GPU | Branding | Cores | % of Top die | Bus | Bandwidth | MSRP |
---|
GP102 | AD102 | 3840 | 100% | -- | -- | 2017 had Titan xP release with the full die at $1,200 |
GP102 | Titan X | 3584 | 93.3% | 384 bit | 480.4 GBps | $1,200 |
GP102 | 1080Ti | 3584 | 93.3% | 352 bit | 484.4 GBps | $699 (2017 release) |
GP104 | 1080 | 2560 | 66.67% | 256 bit | 320.3 GBps | $599 |
GP104 | 1070 | 1920 | 50% | 256 bit | 252.3 GBps | $379 |
GP106 | 1060 6GB | 1280 | 33.3% | 192 bit | 192.2 GBPs | $299 |
If we add in Pascal as well we get a similar story. 1060 has 33.3% of the the full die shaders and matched the 980. The 1070 is around half the shaders of the full die variant and the 1080 was around 66%. Titan X is a bit better than 4090 but the refresh 1080Ti took that spot at a lower price point and Titan xP took the $1,200 spot in 2017.
If that had been followed you would get something like this.
Branding | Cores | % of Top Die | Bus | MSRP | Performance |
---|
4090 | 16384 | 88.9% | 384 bit | $1,599 | 4090 |
4080 | 12288 | 66.67% | 256 bit | $799 | Between current 4080 and 4090. |
4070 | 9728 | 52.8% | 256 bit | $599 | about current 4080 |
4060 | 6144 | 33.3% | 192 bit | $399 | about 3080 / current 4070 |
That stack would be in line with the historical stacks and pricing is basically 2000 series pricing which was a bit higher than 3000 series at each tier and allows for inflation / higher costs while not totally screwing the customer.