(Ignoring your trademark pessimism.) And your point is... ? That doesn't negate anything I've said. Or is Fiji suddenly no longer a 28nm part?
AMDs slides from financial day expects a 2x performance/watt over current 28nm products. Its also confirmed now that 14nm products is more or less just shrinked 28nm products.
(Ignoring your trademark pessimism.) And your point is... ? That doesn't negate anything I've said. Or is Fiji suddenly no longer a 28nm part?
A lot of what was neutered or castrated to make maxwell efficient is going back in because nvidia is looking to make telsa and quadros out of Pascal.
Keep that in mind as you speculate on its comparitive efficiency.
Pascal is going for half precision in HPC. So your post doesn't make much sense.
Pascal is going for half precision in HPC. So your post doesn't make much sense.
Half vs what level in Maxwell?Transistors will be added that don't help the perf/watt metric for that half precise performance.
Tonga isn't that efficient. And Fiji got HBM working for it. Yet the 980TI got only got more memory but also lower power usage and higher performance.
Welcome to reality. R&D matters :|
What’s perhaps more interesting is what happens to DRAM energy consumption with HBM. As we mentioned before, R9 290X spends 15-20% of its 250W power budget on DRAM, or roughly 38-50W of power on an absolute basis. Meanwhile by AMD’s own reckoning, GDDR5 is good for 10.66GB/sec of bandwidth per watt of power, which works out to 30W+ via that calculation. HBM on the other hand delivers better than 35GB/sec of bandwidth per watt, an immediate 3x gain in energy efficiency per watt.
Of course AMD is then investing some of those gains back in to coming up with more memory bandwidth, so it’s not as simple as saying that memory power consumption has been cut by 70%. Rather given our earlier bandwidth estimate of 512GB/sec of memory bandwidth for a 4 stack configuration, we would be looking at about 15W of power consumption for a 512GB/sec HBM solution, versus 30W+ for a 320GB/sec GDDR5 solution. The end result then points to DRAM power consumption being closer to halved, with AMD saving 15-20W of power.
AMDs slides from financial day expects a 2x performance/watt over current 28nm products. Its also confirmed now that 14nm products is more or less just shrinked 28nm products.
It's funny how quickly you change your tune on the effects of HBM1 on power consumption to fit your narrative. When AMD people talk about how awesome it is, you shoot them down, but now it suddenly makes a huge difference and is the entire cause of the difference between Fiji and Hawaii. Good stuff!
GCN 1.3 as it would be called in the common tongue. R&D mainly went to the CPU division.
It starts to show what is expected, consumer parts will be based on very small dies due to cost structure. At least we get great performance/watt, maybe not so much performance.
Nope. You however try to create a strawman. Get over it and move on. IMC+Memory savings from HBM didn´t solve their core issue.
I was talking about thisI think GCN 4 is something like Maxwell for Kepler.
I expect that performance from core count may change a lot, and we could see that 2816 GCN 4.0 core GPU end up being on par with 4096 GCN 3.0 GPU.
Im wondering if this might have something with cache and delivery time of data to front end, and feeding up all of the cores, and emptying the schedule much faster than before.
It's like $30 more expensive at most for a model that will easily be faster than fury x. It's cheaper than the still unavailable dp to hdmi cord....I will agree that the custom OC GTX 980 Ti models are faster than Fury X but they are also more expensive.
Anyway i was not trying to make it a battle between Fury and GTX980Ti but to make a point that Fury with the same number of ROPs as Hawaii is faster than any Hawaii card at 4K and reaches the same level of performance as reference GTX 980Ti.
I was talking about this
"Jen-Hsun claims that Pascal
will achieve over 2x the performance per watt of Maxwell in Single
Precision General Matrix multiplication" These are architecture improvement not base on new node.
http://www.extremetech.com/gaming/2...mance-gains-from-upcoming-pascal-architecture
The real comparison will be GCN 4.0 and pascal.
People take marketing slides as truth now. Seriously?
No, they compared systems.
They compared systems, but the text is:
"The Polaris card on Med Preset @1080p scored 60fps and consumed 86W with drivers 16.10 beta. The GTX 950 card on Med Preset @1080p scored 60fps and consumed 140W with drivers 359.06."
This is on a system with Core i7 4790K with 4x4GB DDR4 (?).
Maybe they overclocked that GTX 950 or maybe they made up the numbers.
Besides, Core i7 4790K alone can consume over 80W so 86W is clearly not the total system power consumption under load.
They compared systems, but the text is:
"The Polaris card on Med Preset @1080p scored 60fps and consumed 86W with drivers 16.10 beta. The GTX 950 card on Med Preset @1080p scored 60fps and consumed 140W with drivers 359.06."
This is on a system with Core i7 4790K with 4x4GB DDR4 (?).
Maybe they overclocked that GTX 950 or maybe they made up the numbers.
Besides, Core i7 4790K alone can consume over 80W so 86W is clearly not the total system power consumption under load.
anandtech said:To that end RTG also plugged each system into a power meter to measure the total system power at the wall. In the live press demonstration we saw the Polaris system average 88.1W while the GTX 950 system averaged 150W. Meanwhile in RTGs own official lab tests (and used in the slide above) they measured 86W and 140W respectively.
People take marketing slides as truth now. Seriously?