05-15-2016, 08:21 PM
Now that there's a rumor that AMD might launch Vega way ahead of schedule (originally early 2017, now October?).. it's probably getting Nvidia ready and prepared for this. Nvidia's plan: Launch Titan Pascal and possibly also GTX 1080 Ti (based on GP100) very shortly after AMD launches Vega. At least NV would paper-launch GP100 without any trouble at all, since GP100 is already in production (quoted from Tomshardware:
Traditionally, the flagship Titan / GTX xx80 Ti cards are clocked maybe 15% higher than Tesla variants. I'd expect it to be clocked almost just as high as GP104 GTX 1080 - at least 1500Mhz core with 1600MHz boost for sure. HBM2 memory might be clocked way higher than 1.4GHz as it's being refined, but I'd try to be conservative with that one - probably not more than 1.6GHz compared to 1.4GHz on the P100.
Notice how the GP100 actually has MORE than 2x as many trannies as GTX 1080?
![[Image: nvidia-gtx-1080-gpuz-official.jpg]](http://cdn.videocardz.com/1/2016/05/nvidia-gtx-1080-gpuz-official.jpg)
Only 7200 million trannies, whereas GP100 has 15.3 billion.
http://www.pcgamer.com/nvidia-pascal-p10...deep-dive/
What's interesting is that GP100 can do twice as many half-precision (FP16) textures as single-precision FP32 (unlike previous generations). Would that really help with games if they have simpler operations that could use half-precision?
http://wccftech.com/geforce-gtx-1080-blo...d-edition/
2.5GHz!?!??? This is insane, and unbelievable!
Quote:The Tesla P100 is in volume production today and will ship “soon." Jen-Hsun Huang said that really means it will show up in the cloud first, and ship via OEMs in Q1 2017.
Traditionally, the flagship Titan / GTX xx80 Ti cards are clocked maybe 15% higher than Tesla variants. I'd expect it to be clocked almost just as high as GP104 GTX 1080 - at least 1500Mhz core with 1600MHz boost for sure. HBM2 memory might be clocked way higher than 1.4GHz as it's being refined, but I'd try to be conservative with that one - probably not more than 1.6GHz compared to 1.4GHz on the P100.
Notice how the GP100 actually has MORE than 2x as many trannies as GTX 1080?
![[Image: nvidia-gtx-1080-gpuz-official.jpg]](http://cdn.videocardz.com/1/2016/05/nvidia-gtx-1080-gpuz-official.jpg)
Only 7200 million trannies, whereas GP100 has 15.3 billion.
http://www.pcgamer.com/nvidia-pascal-p10...deep-dive/
What's interesting is that GP100 can do twice as many half-precision (FP16) textures as single-precision FP32 (unlike previous generations). Would that really help with games if they have simpler operations that could use half-precision?
Quote:The GP100 based P100 has 3584 CUDA Cores. The chip itself has a turbo clock of 1480 Mhz. The total peak single precision performance of the P100 GPU therefore is 10.6 TFLOPs (3584 CUDA Cores x 2 Operations Per Second x 1480 Megahertz). Since we are talking about the exact same Pascal architecture here, we can directly compare performance of the GP104 based GTX 1080 to the P100.
Quote:The GTX 1080 on the other hand only has 2560 CUDA Cores. However, the card can reach clock rates of 2.1 Ghz conservatively and 2.5 Ghz on a liquid cooled setup. At a clock rate of 2.1 Ghz, the GTX 1080 is actually 10.7 TFLOPs of single precision performance! At a clock rate of 2.5 Ghz, the GTX 1080 will blow past the P100 GPU giving performance levels of 12.8 TFLOPs! That is an absolutely insane figure and roughly 20% faster than the P100. To put this into perspective that is almost twice the theoretical single precision performance of the GTX 980 Ti.
http://wccftech.com/geforce-gtx-1080-blo...d-edition/
2.5GHz!?!??? This is insane, and unbelievable!

