NVIDIA Maxwell Speculation Thread

See from 1:07 to 1:40, it makes a lot of sense, you can see why Intel and Nvidia is fanatically increasing their power efficiency. Stop complaining about mobile, mobile focus improves performance. If you go crazy without caring about power usage like what Sunnyvale did, you get Bulldozer & Hawaii.
Here's a old presentation on the subject of power wall and how nvidia planned to conquer it.
http://mediasite.colostate.edu/Medi...aspx?peid=22c9d4e9c8cf474a8f887157581c458a1d#

We should see some interesting things in Maxwell.
 
Last edited by a moderator:
A Chiphell thread that apparently has an overall benchmark of the 750 Ti.

00244120p5ppxpi5lso2lgeuoy.jpg


Since the 760 is slightly ahead of the 660 Ti that would be a pretty big gap between the 760 and the 750 Ti (not that it's unexpected…it's roughly in line with previous performance rumors). I wonder if they are planning to fill the gap with a 700 numbered part.
 
GTX 650 Ti and GTX 650 Ti Boost have the same chip, the GK106, why would this underwhelming performance card GTX 750 Ti offer performance in between those two? This card with GM107 should be called GT 740 or something... :rolleyes:
 
GTX 650 Ti and GTX 650 Ti Boost have the same chip, the GK106, why would this underwhelming performance card GTX 750 Ti offer performance in between those two? This card with GM107 should be called GT 740 or something... :rolleyes:

Why should a card with a chip that is probably smaller outperform a card with a larger GPU? And do you forget that this 750 Ti probably has only 128-bit? Compression can only do so much, how many generations do Nvidia and AMD do this now? 10?
I don't understand this pedantic pessimism towards Nvidia, this negative attitude. The card is supposedly faster than the one it replaces by name by a good chunk. What more do you want? The second coming?
 
Why should a card with a chip that is probably smaller outperform a card with a larger GPU? And do you forget that this 750 Ti probably has only 128-bit? Compression can only do so much, how many generations do Nvidia and AMD do this now? 10?
I don't understand this pedantic pessimism towards Nvidia, this negative attitude. The card is supposedly faster than the one it replaces by name by a good chunk. What more do you want? The second coming?

I think he actually agrees with what you are saying...
 
How do you figure? I see no sarcasm in his post nor any reason for it.

GTX 650 Ti and GTX 650 Ti Boost have the same chip, the GK106, why would this underwhelming performance card GTX 750 Ti offer performance in between those two? This card with GM107 should be called GT 740 or something...

I read that as him saying that expectations are too high for a chip that basically replaces GK107. He is not saying that it is a bad chip, just that nVIDIA is incurring in a mistake by calling it GeForce GTX750Ti.
 
The GTX 550 Ti was a lot slower than the GTX 460 too.
i.e., 50 is a lower number than 60.

650 Ti Boost is more like a variant of GTX 660 (same card with same PCB same everything, just one less SMX enabled)

As boxleitnerb says, it's a 128bit GPU anyway ; it matches the competition closely, Bonaire, bost are the fastest 128bit GPUs ever, after notable "beasts" like 7600GT, Cape Verde. Simply get a 256bit, > 100W GPU if a 128bit, < 100W GPU isn't for you.
 
This card, if indeed based on Maxwell, will just be a guinea pig for Beta testing. By the time Nvidia will release the real chips in Q4 2014, they'll already have all driver issues ironed out and game developers plenty of engine optimizations integrated. This will be obviously very favorable for Nvidia in the inevitable Benchmark wars.
 
I read that as him saying that expectations are too high for a chip that basically replaces GK107. He is not saying that it is a bad chip, just that nVIDIA is incurring in a mistake by calling it GeForce GTX750Ti.

Yup, this is what I don't agree with... Neither I see how exactly the company is positive towards its customers giving in some cases 2 or 3 generations without any performance increases or if present, negligible (at the mobile front is extremely pronounced).

However, I would like to dismiss the word 'expectations' because recently people turned it to a "dirty" word and replace it by 'requirements'

So, basically, if the company is positive towards its customers, there is no reason for anyone to be negative. Which, of course, is not the case ;)
 
Bandwith and process nodes make slow progress, especially the former.
Complaints start to read like those about Intel CPUs, e.g. Haswell is not much faster than Sandy Bridge (needless to say I don't share the view that Intel is intentionally lazy because of the state of competition, or does not improve single-threaded performance on purpose)

Look at laptops, a 35W CPU will suck, even if made of fairies it might be slower than your overclocked i7 920. Likewise a mobile 45W GPU won't be that great next to a desktop card. Though in this role, maybe GM107 will run Crysis 1 well enough.
 
This card, if indeed based on Maxwell, will just be a guinea pig for Beta testing. By the time Nvidia will release the real chips in Q4 2014, they'll already have all driver issues ironed out and game developers plenty of engine optimizations integrated. This will be obviously very favorable for Nvidia in the inevitable Benchmark wars.

Any important/big dev would already have working Maxwell GPUs for about year...
 
I doubt they taped-out that long ago.

If there's a card coming with a Maxwell chip next month, then it probably taped-out around 9-12 months ago. It takes about 2 months to get silicon back post tape-out. Customers probably need the parts 2+ months before launch for all their testing, design, and inventory buildup. So give it another 4-6 months for all the validation and any easy fixes.
 
Any important/big dev would already have working Maxwell GPUs for about year...
Nope. Developers usually don't get CPUs that far ahead let alone GPUs.

If there's a card coming with a Maxwell chip next month, then it probably taped-out around 9-12 months ago. It takes about 2 months to get silicon back post tape-out. Customers probably need the parts 2+ months before launch for all their testing, design, and inventory buildup. So give it another 4-6 months for all the validation and any easy fixes.
At ATI the average time from tape out to production was ~7 months. Fabs take longer now, but if a GPU takes a year from tape out to being on the shelf you've screwed up and had too many bugs. At least for desktop parts. Mobile parts take longer than desktop parts to begin selling to end users.
 
Back
Top