AMD: R7xx Speculation

Status
Not open for further replies.
Semi-related to this, the performance gains in Assassin's Creed from DX10.1 were mostly due the fact that they could remove one pass when rendering the post processing effects
Intriguing, anyone has a clear idea of which DX10.1 feature allowed that? :)
 
Intriguing, anyone has a clear idea of why one should remove a working feature that improves performance? :)
Perhaps they've realised that they can improve the D3D10 version of the code and maybe improve the D3D10.1 version further?

Jawed
 
Semi-related to this, the performance gains in Assassin's Creed from DX10.1 were mostly due the fact that they could remove one pass when rendering the post processing effects

The DX10.1 support, however, will be removed in the next patch but probably will be reimplemented later

I suppose that means they do a separate Z pass when in DX10 mode (rather than using the MRT approach), just to get a useable depth buffer for the post-processing effects. Would be interesting to see if there's any performance difference with post-processing disabled.
 
Nothing's stopping the driver from using such functionality, but it doesn't help either, because you don't resolve the depth buffer for antialiasing. Only the color buffer is resolved.

Oooh, thanks and apologies for the stream of inane questions :)
 
According to Tgdaily ( http://www.tgdaily.com/content/view/37025/135/ ) the Radeon 48xx is in production. That much I'd read before, but this is the part that is exciting:

The Radeon 4850 version is set to come to market with an 800+ MHz core (the final clock has not been specified yet and will not be available until the final qualification is completed), while the 4870 will be the first mass-production GPU with a clock speed higher than 1 GHz. Prototype RV770 boards were clocked at about 1.05 GHz.

It has long been rumoured that they were aiming for those speeds, but that is the first I'd heard that they seem to have hit them. I personally thought it sounded overly optimistic. If that is accurate, good for AMD.
 
That would be impressive if true...if they have gone over 1GHz then 1.05Ghz should be trivial to achieve and that would take 480 shaders to 1TFlop! Sweet.
 
RV770 @ 1.05GHz is precisely what R600 should have been. I am glad ATi has corrected their mistake. Depending on price (and performance) I may wait until my EVGA step-up is at it's end before I swap cards again. I was going to upgrade to a 98GTX/GX2 from my 88GT, but if 4870 is out in force before that happens and it pwns I'll go back to ATi happily.
 
I don't think ATI will ever "own" Nvidia again.

I get the impression Nvidia are just toying with them. There best GPU (8800 Ultra) was released over a year ago now, and the new GPU (9800 GTX) they released barely beats it.

I fell oblidged to buy an ATI card as my motherboard is an AMD chipset, so I am hoping it's a decent card and better than the 8800 Ultra (That will do for me to be satisfied - likely buying two cards and using crossfire will still be cheaper than buying Nvidia's new GPU), but I do not expect any performance advantage to last longer than a month before Nvidia brings out something thats far better.
 
Remembering how R680's clocks turned out in the end, I hope that over GHz clocks can be acheived by overclocking. 875MHz by default would be my guess.
 
Prototype RV770 boards were clocked at about 1.05 GHz.
I think that surely has to be read as just to say 'we got a 1ghz chip doing 1TFLOP', presumably a big heatsink & above stock voltage used?
Would love to be wrong, 1ghz clock & 1TFLOP would be cool :D... assuming there's enough TMU/ROP to make use of it :p
 
I didnt say anything about the actual die-size, w/ smaller i ment the transistor count - and w/ less transistors u can clock the chip higher.
 
w0mbat: ATi's high-end GPUs used higher clock-speed than the smaller ones...

RV360 - 130nm low-k: 500MHz
R480 - 130nm low-k: 540MHz

RV515 - 90nm: 600MHz
R520 - 90nm: 650 (-700)MHz
 
I didnt say anything about the actual die-size, w/ smaller i ment the transistor count - and w/ less transistors u can clock the chip higher.

Someone (think it was silent-guy) mentioned that transistor speed has nothing to do with how big the chip is which makes perfect sense. The limiting factor is how much heat the chip puts out and your ability to cool it.

But I would think clock propogation presents its own set of problems too. Syncing a very high clock across the chip should get more difficult with larger chips right?
 
Status
Not open for further replies.
Back
Top