6G is still not enough? How can they be so thirst of the amount of Global Memory. It looks like Quadros do need so much. But talking about Tesla..actually most GPGPU HPC use C2050 which has 3GB memory instead of C2070 which has 6GB...You're probably thinking in gaming terms
The high-end Teslas and Quadros have 6 GB, so there seems to be some demand for as-large-as-possible framebuffers in HPC/workstation space. A 512 bit interface would increase that to 8 GB.
And, I'm wondering, If they really use 512bit, the "low-end" compared to the high-end Teslas, which I mean the high-end desktop users, will feel sad that a lot of power are consumpted by something they don't need.
Indeed, there are very few games that current GPUs can handle with crancked up AA. (32xRGSSAA)
I'm sure that however highr bandwidth the GPUs are given the developers find a way to exploit, same goes to an amount of memory.
How can you keep on making the same huge chips, within the same power budget that is, without improving efficiency?
UniversalTruth
Veteran
Yes, nv talks about that because they will launch Fermi shrinks for notebooks soon enough.
The guys here (because they are enthusiasts) keep talking about something that is at least 6 months in the future. Luckily, those huge chips need a mature process.
Correctly. If Kepler can provide better performance pre transistor, the performance pre power can increase a lot and that's definately good for mobile sector which , nowadays , just used the chips which can be regard as some kind of cut-off verson of the high-end desktop chips.
It seems it's so hard. But it is indeed useful for notebook GPU.
Only by changing 40nm to 28nm is not enough.
You often need to spend transistors to improve perf/w so perf/transistor probably isn't high on anyone's priority list.
It's is on Intel's list. And, Sandy Bridge suceeed in it.
But...what I'm rethinking now is that maybe partly you're right, it is not that useful for top-end because notebook GPUs can have a higher frequency or an larger die through new tecnology.
Because they have been doing pretty much the same tasks for the last decade, you slowly learn how to take what you have and become more efficient.
Compare that to GPUs that have been "slowly" moving away from fixed function and increasing have more tasks and features in new products while still keeping the same performance or better with previous tasks.
DegustatoR
Legend
Not really...
UniversalTruth
Veteran
I recall that all NV13***(something) chips are 28 nm optical shrinks of Fermi chips.
http://www.computerbase.de/news/2011-11/details-zu-anstehenden-28-nm-grafikloesungen-von-nvidia/
You don't think so. Why?
http://www.computerbase.de/news/2011-11/details-zu-anstehenden-28-nm-grafikloesungen-von-nvidia/
You don't think so. Why?
DegustatoR
Legend
Because there is zero point in doing Fermi 28nm shrinks for December when you have a full Kepler line-up to launch from January to June. I think that we may see one Fermi on 28nm but all the rest will be Keplers.
Maybe as a safety net to be at least able to wage a price war if AMD is on time with 28nm next-gen and Kepler might be late? Remember the GT200 disaster end of 2009 where Nvidia stopped production of 55nm GT200 chips and had nothing to compete from lower midrange upwards.
DegustatoR
Legend
1. It's not late.
2. Doing Fermi shrinks on 28nm is as time consuming as doing Keplers on 28nm thus if you'd do Fermi shrinks then it'd be a given that Keplers will be late.
That had much more to do with them doing GF100 first and GF104 later than not doing Tesla shrinks on 40nm (although you could say that they did do them in the form of GT21x). With Keplers it'll be different since they're doing low and mid first and high-end later.
