My experience with Redwood (Radeon 5570) was pretty good. Games like CoD4 and WaW could get 60 FPS + 2x AA quite easily if you stayed at 720p. Without AA, 1600 x 900 and getting 60 FPS was possible. I will say though, I think multiplatform titles on the PC are getting a bit lazy in the optimization department. Crysis ran quite well though, high settings 720p, no AA. I would run more tests, but I recently replaced the 5570 with a GTX 460 and sold the Radeon to a friend. The real advantage of the 5570 over Xenos to me is the possibility of playing at 1080p. The issue is that so many multiplatform games lack the optimization needed like I said earlier, though Unreal 3 scaled excellently to it. Rainbow Six Vegas 2 with AA, 1600 x 900 resolution and pretty much 50+ FPS was very nice to say the least.
Xenos: How fast is it?
- Thread starter chavvdarrr
- Start date
My experience with Redwood (Radeon 5570) was pretty good. Games like CoD4 and WaW could get 60 FPS + 2x AA quite easily if you stayed at 720p. Without AA, 1600 x 900 and getting 60 FPS was possible. I will say though, I think multiplatform titles on the PC are getting a bit lazy in the optimization department. Crysis ran quite well though, high settings 720p, no AA. I would run more tests, but I recently replaced the 5570 with a GTX 460 and sold the Radeon to a friend. The real advantage of the 5570 over Xenos to me is the possibility of playing at 1080p. The issue is that so many multiplatform games lack the optimization needed like I said earlier, though Unreal 3 scaled excellently to it. Rainbow Six Vegas 2 with AA, 1600 x 900 resolution and pretty much 50+ FPS was very nice to say the least.
5570 has a 150 MHz clock speed advantage over Xenos when at 650 MHz. It has 8 ROPs like Xenos, so that means more 30% pixel fillrate than Xenos. It has 25% more texture fillrate per clock, resulting in a 63% advantage following the higher clock speed. Probably more Z fillrate. It's the 5th generation of ATI's unified shader architecture so I'd imagine that its 400 shaders are much superior. Some of the cards have GDDR5, meaning considerably more memory bandwidth for the most part, and they have 512MB-1GB of dedicated video memory.
So I don't really see how 5570 would have to worry about Xenos.
So I don't really see how 5570 would have to worry about Xenos.
Last edited by a moderator:
Definitely more than just the raw specs - effeciency (as you mentioned) is no doubt superior. On my LP 5570 with GDDR3, multiplatform titles are far superior than the 360 versions by and large (save for GTA4, of course my dual-core X2 is the main bottleneck there).5570 has a 150 MHz clock speed advantage over Xenos when at 650 MHz. It has 8 ROPs like Xenos, so that means more 30% pixel fillrate than Xenos. It has 25% more texture fillrate per clock, resulting in a 63% advantage following the higher clock speed. Probably more Z fillrate. It's the 5th generation of ATI's unified shader architecture so I'd imagine that its 400 shaders are much superior. Some of the cards have GDDR5, meaning considerably more memory bandwidth for the most part, and they have 512MB-1GB of dedicated video memory.
So I don't really see how 5570 would have to worry about Xenos.
As far as I could recall, Capcom did compare Xenos being capable of spewing out the same amount of vertices to that of an 8800, but this was if all of its 48 shaders were just doing that. I remember people here said that xenos can churn out 2x as many traingles than the RSX coz it was like 1 per clock compared to RSX's every other clock per cycle. Then the whole can of worms were opened when people were like , well, the Cell can help the RSX with that...then people got wild and crazy and a bunch of "naughty boys" were given out by the mods... I miss deadmeat... lol... just kidding...
But from what I could recall, before the punishments for trolling/rude behavior were laid out, most people agreed that the Xenos was just a little bit behind the X1900 while the RSX was about the same as a 7800 GTX because they found out about some tweaks done on the RSX where they added an extra MAD or MUL or whatever it was... so they took that into account. But I do clearly remember about the Xenos having triangle per second twice that of RSX... people crowned Xenos as the winner coz of its efficiency and flexibility compared to RSX at the time...
