Silverforce11
Lifer
- Feb 19, 2009
- 10,457
- 10
- 76
Fiji Pro at $499 would be killer if its ~10% slower than XT version (~$699). The WCE can charge a premium but bang for buck, definitely the Fiji Pro. Let's hope its priced well.
Fiji Pro at $499 would be killer if its ~10% slower than XT version (~$699). The WCE can charge a premium but bang for buck, definitely the Fiji Pro. Let's hope its priced well.
Fiji Pro at $499 would be killer if its ~10% slower than XT version (~$699). The WCE can charge a premium but bang for buck, definitely the Fiji Pro. Let's hope its priced well.
$449 is my bet. If I win you buy it for me. If I lose, you buy it for yourself. :whiste:
All down to costs. since its at the end of 28nm life for GPUs I'd expect its low enough
Out of the box the difference will likely be ~17-20%.
1050mhz x 4096 / (1000 x 3584) = 20%
That's kinda the point - AMD underclocks their 2nd tier cards (5850/6950/7950/R9 290 all shared this trait) and with less shaders/TMUs, performance falls against the flagship by more than it really is in the hands of overclockers. That's where overclocking comes in - we get those clocks up to 1150mhz+!
Fiji Pro at $499 would be killer if its ~10% slower than XT version (~$699). The WCE can charge a premium but bang for buck, definitely the Fiji Pro. Let's hope its priced well.
still kinda afraid that they didn't even have the card booting 2 weeks ago. that means no launch in June IMHO
Fury Pro unlocked cores via modded bios? what!!? That would be awesome. One can hope. Maybe, just maybe.
4096/3584-14% SP difference
5850/5870-11%
6950/6970-9%
7950/7970-14%
290/290x-10%
Gtx970/980-23%,224bit vs 256bit
980TI vs TITANX-9%
I am not an engineer but something about the specs strikes me as odd. Why would Fiji PRO have 128 ROPs, double that of Hawaii but yet shaders and TMUs only increase 27%? With colour compression of Tonga's GCN 1.2, even with 64 ROPs, the pixel fill-rate would increase 100%. Why are they doubling the ROPs yet again but only increasing shaders and TMUs so little? Is the pixel-fillrate such a massive bottleneck for current GCN design? Doesn't appear to be a well balanced design on paper to cram that many ROPs, but I am just thinking out loud....
Similarly for Fiji XT, they are increasing shaders/TMUs by 45% (ok by 53% if we account for 1.05Ghz clocks vs. 1Ghz on the 290X) but yet ROPs double. Wouldn't it make sense to have 96 ROPs + Tonga's colour compression? That already sounds like overkill as is, nevermind 128 ROPs. My logic is failing me on this one as I can't reconcile this spec. Furthermore, 128 ROPs of Fiji is 33% more than paper specs of GM200 but yet performance estimates put this card ~ Titan X +/- 10%. Are AMD's ROPs that inefficient compare to NV's?
I know it's often just pointless to compare paper specs of NV vs. AMD but I even when comparing that number of ROPs to Hawaii, it seems like overkill, no?
Clock speeds; everyone always forgets clock speeds. GM200 custom models should easily hit 1.4-1.5 GHz at least with, what, 1.15-1.2 GHz boost speeds even on stock reference designs? If Fiji's anything like the 285 and 290X, it won't overclock to nearly the same level as easily (average for those is where, about 1.1-1.15 GHz?). Frequency is a multiplier with ROPs, TMUs, and shaders ergo you have AMD going with a "wider" but "slower" GPU design and Nvidia with a "narrower" but "faster" one. It seems to apply to everything with GM200 vs Fiji; 33% more ROPs, TMUs, and shaders in Fiji but with less individual throughput/efficacy in no small part due to frequency disparity.I am not an engineer but something about the specs strikes me as odd. Why would Fiji PRO have 128 ROPs, double that of Hawaii but yet shaders and TMUs only increase 27%? With colour compression of Tonga's GCN 1.2, even with 64 ROPs, the pixel fill-rate would increase 100%. Why are they doubling the ROPs yet again but only increasing shaders and TMUs so little? Is the pixel-fillrate such a massive bottleneck for current GCN design? Doesn't appear to be a well balanced design on paper to cram that many ROPs, but I am just thinking out loud....
Similarly for Fiji XT, they are increasing shaders/TMUs by 45% (ok by 53% if we account for 1.05Ghz clocks vs. 1Ghz on the 290X) but yet ROPs double. Wouldn't it make sense to have 96 ROPs + Tonga's colour compression? That already sounds like overkill as is, nevermind 128 ROPs. My logic is failing me on this one as I can't reconcile this spec. Furthermore, 128 ROPs of Fiji is 33% more than paper specs of GM200 but yet performance estimates put this card ~ Titan X +/- 10%. Are AMD's ROPs that inefficient compare to NV's?
I know it's often just pointless to compare paper specs of NV vs. AMD but I even when comparing that number of ROPs to Hawaii, it seems like overkill, no?
