-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Poll: 390X: How fast do you think it will be.
- Thread starter humbug
- Start date
I wouldn't put to much stock in that. More to it than that.
The 295X2 = 1024Bit
The 390X = 4096Bit
Going by number of shaders on 290X vs number of shaders for 390X... I would expect performance to be 5-15% slower than Titan X depending on game
According to this The vs the 290X Titan-X is 43% faster @ 1440P and 4K
290X = 2816 Shaders
390X = 4096 Shaders = +45%
290X is GCN 1.1
AMD's latest Arch to date and released is GCN 1.2 (Tonga) it has <4x the Tessellation Throughput and +60% the Texture Bandwidth of GCN 1.1.
Tonga (R9 285) 1796 Shaders @ 920Mhz - 256Bit Bus is is as fast as the 384Bit 950Mhz 280 to faster than the 2048 Shader 1050Mhz 384Bit 280X.
Overall GCN 1.2 is about 15% faster than GCN 1/1.1
So the Titan-X is 43% faster than a 290X, add +45% Shaders add the 15% over performance increase to all those Shaders, without Shader full scaling its going to be about 45% faster than the 290X, perhaps a little more than that.
So that equates to about + a few % faster than the Titan-X
Completely ignoring any performance benefit HBM might have.
Board updated.
Last edited:
For sure but not unrealistic.
A 1796 Shader GCN 1.2 @ 920Mhz 256Bit 190 Watt 285 is nearly as quick as its 2048 Shader 1000Mhz 384Bit 220 Watt big brother, sometimes a little quicker, it wouldn't take a lot of tweaking to get a 2048 Shader 512Bit 4GB Tonga around 290X performance at 180 Watts.
And it would have to be to compete with the GTX 970, or they might as well not bother.
http://www.techpowerup.com/reviews/Sapphire/R9_285_Dual-X_OC/25.html
Last edited:
They don't need to shrink dies, Hawaii is 40% more GPU on Shaders alone vs Tahiti, it also has a larger IMC, 2x the ROP's, 4x the ACE units
Tahiti is 365mm2, by all accounts Hawaii should be about 550mm2 (at least) like NV GK110, its not, its 438mm2, well over 100mm2 smaller than it should be as a Tahiti architecture.
The reason its so small in relative terms is because they redesigned it and use much denser transistor packing, there have been rumours of AMD packing even denser transistors with the 300 series, but even without that with 4096 Shaders it would be ~635mm2, the Titan-X is 601mm2.
Last edited:
I don't know why you are bringing CPU's on different manufacturing nodes even relative to eachother, your just comparing Apples to Oranges, its not relevant all it does is add an irrelevant curve ball to confuse the argument.
Its actually the other way around, AMD have packed more GPU on the same size die than Nvidia (certainly Kepler), the clock rate differences are nothing to do with with those dies or the transistors on them.
Nvidia have a new firmware power delivery system with Maxwell, its able to switch between high and low power states at very high frequency resulting in low average power consumption.
But its only able to do that in a (goldy locks zone), overclocking, even some factory overclocked 980's run at 250 Watts average power draw, just like a 780TI, tho the 980 having higher clock rates.
There isn't actually that much architectural difference between Kepler and Maxwell, aside from Maxwell having better Texture compression technology (just like Tonga)
Its able to clock higher than AMD's current line of GPU's, and Kepler purely because of its power efficient firmware, that allows it to run higher clocks.
A 1500Mhz 438mm2 290X would be faster than a 1500Mhz 398mm2 980.
There is nothing to stop AMD from introducing their own power efficient systems.
Pre- 290X it was said by some in here that such a GPU could not exists on the AMD side, to match or beat a GTX Titan because it would have to be to big, such was the confidence of these people.
In the end AMD had a GTX Titan of their own and it was much smaller than the Titan. Shock!
Last edited:
Hawaii clocks like Tahiti before it.
Your completely ignoring the power requirements of overclocking, if Hawaii, and Tahiti for that matter requires 400 watts to run at 1500Mhz its never going to do it outside of LN2 because the chip couldn't take the heat on standard cooling, and thats ignoring the stress on the PCB and its components.
