AMD: R7xx Speculation
- Thread starter Unknown Soldier
- Start date
- Status
This is what I am talkin 'bout.
This looks to be an exact repeat of the pre 2900 hype. It was supposed to destroy NVDA and a lot of people were expecting it. It never happened. From what I have seen it is going to happen again.
People are going to be let down even though the 4800s are still going to be good. The hype needs to fade a bit.
Only YOU seem to be believing the RV770's are going to beat the Nvidia cards. Nobody here believes that - we all expect great performance for price cards at their range, and nothing out there makes us believe otherwise (or that it's going to beat the GT200's head to head). We're not the ones overhyping the card as anything more than what their price range suggests
Jawed
Legend
In theory Perlin Noise is the purest ALU test, as the ALU:TEX ratio of the shader is 9:1 in D3D assembly.
When compiled it comes out to 197 ALU instruction groups (clocks). Assuming 32 TUs, for a 5:1 ALU:TEX ratio, this shader is actually TEX bound! But only just.
So that explains why it's 95% of the theoretical speed-up over HD3870 (HD4850 should be 2.01x faster, 352fps).
Jawed
Last edited by a moderator:
Well, what do you have to back up your claims slappi?
If you are just speculating here then what is so much better about you claiming doom and gloom compared to others claiming it might outperform G200 in X2 configurations?
You sure sound a lot more passionate than other people here.
That was GDDR4. I don't feel like searching all the PDF's today, but you can see on Samsungs pie chart they list 1/3/5 as low latency.
For comparison, from this Sharky Extreme review:
3850 149
3870 173 (+16%, = freq. diff. b/w 3850 & 3870)
8800GT 145
8800GTS 177 (+22%, = freq. & ALU diff. b/w 8800GT & 8800GTS)
Hanners got 345 out of a 9800GX2. The GX2 should have +85% over an 8800GTS, but it's scoring +95%--curious, as each GX2 core is clocked lower than that in an 8800GTS (seventh paragraph down) and Hanners has an 8800GTS scoring 175 on a similar system (same Q6600, diff. MBs). I guess it's the newer drivers: FW174 vs FW169.
no-X's #s: 335/175 = +91%. 800*625 / 320*775 = +101%. (Edit: as Jawed said.)
Though I think it's a fillrate test we're all waiting for.
3850 149
3870 173 (+16%, = freq. diff. b/w 3850 & 3870)
8800GT 145
8800GTS 177 (+22%, = freq. & ALU diff. b/w 8800GT & 8800GTS)
Hanners got 345 out of a 9800GX2. The GX2 should have +85% over an 8800GTS, but it's scoring +95%--curious, as each GX2 core is clocked lower than that in an 8800GTS (seventh paragraph down) and Hanners has an 8800GTS scoring 175 on a similar system (same Q6600, diff. MBs). I guess it's the newer drivers: FW174 vs FW169.
no-X's #s: 335/175 = +91%. 800*625 / 320*775 = +101%. (Edit: as Jawed said.)
Though I think it's a fillrate test we're all waiting for.
Quite impressive, if true. A 9800 GX2 scores only about 382fps, so a HD4870 should outclass it in this test.
Perlin Noise is a very interesting feature test in both 3DMark06 and Vantage - little noticed though by those who only look for 3DMark-Scores and Crysis-Fps.
Last edited by a moderator:
They are nice.Though I think it's a fillrate test we're all waiting for.
Some mathematical alhorithm that, I think, is used to create randomness in procedural textures and other things.
9800GX? Did you mean 9800 GX2? My 9800 GTX OC clocked at 770/1150 only scored 225 at default settings.
-FUDie
I don't think you understood my question. I was asking how often the SF/interpolation ALUs would be idling in the first place to allow the MUL to be used. Without knowing how much the hardware is actually being put to use there's no way to make any assumptions about efficiency.
Yeah, sorry for the typo - will be fixed immediately.
Yeah it scales almost 1-for-1 with increases in shader units or clock. It's downright amazing how AMD just blew apart the teraflop barrier without breaking a sweat. It will be interesting to see how Nvidia marketing tries to play that down.
I really hope they're doing more with those shaders this time around though. I can't see them being too useful in gaming scenarios otherwise.
Though with all this talk of 3dmark numbers and massive flops, why hasn't there been a single leak of real game performance yet. Run the Crysis numbers!
Number of TMUs and ALU:TEX isn't out of up to now ATis trends.You have more infos? Share plz
Jawed
Legend
GTX 280's halo effect should do all the talking. Really it's a question of whether it's closer to 1.5x faster than 8800GTX or 2x faster. It seems that NVidia has become more dependent on GPU features that aren't merely TEX and Z rate. In other words there'll be more focus on D3D10 and extreme shaders.
Arguably ATI is making a larger relative stride, 2x TEX rate, 2x (we presume... sigh) the Z rate and 2.5x FLOPs - but NVidia appears to be fixing things that were decidedly ineffective in G80. So in comparing scaling over the previous GPUs it's going to be close and very interesting.
Jawed
leoneazzurro
Regular
it was 3:1 in R580, 4:1 in R600-RV670, so in RV770 is plausible a 5:1.
10:1 is, I think, too much (if they are not EXTREME TEX units, of course
)- Status
Similar threads
