Joined
·
16,904 Posts
[sisoft] Possible GM200 specs leak
1 - 20 of 55 Posts
Quote:
![thumb.gif]()
Quote:
yeah most likely closer to 650 sq mm. must be a beast of a performer.Originally Posted by Clocknut
600-650Mm2 size confirmed. lol
Quote:
GM200 will be most likely GM204 + 35% perf . Nothing scales perfectly. Also clocks will be lower on GM200 vs GM204. The Quadro/Tesla specific features and double precision will add die size too. So expect this chip to be >50% increase over GTX 980. most likely closer to 650 sqmm.Originally Posted by Alatar
Source
Will update the thread when I see an actual article with this. Seems like GM200 will be GM204 + 50% + DP. Will be interesting to see how the final cards clock. The one being tested here is probably an ES for a pro card.
Joined
·
1,216 Posts
3072SP 1400Boost 12GB Ram lol. This would decimate whatever currently out. Although I wonder why 6GHZ vs 7?
Joined
·
16,904 Posts
Quote:
That said with 50% more memory bandwidth, 50% more SMMs and 50% more ROPs I actually do expect to see more than 35% higher perf when OC'd. Probably not at stock. I'd imagine NV is going to go air cooling and reference GM200 will probably have quite bad throttling.
Specs, not perf. Thought it would have been relatively clear since I included DP in there.Originally Posted by raghu78
GM200 will be most likely GM204 + 35% perf . Nothing scales perfectly.
That said with 50% more memory bandwidth, 50% more SMMs and 50% more ROPs I actually do expect to see more than 35% higher perf when OC'd. Probably not at stock. I'd imagine NV is going to go air cooling and reference GM200 will probably have quite bad throttling.
Joined
·
39 Posts
The price tag on this beast is going to be nuts if those numbers are even close to being correct spec-wise. 1500-2000 probably? Core for Core Maxwell is already 40% stronger than Keplar is not to mention this is coming in with 3072 vs 2800s on the GK110, on top of stock clock being 1100 with a boost of 1400 just out of the box. However, this is just speculation we dont really know anything outside of the numbers they are showing.
Quote:
Can you confirm the source for 50% more ROPs. btw your OP says 384 bit 6 Ghz memory. so it remains to be seen how much bandwidth increase is, though I doubt Nvidia would increase shaders by 50% while increasing memory bandwidth by just 30%. This would mean GM200 has less bandwidth per shader than GM204. I can see Nvidia running the Quadro/Tesla models at 6 Ghz while running the Geforce models at 7 Ghz as it does with GK110.Originally Posted by Alatar
Specs, not perf. Thought it would have been relatively clear since I included DP in there.
That said with 50% more memory bandwidth, 50% more SMMs and 50% more ROPs I actually do expect to see more than 35% higher perf when OC'd. Probably not at stock. I'd imagine NV is going to go air cooling and reference GM200 will probably have quite bad throttling.
Joined
·
3,472 Posts
Quote:
flagship Geforce hasnt been revert back to 6GHz b4.
usually the big daddy chip are around 50% bigger spec & I dont think Nvidia would go for anything less than 7Ghz GDDR5.Originally Posted by raghu78
Can you confirm the source for 50% more ROPs. btw your OP says 384 bit 6 Ghz memory. so it remains to be seen how much bandwidth increase is, though I doubt Nvidia would increase shaders by 50% while increasing memory bandwidth by just 30%. This would mean GM200 has less bandwidth per shader than GM204. I can see Nvidia running the Quadro/Tesla models at 6 Ghz while running the Geforce models at 7 Ghz as it does with GK110.
flagship Geforce hasnt been revert back to 6GHz b4.
Joined
·
16,904 Posts
Quote:
Quote:
My guess is that GTC 2015 is the latest when we're going to see a release.
There's no source but it's somewhat obvious. They're going to keep the same ration of MCs and ROPs. I really don't see nvidia going and pulling a tahiti with ROPs, it's going to be exactly like it was with GK104 to GK110 as far as MCs and ROPs go, same ratio. Meaning that GM200 will have 96.Originally Posted by raghu78
Can you confirm the source for 50% more ROPs.
Quote:
The OP specs are probably for an ES of some pro card. 6ghz mem will usually result in better efficiency so it's understandable. But I don't see any reason for 6ghz on the consumer model (at least the top one) when they've already done 7ghz with the 780Ti and Titan Black.
My guess is that GTC 2015 is the latest when we're going to see a release.
Joined
·
16,904 Posts
Quote:
![tongue.gif]()
That said I'm really not looking forward to the market segmentation we're going to probably see. Just please release the full chip with 6GB as the first product and let evga make a classified out of it.
Maybe this will drive a 4K oculus with acceptable frame ratesOriginally Posted by zinfinion
Maybe I can finally play ETS 2 at 60FPS in a city while raining with 400% scaling!!
That said I'm really not looking forward to the market segmentation we're going to probably see. Just please release the full chip with 6GB as the first product and let evga make a classified out of it.
Joined
·
1,114 Posts
Quote:
What do you mean?Originally Posted by Alatar
That said I'm really not looking forward to the market segmentation we're going to probably see. Just please release the full chip with 6GB as the first product and let evga make a classified out of it.
Joined
·
1,233 Posts
Quote:
Just please let MSI make a lighting version (msi no trolling please). They haven't made one since the 780.Originally Posted by Alatar
Quote:
Maybe this will drive a 4K oculus with acceptable frame ratesOriginally Posted by zinfinion
Maybe I can finally play ETS 2 at 60FPS in a city while raining with 400% scaling!!![tongue.gif]()
That said I'm really not looking forward to the market segmentation we're going to probably see. Just please release the full chip with 6GB as the first product and let evga make a classified out of it.
Joined
·
13,043 Posts
3MB L2 cache... Is nVidia building a CPU?
Joined
·
2,927 Posts
still 28nm eh? that's gonna be 4 years on the same process...time's are tough
Joined
·
4,822 Posts
I'm going to wait till this drops to $200
![tongue.gif]()
Joined
·
4,193 Posts
Quote:
So it's fairly possible that the memory is not clocked at 7GHz because of that.
It's not confirmed and they increased the memory cache to increase efficiency for maxwell (and reduces the memory bottleneck).Originally Posted by TheBlademaster01
3MB L2 cache... Is nVidia building a CPU?
So it's fairly possible that the memory is not clocked at 7GHz because of that.
1 - 20 of 55 Posts
Join the discussion