That is why you never see any 1500Mhz 290 outside of LN2 benching.
Bring the power requirements down and it will do it, thats why Maxwell does it, nothing to do with the process manufacturing, they are the same for Hawaii and Maxwell.
Last edited:
"Why you keep comparing hawaii vs tahiti"
Already explained this.
Because you said denser packing is a problem for AMD in clocking, Hawaii has a lot denser transistor packing than Hawaii and they clock the same.
"You said amd don't need to shrink dies"
I have already explained this too.
Hawaii has about 50% more GPU on a die thats grown by 20% to accommodate it, they have grown significantly without a die shrink, no reason why they couldn't do it again. On current speculative specifications 4096 Shaders a 390X would only be marginally bigger than a Titan-X.
"Why can all maxwells clock high, why can a baby maxwell clock high then? No clever power regs on the 750s?"
Yes there is, the same thing as on big Maxwell.
"Nvidia have for more efficient shader throughout with less resources. Amd have to use more shaders andbbigger chips to compete."
Apples and Oranges again.
AMD's shaders are not the same as Nvidias, AMD have more shaders per die size than Nvidia, Nvidia use a (bigger shader) so to speak.
Last edited:
Did you read anything above you?
You keep comparing Apples to Oranges.
The GTX 780TI had 2880 Shaders, it was 550mm2 in size, the 290X had 2816 Shaders, its much smaller than the 780TI @ 438mm2, they both have about the same performance and power consumption.
Your problem is you keep looking at the numbers and then stating results that you yourself can't even equate.
In Kepler vs Hawaii, at least, there is no efficiency advantage or process disadvantage for AMD, as a matter of fact AMD get more from less vs Kepler.
Let GCN 1.3 arrive before proclaiming it to be inadequate, that's exactly what was said about a GTX Titan rival, until it arrived and was in some ways better than even i predicted it to be.
Last edited:
Also do now with the 295X2.
Updated
no i meant it will be a 295 x 2 but just as one cardlol
I thought it didn't read right earlier on.
Sorry, missed this. will update.
Power-consumption is the only problem i foresee, if the specs are correct i see no reason why it can't edge out the TX, i voted 1, but only just, it will be tight over that +5%
Power-consumption doesn't scale linear like that with Shaders, +45% Shaders should result in about an extra 30% powerdraw, which should bring it to 350 Watts, 270 + 80.
300 Watts or more is too high, i wouldn't like to see it much more than 280 Watts.
i think HBM can save about 30 Watts, they need to get it down another 30 to 40 Watts, about 10 or 15%.
If they can do that and do it at an architectural level (not some firmaware that doesn't work outside of some goldy locks zone) then they are onto a power house that can be cooled just as easily as the 290X
Last edited:
There is another problem that could crop up, with the memory so close to the GPU things are going to get a bit warmer than they would with GDDR5. Even with an AIO cooler the extra heat could hold the GPU back.
As to the TXs they are a tough act to beat as they are probably the best overclockers in the Maxwell range. Their stock clocks are lower than the 980s but overclocked they get very close to the 980s max clocks.
Only time will tell how the 390X goes but if they are faster than the TXs I will probably be one of the first to say so.
I also said that multiple 290Xs @2160p were faster than their 980 counterparts when everyone else was saying that all the 980s needed was a driver update. I am still waiting for the drivers 7 months later.
Agreed keeping those HBM's cool so close to the GPU could be a problem, maybe if they separate the IC cooling from the GPU plate it would help, having an extended GPU plate to go over the IC like they do with GDDR5 isn't too good as you get heat transfer.
The HBM IC's themselves are low power and shouldn't generate much heat on thier own, its just about stopping the heat transfer from the GPU.
Last edited:
Whatever the possible problems maybe I think the 390X is going to be a very fast card and in the long run the way forward.
As good as the TX is it is not going to get any better, I think this card is a good place to say goodbye to 28nm/GDDR5.
The King is dead long live the King.
Yeah 28nm has actually done pretty well, it will take a while for the next manufacturing node to come as good, it may not be much good at all in relative terms with the first series of GPU's