpu speculative parts in the paper don't seem to contain any insider info, just conjecture based on the GTC slideware...
According to GTC the difference between Pascal and Maxwell should be ~80% and not this ~40% like in this paper. And I do not remember on a GTC, where they talked about GK210. ;)
 
You're wrong on this one.

No major tech site has mentioned it's existence.

Here is the entire 700 series desktop product lineup: http://www.geforce.com/hardware/desktop-gpus/geforce-gt-720

Only two mentions of Maxwell : GTX 750 and GTX 750 TI.

The 740 is Kepler GK107, the 730 is a mix of Kepler GK107 and Fermi GF108, and the 720 is Kepler GK208.

No GM108. Where are these die measurements and die labeling pics? If there are supposed measurements, I think it's of GK208.
 
But GK208 was a Tegra testbed / mobile part and it just got a desktop release.

Perhaps GM108 does exist in more than just code name, but the fact that no major tech site has mentioned it, let alone the fact that Nvidia continues to release desktop parts based on older architectures that are equally slow (or slower) instead of this makes no sense.
 
But GK208 was a Tegra testbed / mobile part and it just got a desktop release. Perhaps GM108 does exist in more than just code name, but the fact that no major tech site has mentioned it, let alone the fact that Nvidia continues to release desktop parts based on older architectures that are equally slow (or slower) instead of this makes no sense.
I could have sworn that GK208 was released for desktop quite some time ago (a year?). There was a decent amount of excitement about it in the CUDA world since it has some of the advanced compute features of GK110 that are not present in other Kepler chips.

I don't really believe in real chips being a testbed. If they exist, it's because they are considered valuable as a standalone product. Nobody pay $xM for tape-out and qualification unless it has value to exist on its own. The Kepler architecture has been on the market for a long time before GK208. Whatever changes were made could have been simulated without the need for real silicon.
 
Perhaps GM108 does exist in more than just code name, but the fact that no major tech site has mentioned it, let alone the fact that Nvidia continues to release desktop parts based on older architectures that are equally slow (or slower) instead of this makes no sense.
Ever seen some hard facts on GF117/GF119? Just the same story.
 
Alright it's possible, I just find it highly unusual they're still producing other chips that essentially overlap with GM208.
 
Alright it's possible, I just find it highly unusual they're still producing other chips that essentially overlap with GM208.
I don't think it is all that strange: the performance of business/home non-gaming laptops has been fine for years (I have a two of 5yo MacBooks that are going strong just fine). I can imagine that there must be plenty of laptop producers who rather keep on selling the same thing than develop a new product.
 
Maybe GM108 has no diplay pipelines like GF117 and can only be used in platforms with IGP?
According to nvidia specs, it has display outputs (and video accel) - I guess could always be some copy / paste bug. I didn't actually realize nvidia has official pictures of those chips - http://www.geforce.com/hardware/notebook-gpus/geforce-830m/product-images, interestingly this actually looks pin-compatible with gm107, despite the latter obviously featuring a 128bit bus. Looks indeed quite small there though I can't really see the die edges.
 
Back
Top