Nvidia fell for an orchestrated pricing trap says AMD exec

Well for Ryzen 3 launch as well as Navi - a lot of chips will be pushed out of TSMC 7nm. To push top to bottom on a new process all at the same time I think would be asking too much. When will 5800 series hit? As soon as possible and that probably depends more on TSMC capacity and maturity of the process.

Navi 5700 XT looks to be pushing a little over 200w (under rated at least my sample). Now a 384 bit memory bus - 12gb of ram version 64, 60,56? CU's or 4096 RNDA shaders, performance 1.3x-1.5x faster than the 5700 XT I see possible with a 300w or less part. Who doesn't think AMD does not have this already figured out and testing? RNDA works and looks rather potent, way better than expected.

Now when Ryzen 3 sells or the inrush slows down a little (it will be hot rest of year but should slow down), the big orders for EPYC are well understood. When do you launch small Navi and big? Drivers are maturing, developers have samples, TSMC capacity and maturity should go up -> I think rather soon as in end of this year early next. Will AMD use Samsung for big Navi? or anything else? Plus Google will be or has ordered a hell a lot of specialize Stadia hardware from AMD that has to be delivered this year. If anything AMD is probably held up by process availability. To me Navi looks like Maxwell when it hit. For Nvidia it totally changed their standings to a clear dominant position - Navi or RNDA may do the same for AMD.
 
To me Navi looks like Maxwell when it hit. For Nvidia it totally changed their standings to a clear dominant position - Navi or RNDA may do the same for AMD.

It may, but it is still an uphill battle for AMD as Nvidia won't just set around like Intel. The big if at the moment is we just don't know how big of an upgrade when Nvidia release Ampere. Recent history have shown whenever Nvidia changes process node, there is a big increase in performance for them.
 
It may, but it is still an uphill battle for AMD as Nvidia won't just set around like Intel. The big if at the moment is we just don't know how big of an upgrade when Nvidia release Ampere. Recent history have shown whenever Nvidia changes process node, there is a big increase in performance for them.
The perf/watt edge still goes to Nvidia Turing for the most part and that is on a 12nm process but a much larger die. AMD could have traded some die space as in larger to up the clocks some which larger die would allow better cooling. With a better power budget allows NVidia to make a chip as big as needed to be to maintain top performance. Memory bandwidth looks like AMD could improve upon for more performance, faster DDR 6, wider bus etc. Slow down Navi GPU a little but increase memory bandwidth and the Perf/Watt should improve.

I still think there is a lot of maturing yet for Navi for developers and even in the drivers - so 5%-10% is probably not unreasonable here. My sample undervolts well too, so once process maturity occurs, the voltage and thus power will most likely be lowered for Navi in general. With AMD what seems like unneeded overvolted GPU's hints at AMD accepting lower quality chips to be accepted that need the higher voltage to work in spec. Better binning and skews, Nvidia had A silicon for Turing faster parts (think they got rid of that now because of process maturity) allowed slower spec ones to still be sold while AMD appears to bin less and rely on higher voltages to ensure they work.

Yes in the GPU space, uphill battle against Nvidia but also Intel is like waking up a sleeping giant. All in all we all benefit when both companies are competing fiercely against each other.
 
The perf/watt edge still goes to Nvidia Turing for the most part and that is on a 12nm process but a much larger die. AMD could have traded some die space as in larger to up the clocks some which larger die would allow better cooling. With a better power budget allows NVidia to make a chip as big as needed to be to maintain top performance. Memory bandwidth looks like AMD could improve upon for more performance, faster DDR 6, wider bus etc. Slow down Navi GPU a little but increase memory bandwidth and the Perf/Watt should improve.

I still think there is a lot of maturing yet for Navi for developers and even in the drivers - so 5%-10% is probably not unreasonable here. My sample undervolts well too, so once process maturity occurs, the voltage and thus power will most likely be lowered for Navi in general. With AMD what seems like unneeded overvolted GPU's hints at AMD accepting lower quality chips to be accepted that need the higher voltage to work in spec. Better binning and skews, Nvidia had A silicon for Turing faster parts (think they got rid of that now because of process maturity) allowed slower spec ones to still be sold while AMD appears to bin less and rely on higher voltages to ensure they work.

Yes in the GPU space, uphill battle against Nvidia but also Intel is like waking up a sleeping giant. All in all we all benefit when both companies are competing fiercely against each other.

Agree, AMD could still do what you suggest here, trade some die space to clock higher. I feel like overvolting have been MO for both AMD and Nvidia (Nvidia also benefits from under volting too) to ensure majority of their chip gets sold.
 
Back
Top