Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
A rebrand = using the same silicon with a different name. New features, hardware features that use different silicon doesn't equate to a rebrand, ever.
Every single generation we've ever had you get a new card with a similar amount of cores to the previous high end(okay some exceptions but boo hoo).
IE if this was on 20nm, there would be a card with around 280x performance but roughly half the size(due to the process) and roughly half the power. There would be another card that is roughly the speed of a 290x but half the size/power, then you would only get one new high end card that actually goes and adds 70-80% performance.
New generation, different architecture(even if it's similar/a progression is not the same), different hardware features is not a rebrand.
There won't be a 285x with 384bit bus and a 285 based on the same core with a 256bit bus... it's just not going to happen. You also aren't going to intentionally have one card that has 12% less shaders or so with 33% less bandwidth or conversely a card with 15% more shaders and 50% more bandwidth.
AMD has, from my recollection rarely if ever done cut down memory bus parts. Nvidia has more form for cutting down bandwidth and it likely saves them a little power efficiency.
on top of that firepro/compute/professional cards bandwidth is very important. You'd be much more likely to see cut down shader + full bus parts than cut down bus + full shader parts.
Tonga appears to be targeted at the 150W ballpark, 384bit bus simply uses up too much power to hit such a target. Big bus + midrange part is never done, ever, because midrange parts need to be lower cost, lower die size and lower power usage because they have to fit in Dell machines that cost less and they have smaller PSU's, less cooling, etc, etc.
Low end 64-128bit, mid range, 256bit, high end 384-512bit. You don't make a midrange size die, with midrange size power usage, midrange shader count...... and give it a high end bus.
