AMD: Volcanic Islands R1100/1200 (8***/9*** series) Speculation/ Rumour Thread

Couldn't NVIDIA also use 250 W as the "marketing" TDP for a GM200 part?

 

181W for be precise .. It is writed in the bios so we cant be wrong lol . ( power limit for turbo is set at 181W on "standard gpu / clock" it goes way higher on the AIB retail ( i have see an MSI with 220W limit )
It is marketing, but the magic number of 165W is the average TDP effectively.

 

Does TDP usually refer to max instantaneous consumption or average?

 

TDP is an average: thermal power is dissipated with a time constant of seconds, so transient power is much less important than instantaneous power.

Dunno why people here keep getting TDP and other things confused.

 

Gigabyte 970 is 250W and 980 300W (The windforce models)

 

Not surprising considering the G1 (970) board's base clock (1178) starts where the reference cards boost sits.
IIRC, the 970 G1 actually has an upper power limit of 280W (112% power)

 

Yeah the +xx% power changes the TDP over what BIOS states and 250*1,12 is 280W

 

My room heats up like a sauna when I game with my 290X, yet remains cool when I use my GTX 980.

 

Ahh..thanks. I had a feeling myself that it was 180W but thought I would double check and since NV said 165W I went with it. Gotta love marketing

Yep if we take 180W then indeed the difference is lower. Usually the bigger die GPUs are clocked a bit lower and have higher power efficiency. But in this case, they may not have much scope for that as the difference between GM204 and GM200 are much lower than the previous generations.

I think they could mange 24 SMM's even with DP FP. They do have ~40% more die area to play around with after all. GM200 should have higher density as well.

I had another idea..I speculate that is also possible that Nvidia has gone for a 512 bit MC. At the launch of Hawaii, AMD mentioned that the area required for a slower 512 bit MC was actually lower than for a high speed 384 bit one. It would also let them stick with 2 MB L2 cache as with 384 bit it would have to be 3MB. Apart from lower die size, the other major benefit of this approach would be the ability to offer large amounts (16 GB) of VRAM for the professional segment.

 

Just for the sake of it, what would the required clocks be if it was Tonga Pro with 1792 stream processors?

 

977MHz.

 

https://finance.yahoo.com/news/amd-firepro-awarded-top-spot-211500900.html
http://www.green500.org/lists/green201411

AMD FirePro S9150 (Hawaii) hits the top of the new Green 500 List, with 5,271.81MFLOPS/W

 

That's impressive. 2.5 TFLOPs DP at 235W goes a long way. The second place system appears to be a pure-x86 system (am I reading that right?) which is extremely impressive too. The NVidia system in 3rd place is outperformed by 19%, which is hardly trivial.

 

The second system seems to use a many-core chip I'd never heard of: http://pezy.co.jp/en/products/pezy-sc.html

 

Some 16 guys company... The chip is 60W (25 DP-GFLOPs/W). More details: http://news.mynavi.jp/articles/2014/09/17/pezy/

 

Supercomputing-specific chips have a habit of disappearing

 

http://wccftech.com/amd-radeon-mxm-gpus-litho-xt-strato-pro/
Next mobile line is apparently named after geospeheres of earth. Any idea which GPUs these would correspond to on desktop side?

 

Thought I could help but no, just looking up their mobile GPUs almost all of them seem to be available in 2/4GB versions excluding the M290X and (unreleased, rumoured?) M295X, both having 4GB.

 

M295X is full Tonga, it is in the new 5K iMac.

 

Actually as far as I know, this hasn't been ever confirmed, only assumed
edit: the full part I mean