I am not an engineer but something about the specs strikes me as odd. Why would Fiji PRO have 128 ROPs, double that of Hawaii but yet shaders and TMUs only increase 27%? With colour compression of Tonga's GCN 1.2, even with 64 ROPs, the pixel fill-rate would increase 100%. Why are they doubling the ROPs yet again but only increasing shaders and TMUs so little? Is the pixel-fillrate such a massive bottleneck for current GCN design? Doesn't appear to be a well balanced design on paper to cram that many ROPs, but I am just thinking out loud....
Similarly for Fiji XT, they are increasing shaders/TMUs by 45% (ok by 53% if we account for 1.05Ghz clocks vs. 1Ghz on the 290X) but yet ROPs double. Wouldn't it make sense to have 96 ROPs + Tonga's colour compression? That already sounds like overkill as is, nevermind 128 ROPs. My logic is failing me on this one as I can't reconcile this spec. Furthermore, 128 ROPs of Fiji is 33% more than paper specs of GM200 but yet performance estimates put this card ~ Titan X +/- 10%. Are AMD's ROPs that inefficient compare to NV's?
I know it's often just pointless to compare paper specs of NV vs. AMD but I even when comparing that number of ROPs to Hawaii, it seems like overkill, no?
1.5Ghz overclocks doesn't seem that magical once you consider the fact that on stock cards they can boost up to 1.2Ghz (1.5 divided by 1.2 = 25% OC). It's not really 50% OC... Check out the results from OCing, most sites noted around 20% more fps.Clock speeds; everyone always forgets clock speeds. GM200 custom models should easily hit 1.4-1.5 GHz at least with, what, 1.15-1.2 GHz boost speeds even on stock reference designs? If Fiji's anything like the 285 and 290X, it won't overclock to nearly the same level as easily (average for those is where, about 1.1-1.15 GHz?). Frequency is a multiplier with ROPs, TMUs, and shaders ergo you have AMD going with a "wider" but "slower" GPU design and Nvidia with a "narrower" but "faster" one. It seems to apply to everything with GM200 vs Fiji; 33% more ROPs, TMUs, and shaders in Fiji but with less individual throughput/efficacy in no small part due to frequency disparity.
Both companies also bulked up significantly on ROPs beginning with Hawaii in 2013 (37.5% more shaders than Tahiti, 100% more ROPs). Nvidia's move to 96 ROPs with GM200 is effectively well over double GK110's theoretical capacity taking Maxwell's OC headroom into account, even. Perhaps pixel fillrate (like bandwidth) is becoming an increasingly important limitations for higher-resolution demands.
Edit: About Fiji Pro specifically, it's not by design but AMD have never disabled any ROPs on cut-down GCN parts and that's what Fiji Pro is after all. Maybe it would have ramifications on other parts of the chip's performance (GTX 970 style) if they disabled some so they just leave what's there intact. Hawaii Pro had 25% more shaders and 100% more ROPs than Tahiti, history repeats I guess.
I didn't say 1.5 GHz was 50% higher than stock or anything magical, but it (and the higher stock frequencies too) is way higher than GCN parts can normally do which is a crucial factor in Maxwell performing well above what other chip specifications indicate.1.5Ghz overclocks doesn't seem that magical once you consider the fact that on stock cards they can boost up to 1.2Ghz (1.5 divided by 1.2 = 25% OC). It's not really 50% OC... Check out the results from OCing, most sites noted around 20% more fps.
Hawaii cards can get 1.2Ghz no problem by pumping the voltages. That's 20% OC and you don't need special cards for that, reference PCB card will do just fine such as the reference PCB card Sapphire Tri-X. All you need is good cooler. Hawaii cards also noted to scale well with core clocks too.
Fudzilla is reporting that the Grenada and Antigua chips in R9 390(X) and R9 380 are respins of Hawaii and Tonga. Assuming this is accurate (and it might not be), that could mean a variety of different things. We could be dealing with just a new stepping with slightly better power usage characteristics (least exciting but most likely, IMO), or it's possible that actual changes were made to the chip itself - probably not a full GCN 1.2 update for Hawaii (though anything is possible given the lack of solid info so far) but perhaps updated fixed-function blocks such as an updated UVD for HEVC decoding.
I'm expecting something like the 4870 -> 4890 treatment RV770 got back in the day, Hawaii is still relevant enough not to fall too much behind Tonga in features. Such tweaking would put it ahead of the 980 easily.
I'm concerned there is no mention of Pitcairn... if anything that's the chip that needs updating the most.
Meh, these tech new sites are in the dark like the rest of us, only difference is they lack logic and speculate like NUTs, completely wildly with so many different outcomes.
Anyone with some common sense will have realized immediately, AMD simply cannot re-badge Hawaii, it's not competitive due to the massive power consumption delta. The R290/X SKUs are selling for $200-280. If they are releasing a new series of SKUs and want to sell it for $329/389, it has to be significantly improved to compete* against 970/980