![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
radeon hd x2900 xt
- Thread starter JVC
- Start date
debello_64
Gawd
- Joined
- Oct 18, 2004
- Messages
- 672
The 2900XT has flames, that has to be worth atleast +5fps
Mr. Stryker
Supreme [H]ardness
- Joined
- Mar 15, 2005
- Messages
- 5,220
Drivers. And some design flaws I believe. But it still works for the price.
This has been answered to death, by here goes again
Basically, filling rate. R600 bets on tremendous shading power, yet lacks in filling rate, which is about the same as a GTS 640. AA has some problems too, since the performance hit, is very large when AA is on, in most situations.
Also, and this is my personal opinion, I believe there's some serious efficiency problem, in how the scheduler distributes the workload, to all the Stream Processors. There are some games out there, that take advantage of shader power, yet the XT still lags behind the GTS 640, in most cases. Maybe just because those games, even though shader intensive, also need good fill rate capabilities, but still, it's 320 SPs vs 96 SPs (i.e. 192 SPs as ATI counts them), which should make a difference.
I'm sure those here will corect me but as I understood it's a two-part issue:
1. ATI (and Nvidia) "simulate" performance of a new line on supercomputers before they submit the final design to the fab to create the chips. How well the GPU will clock up and how efficient the GPU turns out to be is not an exact science so there is a plus or minus 10-20% on the final results. ATI predicted (and prayed LOL) that the GPU would be on the plus side but it ended up being on the minus side. In other words final silicon was slower than they predicted and they were unable to bin enough higher clocked GPU's and had to lower the final clock on the cards thus lowering performance.
2. ATI was betting than DDR4 would be faster and more stable by the time the GPU's would be ready. This was not the case and they had to ship with either DDR3 or DDR4 with slower timings.
Now none of what I just said is 100% but just what I have heard which is no better than hearsay Please filling the blanks and cross out my errors LOL.
Thanks
-JB
1. ATI (and Nvidia) "simulate" performance of a new line on supercomputers before they submit the final design to the fab to create the chips. How well the GPU will clock up and how efficient the GPU turns out to be is not an exact science so there is a plus or minus 10-20% on the final results. ATI predicted (and prayed LOL) that the GPU would be on the plus side but it ended up being on the minus side. In other words final silicon was slower than they predicted and they were unable to bin enough higher clocked GPU's and had to lower the final clock on the cards thus lowering performance.
2. ATI was betting than DDR4 would be faster and more stable by the time the GPU's would be ready. This was not the case and they had to ship with either DDR3 or DDR4 with slower timings.
Now none of what I just said is 100% but just what I have heard which is no better than hearsay
Please filling the blanks and cross out my errors LOL.Thanks
-JB
Brent_Justice
Moderator
- Joined
- Apr 17, 2000
- Messages
- 17,755
Primarly it is much too power hungry for the performance. It needs to be .65nm or less. If they had reached their originally intended clock speeds of 1GHz+ performance wouldn't be the issue so much. It suffers with the lower clock speeds. Also, they need to up the ROPs and TMUs for today's games. And of course driver tweaking of the dispatch processors performance.
What makes you say that?
An old Pentium running at stock 3Ghz is a lot slower than a Q6600 running at 2.4Ghz. It would be naive to say the slow, obsolete pentium has "better specs" because one measurement, in this case processor speed, appears larger. There are many other factors and the actual design matters more.