That's wrong. 55 nm is a half-node, so you cannot calculate (55*55)/(65*65).
This half-node is 10 per cent linear:
65*0,9 = 58,5
(58,5*58,5)/(65*65) = 0,81
576 sqmm*0,81 = 466,56 sqmm
Nvidia GT300 core: Speculation
- Thread starter Shtal
- Start date
- Status
That's wrong. 55 nm is a half-node, so you cannot calculate (55*55)/(65*65).
This half-node is 10 per cent linear:
65*0,9 = 58,5
(58,5*58,5)/(65*65) = 0,81
576 sqmm*0,81 = 466,56 sqmm
That's wrong. 55 nm is a half-node, so you cannot calculate (55*55)/(65*65).
This half-node is 10 per cent linear:
65*0,9 = 58,5
(58,5*58,5)/(65*65) = 0,81
576 sqmm*0,81 = 466,56 sqmm
Indeed this seems to be the case according to info on the TSMC site.
Slightly misleading to call this 55 nm where it is 58.5 nm.
Also according to TSMC you get 2.35 higher transistor density from 65 to 40 nm, so not the 2.64 you would expect.
So for 2.2 B and 1.3 B transistors you would need about 385 sqmm and 227 sqmm on 40nm.
55nm is the gate length, density of transitors is another problem.
But isn't transistor density also a result of design? So it's feasible that future Nvidia parts will be more dense than the relatively sparse GT200. In any case it's gonna be quite interesting to see which of these companies make better use of the transistor budget afforded to them by 40nm.
That's what has been said before. For example, RV770 fitting more transistors on a smaller die than G92b.
BTW, should we expect significantly higher clocks on 40nm? Or is it all about density? This Altera document speaks pretty ominously about significant increases in leakage with the smaller process.
It seems to me that the combination of more tightly packed transistors + higher leakage current doesn't lend itself well to higher clocks assuming die sizes stay around what they are today.
It seems to me that the combination of more tightly packed transistors + higher leakage current doesn't lend itself well to higher clocks assuming die sizes stay around what they are today.
B3d's past updates on TSMC's process node had numbers showing large density gains, but notably modest power improvements.
GPUs on 40nm are going to face the problem that their transistor budgets are tracking close to doubling over 55nm, but power improvements have not kept pace.
GPUs on 40nm are going to face the problem that their transistor budgets are tracking close to doubling over 55nm, but power improvements have not kept pace.
Megadrive1988
Veteran
How do we know GT300 isn't a DX10 or DX10.1 upgrade of Nvidia's current architecture like G80 ==> GT200 ?
I don't think any of us "knows" that for sure. Rumours place GT3x0 in late 09' and that's the timeframe all D3D11 GPUs are supposed to be projected for. Now with or without delays for A, B or C, is there any particular reason that would justify a <D3D11 GPU from any of the 3 IHVs for late 09?
That actually is a very interesting find! At least given that GPU makers should be aware of the principal issues also.
Jawed
Legend
The redundancy-yield graph, Figure 5, is interesting because at around 250mm2, the improvement is less than 3x early in a process's life and around 40% once the process is mature. This is for a chip that is essentially just a huge uniform grid, I guess, implying that redundancy is far more effective than it would be for the "pools of functionality" found in GPUs.
Altera seems to be running about 1 year ahead of AMD/NVidia GPUs on 40nm. Maybe only 9 months. Still, that implies a significant difference in process maturity for them in comparison with Altera.
Jawed
Altera seems to be running about 1 year ahead of AMD/NVidia GPUs on 40nm. Maybe only 9 months. Still, that implies a significant difference in process maturity for them in comparison with Altera.
Jawed
GT212 will be higher clocked at chip, shader and memory (effective frequency
) than GT200b.So, yes. Although this chip will be more complex than GT200b it will clock higher.
Do you mean GT212? This chip will not have so much transistors.
Yes, I was refering to the 2.2B. Obviously GT212 has more than 1.4B.
An Nv-CJ? *SCNR*
Anyways, in the german 3dc-forums, he (same name at least) claims to know for certain that there are GT300 samples already deployed.
L'Inq:
http://www.forum-3dcenter.org/vbulletin/showthread.php?p=7015310#post7015310
Which of course makes sense, since we all know how early preferred devs get a glimpse at upcoming technology to make sure, nv has something impressive to show at their editor's days. That said, i wouldn't bet any serious money on specs derived from those preview boards - at least not, if i take the stories about 7800 GTX into account...
Uhmm I'm not questioning the possibility at all that he might or might not have some sort of insider knowledge, but what he said so far doesn't take that much more than common logic. Alas for all 2009/10 GPUs if 40nm wouldn't provide higher frequencies with higher transistor counts than today (yet not necessarily as large die area).
Personally I'm always very sceptical with targetted frequencies especially with GPUs up to very close to release. If memory serves well NV was also targetting for a 2.5x ALU:TMU frequency ratio for GT200b@55nm, yet apparently they couldn't reach that one either for the 2nd time in a row. I'd easier play russian roullette than gamble on final core frequencies LOL
- Status