uhm, the Vega Core specs don't add up- it's got higher clockspeed and core count than the Eclipse but lower performance numbers?
I think the leaks we have seen suggest the core should run around 1200mhz, giving GTX 1070 level perf (unknown number of shaders, though I suspect it'll be somewhat cut down).
Does this mean they are going with the slower HBM2 I thought it was 512Gb/s ?? I was expecting high clock speeds alright... the 1200mhz was only early engineering sample. An the Vega architecture is designed to run faster than Polaris. But I think they were supposed to be using stacks of 256Gb/s HBM2.
Well it looks like 256gb/s HBM2 isn't available yet, the current HBM2 runs at 1.6ghz and not 2ghz required for that 256gb/s speed.
Still, 400+ gb/s bandwidth should be ample (just look at a GTX 1080 vs a Fury X- the latter has 512 gb/s of bandwidth vs 320 gb/s on the 1080, yet the 1080 is a much faster card. Yes high bandwidth helps however it has to be balanced against the rest of the card. Fury had way more bandwidth than it really needed given it's shader performance).
As for the speeds- what I'm saying is for the *bottom model* (the cheapest 'Core') vega is likely to use a lower clockspeed (e.g. 1200mhz) and that will go up for higher sku. That is the point of the table- they are suggesting that Vega will have 3 gaming models aimed at different price / performance points. The issue I have is the table also suggests that the entry level 'core' that is intended only to go up against the GTX 1070 is faster than the faster, more expensive part aimed at the GTX 1080. That doesn't make sense.
To my mind the specs would be something like this:
- Vega Core: 1200 mhz clock, Circa 3500 shaders, 400gb/s memory bandwidth
- Vega Eclipse: 1400 mhz clock, Circa 3500 shaders, 480gb/s memory bandwidth (based on the specs of the recently announced FE)
- Vega Nova: 1500+mhz clock, full 4096 shaders, 480gb/s memory bandwidth