Radeon RX 5700 Thread

[Spyhop]

Remember the days when you could get a card that was 50% faster, now its all x faster per clock, x faster per watt x faster per mm squared bla bla. Always some random caveat that gets slapped on the end so its not clear cut til reviews hit.

Only way to get noteworthy bang-per-buck these days is strategically buying used.

I'm mainly interested in Navi for what it does to the Vega market!

 

[drunkenmaster]

RTX 2070 with no spec changes on 7nm would be 85w - 7nm offers 50% lower power draw

The current Turing cards done on 7nm you'd be able to put a full size desktop RTX 2080 into those thin laptops and possibly beat a Radeon 7 desktop card lol

If all the power came from the die, still no, because I/O on the die still isn't scaling as well. The memory which is a significant chunk of power and the high bandwidth requirements means a lot of I/O power going off die still. Neither of those scale, one at all and one not well below 14nm just because the main chip is 7nm.

Put this way, a 150W tdp gpu is likely using anywhere from 80-100W on the actual gpu die. If you manage 50% less power at the same clocks and performance on a die half the size then you reduce that to 40-50W, but still add the 50W for memory, mosfets and a good chunk of I/O power which won't change.

CPUs are different in that the cpu tdp is on package only, it doesn't account for memory or the board power from mosfets/etc.

HBM is a huge power saving, which could mean taking for instance a Vega 64, that memory usage on board actually goes from say 30W to 50-60W for less bandwidth but from gddr6. That would make a much worse difference going from an HBM card to a gddr5/5x/6 card than from a gddr based card to a new card with a 7nm core and the same memory.

 

[Poneros]

Remember the days when you could get a card that was 50% faster, now its all x faster per clock, x faster per watt x faster per mm squared bla bla. Always some random caveat that gets slapped on the end so its not clear cut til reviews hit.

Soon to be: faster when RT'ing. No other big jumps will exist for "traditional" rendering.

 

[Beren]

It is an interesting Die Size at 275mm2 - Polaris is 232mm2 on 14nm - Vega VII is 331mm2

This is going to be a very significant step up from Polaris for the mid range, we just need to understand what clocks they can hit on the process.