NVIDIA Kepler speculation thread

By my measurements I also get a 1.25 scale factor, and comparing just the grey parts, the GK208 is about 1/3 of the way from the GF117 to the GK107, and ~1.6% bigger than the GF119. After scaling the GK208 picture in Keynote, I got 1.82" x 1.94" for the GK107 (grey part) and 1.58" x 1.59" for the GK208 (grey part). That would give 84 mm^2 for the GK208. But some time ago I tried to estimate GF119's die size from GF108's using this method (both die sizes are known) and I got 83 mm^2, which is a bit off from the actual die size of 79 mm^2, maybe because the grey stuff is a bit bigger than the actual die. So adjusting for that, I get 80 mm^2 for the GK208. [Remark: This method overestimates the die size of GK107.]
 
Eep, not many ITX PSUs have 8-pin PCI-E power connectors lol. 2x 6pin to 8pin adapter would work though I guess. Would make one hell of a HTPC + Couch gaming setup.
 
As Malo said there are still power considerations to be accounted for. It is impressively small for a card with such performance but it still needs a fairly hefty PSU by SFF standards.

You can pack a lot of power into SFF but obviously there is a cost if you want high performance. I still rate noise above all else in this form factor however because I can game on my PC when I need to. Small, near-silent "good enough" performance is the future of SFF for sure.
 
It's about time high end graphics cards start actually getting smaller in size. Now if only case makers would start dropping 5.25 bays, and we could actually have have competitive options between cases that aren't the size of a small child.
And where would you put your DVD/Blu-ray reader/burner?
 
I would like it this way :)

ibm_ps2mod80.jpg
 
:oops: Why didn't you give it to me instead of throwing?

How do you buy and install a 10-15 GB game? Do you wait all the way to download it from somewhere? :LOL:

At least in case of AAA titles, Steam and to my understanding Origin etc offer pre-loading option if you want to.

Personally I can't even remember when I last installed a game from optical media.
 
GPU's are definitely heading in the same direction as CPU's. Nvidia are championing it, just as Intel did with Sandybridge. Soon, 95% of the gaming market will be buying these 'good enough' upper-mainstream cards for close to traditional high end prices and the traditional high end chips will come out a year late and be priced at twice what they once were.
 
GPU's are definitely heading in the same direction as CPU's. Nvidia are championing it, just as Intel did with Sandybridge. Soon, 95% of the gaming market will be buying these 'good enough' upper-mainstream cards for close to traditional high end prices and the traditional high end chips will come out a year late and be priced at twice what they once were.

CPU's are completely stagnate. At most we get 10-15% performance gains every 12-18 months. GPU's are still getting 60-100% performance gains every 24 months. GK110 is twice as fast as gtx580, and even if you think that is an unfair comparison somehow, gtx680 is still 35% faster than the significantly larger and more power hungry gtx580.

There was a big shift by Nvidia to concentrate on performance per watt after Fermi. They're still going to concentrate on that metric, but since the shift has already occurred the performance penalty has already been suffered. Nvidia already has a much more efficient design to build off moving from Kepler to Maxwell, so sacrificing performance to achieve a better perf/watt metric will be easier to avoid. Maxwell will likely have it's ~300mm^2 die as it's flagship Geforce chip for quite awhile just as Kepler did, but it's also going to be 70-80% faster than GK104 (just like GK104 was over GF114), so we'll have a $500-600 solution 20% faster than Titan by ~Q2 of next year.

So please, unless you were being overly sarcastic, stop complaining or just go by a console and be done with it.
 
GPU's are still getting 60-100% performance gains every 24 months. GK110 is twice as fast as gtx580, and even if you think that is an unfair comparison somehow, gtx680 is still 35% faster than the significantly larger and more power hungry gtx580.

There was a big shift by Nvidia to concentrate on performance per watt after Fermi. They're still going to concentrate on that metric, but since the shift has already occurred the performance penalty has already been suffered. Nvidia already has a much more efficient design to build off moving from Kepler to Maxwell, so sacrificing performance to achieve a better perf/watt metric will be easier to avoid. Maxwell will likely have it's ~300mm^2 die as it's flagship Geforce chip for quite awhile just as Kepler did, but it's also going to be 70-80% faster than GK104 (just like GK104 was over GF114), so we'll have a $500-600 solution 20% faster than Titan by ~Q2 of next year.

I don't see how any of what you said justifies the doubling of price brackets for each performance level.

With CPU's Intel have lowered the ceiling in an effort to make the floor look more appealing, thus people are happier to pay $200-350 for a mainstream chip because it has same image a $500-600 chip had pre-2010.

Nvidia on the other hand have mimicked AMD's R700 approach without also lowering prices accordingly. Of course, this was possible for NV because AMD failed to deliver on performance, so they are almost as much to blame. At least now their prices are a bit more in line with past practises.

Like Intel, however, NV have repacked a 'server' chip for consumers to retain their high end, but now have a mantra of low urgency on releasing such chips. As such they will likely now overlap generations. I do not think that Maxwell GTX780 will overtake Titan myself, and Titan will remain valid on a very small scale until the next, Maxwell-based iteration of it is released - a year after GXT780. Sound familiar?

EDIT: With that said, I want to add that I understand the shifting market. My next build will be mATX, and possibly even an ITX system to supplement it. I don't see how my yearning for a proper high end would suggest that I should just buy a console and be done with it, quite the contrary. Regardless of this view, I am someone who buys in the upper end of the sweet spot, so the strategy of efficiency does cater to me - the high asking prices being coupled with it does not.
 
Last edited by a moderator:
if i cant put an 360mm rad in it .. no way . :D

Now, joke aside, as a little HTPC/gaming plateform on the living room .. ofc thats another story.
 
From Fudzilla: "Desktop Geforce 7 comes by Computex."

They don't have any specifics on the 700 series variants to be launched then. Maybe the "Titan LE" will end up as a 700 series part? The particular rumors about it so far (HWiNFO, alleged GPU-Z and board pictures) don't jive well with a rumored ~Q3 release date in my mind. I would assume a somewhat sooner release just going from the content of the rumors (except release date info).
 
From Fudzilla: "Desktop Geforce 7 comes by Computex."

They don't have any specifics on the 700 series variants to be launched then. Maybe the "Titan LE" will end up as a 700 series part? The particular rumors about it so far (HWiNFO, alleged GPU-Z and board pictures) don't jive well with a rumored ~Q3 release date in my mind. I would assume a somewhat sooner release just going from the content of the rumors (except release date info).

This is my guess. Titan LE or whatever ends up being the flagship 700 series. I don't think nvidia ended up tweaking any of there existing Kepler chips. Instead, I think they will just be utilizing the improves yields with a more mature 28nm node and will give each chip (gk104, gk106, etc.) small core speed bumps while maintaining similar or even tighter TDP's.
 
Back
Top