Introducing the Radeon X1650 XT: A New Mainstream GPU from ATI
by Josh Venning on October 30, 2006 6:00 AM EST- Posted in
- GPUs
F.E.A.R. Performance
F.E.A.R. is another one of our standard benchmarks for testing GPU performance. When it was first released, it was the most graphically intensive first-person shooter (or of just about any genre for that matter) on the market. We also found that the game had a few quirks, notably a poorly implemented "soft shadows" effect option that caused a big performance hit without really working graphically at all.
The test we use is the built in performance analysis test that the developers conveniently included in the performance section of the game menu. This is basically a flythrough of a few different rooms consisting of a gunfight, a pool of rippling water and an explosion with flames at the end. We tested F.E.A.R. with the highest quality settings enabled, naturally with the exception of soft shadows which were disabled.
With F.E.A.R. the 7600 GT goes back to getting a few FPS better than the X1650 XT across all resolutions. Around 25 FPS and above is what we consider playable for this game, and most of the cards here achieve this up to 1600x1200. The 6600 GT doesn't quite manage this resolution, and the X1800 GTO is borderline playable. These happen to be cards that have been on their way out for a while now, so this isn't such a surprise.
F.E.A.R. is another one of our standard benchmarks for testing GPU performance. When it was first released, it was the most graphically intensive first-person shooter (or of just about any genre for that matter) on the market. We also found that the game had a few quirks, notably a poorly implemented "soft shadows" effect option that caused a big performance hit without really working graphically at all.
The test we use is the built in performance analysis test that the developers conveniently included in the performance section of the game menu. This is basically a flythrough of a few different rooms consisting of a gunfight, a pool of rippling water and an explosion with flames at the end. We tested F.E.A.R. with the highest quality settings enabled, naturally with the exception of soft shadows which were disabled.
With F.E.A.R. the 7600 GT goes back to getting a few FPS better than the X1650 XT across all resolutions. Around 25 FPS and above is what we consider playable for this game, and most of the cards here achieve this up to 1600x1200. The 6600 GT doesn't quite manage this resolution, and the X1800 GTO is borderline playable. These happen to be cards that have been on their way out for a while now, so this isn't such a surprise.
33 Comments
View All Comments
guidryp - Thursday, November 2, 2006 - link
They spec like this:1650XT: 8 vertex Pipes, 24 pixel pipes, 8 Raster pipes, 575MHz, Mem 675MHz X 128 bus.
7600GT: 5 vertex Pipes, 12 pixel pipes, 8 Raster pipes 560MHz, Mem 700MHz X 128 bus.
And the ATI card barely holds it's own? I was expecting a walkaway after reading the specs.
coldpower27 - Saturday, November 4, 2006 - link
It should be 24 Pixel Shaders vs 12 Pixel Shaders.while both have 8 ROP's, it is probably the X1650 XT only has 8 TMU while the 7600 GT has 12 as both are half their flagship derivatives. Ignore vertex amounts those tpyically aren't half and don't contribute to much on the most part to performance it seems anyway.
X1900 XTX 48 Pixel Shaders, 16 Rasterization Operators, 650MHZ, Mem 775MHZ x 256 Bit Bus
7900 GTX 24 Pixel Shaders, 16 Rasterization Operators, 650MHZ, Mem 800MHZ x 256 Bit Bus
The X1900 XTX doesn't walkaway from the 7900 GTX on the whole either.
trinibwoy - Wednesday, November 1, 2006 - link
Do you guys do other testing that you comment on that is not represented by the graphs? The numbers show a 1 fps difference, yet you use terms like "significant" and "clearly beats". Maybe some median low fps numbers would help demonstrate what you're saying.soydeedo - Monday, October 30, 2006 - link
hey i know i can make an educated guess as to where the x1650xt would end up on q4 benches compared to nvidia's offerings, but i'm still curious why this game was not included in the testing? with quakewars around the corner i think people are still interested in doom 3 engine performance.johnsonx - Monday, October 30, 2006 - link
I suppose this name is part of ATI's general trend lately. It used to be that the XT moniker meant the same GPU with slightly higher clocks. Now it seems like the XT parts are a separate family. The X1300XT has nothing to do with the other X1300's (rather it's a rebadged X1600Pro), the X1900XT has more pipes than non-XT members of the X1900 family, and now the X1650XT has nothing to do with the rest of the X1600/1650 family.It all makes it a bit hard to choose.
Kougar - Monday, October 30, 2006 - link
On page 10 it is mentioned thatThis is completely going against the bar graphs, specifically the HL Episode One graph. The x1650XT got up and began walking away from the 7600GT without AA, but with AA it tripped and slide into place just behind the 7600GT. At resolutions below 1600by1200 it even began losing by a sizeable margin.
Josh Venning - Monday, October 30, 2006 - link
This paragraph has been tweaked a little bit. In HL2 Episode one the X1650 XT only does better than the 7600 GT at the highest resolution with AA enabled, but in Battlefield 2 it performs a little better over most of the resolutions.Cybercat - Monday, October 30, 2006 - link
How many vertex units does this thing have?JarredWalton - Monday, October 30, 2006 - link
See (updated) table on page 2: it has 8 vertex units, 24 pixel pipes, and 8 ROPs. Basically, lots more power than the X1600 XT. I would guess the pixel pipes are more like R580 pipes (i.e. more shader power, but not necessarily the same as an NVIDIA pixel pipeline in raw power).Cybercat - Monday, October 30, 2006 - link
Alright, cool. I keep a chart with stats of graphics cards, so I'm just making sure I have the vertice throughput correct. Other than the useless X1650 Pro, ATI seems to have a much more competitive mainstream line now. There is now more confusion than ever, though.