But from what I could recall, before the punishments for trolling/rude behavior were laid out, most people agreed that the Xenos was just a little bit behind the X1900 while the RSX was about the same as a 7800 GTX because they found out about some tweaks done on the RSX where they added an extra MAD or MUL or whatever it was... so they took that into account. But I do clearly remember about the Xenos having triangle per second twice that of RSX... people crowned Xenos as the winner coz of its efficiency and flexibility compared to RSX at the time...
Xenos might get close to the 8800 in setup rate since that was limited to clock rate in all GPU's until recently. In terms of vertex shading performace though, it wouldn't even come close to an 8800.
R600 was able to slap G80 around in some pure geometry tests. I think I posted this earlier in the thread. Obviously this didn't do much for real game performance though.
Xenos' unified shaders are a sort of a hybrid of R580 and R600. I really wonder how efficient Xenos is compared to R600. I doubt that ATI became less efficient with R600. R600 also has a 240 MHz advantage in core clock speed over Xenos.
For overall performance, remember too that Xenos is only a 8 ROP GPU so it has rather unimpressive pixel fillrate compared to any high end PC GPU since X800. Considering the bandwidth available, 16 ROPs would be rather pointless most likely.
Xenos' unified shaders are a sort of a hybrid of R580 and R600. I really wonder how efficient Xenos is compared to R600. I doubt that ATI became less efficient with R600. R600 also has a 240 MHz advantage in core clock speed over Xenos.
For overall performance, remember too that Xenos is only a 8 ROP GPU so it has rather unimpressive pixel fillrate compared to any high end PC GPU since X800. Considering the bandwidth available, 16 ROPs would be rather pointless most likely.
Beginning with the 4xxx series, wasn't ATi using a new ROP design that IIRC could process two pixels per clock per ROP? I remember seeing articles on how 2x MSAA was being applied on a number or games with little to no hit on framerate. I always saw it as a reason as to why the 4670 I used to have whipped the 9800M GS in the laptop I used to have, since I knew the 4670 had 8 ROPs and the 9800M GS had 16. With or without OC, the 4670 scaled up to higher resolutions much better than the 9800M GS did.
I remember trying GTA4 on the 5570 I used to have, and the performance wasn't there, even with lower settings (with an Athlon II x4) but I think it was the fact that I only had 2 GB of system RAM, on a WIn7 machine which caused the real issues.
I wish I kind of had not transfered my LP system into a full size case to give it the ability to take a GTX 460, as it was expensive getting a new case, PSU and of course the video card itself. If I had been patient I could've gotten the low profile 5750 or GTS 450 that is now available and would've been good enough for most of my needs.
And back to the real topic (Xenos vs...........whatever), I've always been inclined to compare SIMD counts, so I tend to put the Xenos as being about 3/4 the speed of a Radeon 2900 or 3800 or 46xx series GPU at a similar speed, without acknowledging the increases in core clock, memory bandwidth advantage, amount of memory available, console's extra optimizations, whatever. Those GPUs as well as the good 'ol Geforce 9600GT - G94 (which the 9800M GS is a mobile version of), all pretty much guarantee us "console parity" for the most part when multiplatform games are similarity configured to match each other graphically. At least Source, Unreal 3 give us more efficiently coded engines to make comparisons with, in which case the 9600GT really just blows the Xenos out of the water. The eDRAM die does make 2x and 4x MSAA a better fit for the 360 at 720p, but the 9600GT can actually manage 1080p rendering. Any reason that the 9600GT has to be brought down to 720p for decent FPS, Xenos would not be able to match graphically in the first place for the most part. CoD4 did give me that experience though. The 9800M GS did not really do what I expected at 1080p with no AA, so to get 60 FPS with no slowdown, I had to bring the game down to 720p with 2x AA.
Last edited by a moderator:
9800M GT is a lot like a desktop 9600 GT but the core clock is reduced about 125 MHz and RAM clock 100 MHz. Radeon 4670 is slower than a 9600GT according to reviews but your 9800M GT was probably similar or slightly slower than it.
http://techreport.com/articles.x/15559
9600GT should blow Xenos away considering how much more fillrate, RAM and bandwidth it has available. Also considering it can beat a 4670, which is frequently faster than a 3850/70 and 2900XT.
http://techreport.com/articles.x/15559
9600GT should blow Xenos away considering how much more fillrate, RAM and bandwidth it has available. Also considering it can beat a 4670, which is frequently faster than a 3850/70 and 2900XT.
The Geforce 9800M GT was a G92 based part, with 96 shaders, not the 64 of G94 that made up the 9800M GS. Just got to love Nvidia's naming conventions for their parts, especially in laptops *sarcasm* Clock speeds for the 9800M GSin my Asus were just as Wikipedia as them listed: 530 Core/1325 Shader and 1600 Memory. The desktop Radeon 4670 had the added advantage of the core being 750 MHz and I guess for what I was playing, the-just-over-half-the-memory-bandwidth-of-the-9800M GS must have not been much of an issue, even when scaling up the render resolution and other graphics options. I think the bandwidth was ~51 GB/s in the 9800M GS compared to ~29 GB/s for the 4670.
G94 = 9600 GT which was what my point was mostly. The desktop 9600 GT is always (?) faster than 4670 according to reviews. So I'm surprised that you'd see 4670 ever being dramatically faster than G94. I think the 4770 was more of a 9600GT competitor.
I have a 4670 and a factory overclocked 3850 512MB and sometimes the 3850 is faster because of the ~80% bandwidth advantage and 16 ROPs, I think. 9600GT is always faster than the 3800 cards AFAIK.
I have a 4670 and a factory overclocked 3850 512MB and sometimes the 3850 is faster because of the ~80% bandwidth advantage and 16 ROPs, I think. 9600GT is always faster than the 3800 cards AFAIK.
Last edited by a moderator:
Your assumption on the 9600GT vs the Radeon 38xxs is about correct, though I suspect a good deal of it is related to Nvidia just having a plain upper hand in developer relationships, and still at that time, much more on top of improved driver releases than DAAMIT.
You might want to look at this regarding ROP performance of Radeon R700 series hardware<---Click
I guess it could even hint to how well AMD would do in pushing high resolutions on the Evergreens when AA is applied . While yes, Geforce cards win all out in the end, the Radeons do it with many less transistors and of course don't need a much faster separate clock domain for the shaders/stream processors. That being said, I do wonder how the Geforces' power usages and benchmarks would compare if the entire GPU ran at one clock speed like AMD's. Significantly less of course, but it would be an interesting dynamic to compare since the TMUs and ROPs would have no change in clock. Then there is all the I guess, "extra" circuitry dedicated to GPGPU functions perhaps outside of the DirectX, DirectCompute, OpenCL specifications, like for CUDA.
You might want to look at this regarding ROP performance of Radeon R700 series hardware<---Click
I guess it could even hint to how well AMD would do in pushing high resolutions on the Evergreens when AA is applied . While yes, Geforce cards win all out in the end, the Radeons do it with many less transistors and of course don't need a much faster separate clock domain for the shaders/stream processors. That being said, I do wonder how the Geforces' power usages and benchmarks would compare if the entire GPU ran at one clock speed like AMD's. Significantly less of course, but it would be an interesting dynamic to compare since the TMUs and ROPs would have no change in clock. Then there is all the I guess, "extra" circuitry dedicated to GPGPU functions perhaps outside of the DirectX, DirectCompute, OpenCL specifications, like for CUDA.
Last edited by a moderator:
It would be as fast as a GeForce 6800GT but more importantly the GPUs purpose to use the DirectX 9, direct3d API based development tools so its speed is irrelevant because there are no direct GPU calls.
