What are you comparing the performance against? The closest comparator would be RV515 to RV530 in terms of PS difference.SugarCoat said:
X1800xl
- Thread starter Unknown Soldier
- Start date
SugarCoat said:
I don't think so. There were metal layer changes before the root problem was discovered in the layout libraries. I think I am correct in summarizing it like this:
1. First spin - Not clocking well at all.
2. Metal layer changes - alleviates problem but does not solve it (ATI: we're in some deep doo-doo here!)
3. Problem is discovered in the libraries causing the problem (ATI: doh! We just did some completely unnecessary metal layer changes. There was a fundamental problem in the layout thanks to our borked libraries...*sigh*)
So, I think it was a case of the dam breaking and ATI re-spinning to bulk up the levees, to use an analogy. This was not entirely successful, but it helped. Once the real problem was discovered it was solved at the root (the dam). I am not sure if the re-spins helped in any other ways or if ATI has removed those jury-rig fixes. At any rate, the XLs should be the bulked up metal layer changes before the root cause was removed. If anything, they have more metal layer changes than the XT (the fixed one, which should have none, depending on how they are counting), but the XT may have inherited those properties from the intermediate changes.
I think this is more a case of RV530 not having enough fundamental rendering power. It is bulked up on shader performance, but it just doesn't have the power to plow through basic vertex/texture ops. If you look at the R580, which is basically a R520 with added shader power to the ratio that the RV530 has, it will still be a monster because it has 16 full pipelines. I think the basic idea can be summed up that the RV530 would be able to handle Quake 2/3 (exaggerating for illustrative purposes!) with shaders galore, but when it comes to massive games like Far Cry and beyond it's not the shaders that are halting it, but the basic geometry and texturing. I am quite convinced that R580 will be collecting dog-tags and I got this impression from looking at RV530. X-bit labs shows the RV530 outperforming the 7800 GTX in dynamic branching, for example. If the RV530 was 8-1-2-1 instead of 4-1-3-2, it would be much more powerful in real life usage, I believe. That would be something like a cross between a Radeon 9800 and a X800. Alas, it looks like it is a Volkswagen Beetle with a Porsche engine installed and the "chassis" won't be able to show off all that power under the hood (in terms of shader performance).
So, no, I think R580 will have all the ingredients to make it a competitor, not just an exaggerated shader "engine".
Last edited by a moderator:
Vertex performance is not really going to be an issue for RV530 since it has 5 VS and with it clocked as it is this puts it pretty much on par with an X800 XT in terms of VS performance (and the higher clock probably means more setup performance actually). The ratio of PS-to-Tex is where ATI see shading going in the future, so current review test apps are not going to be the best cases for it.
Jawed
Legend
The XL's clocks and the difference in MSRP between XL and 256MB XT ($50) implies, to me, that both will be seeing much lower prices soon and that an XTPE is due.
$50 for a 25% increase in core clock is nutty good value (let alone the increase in memory clock, too). No-one should be buying the XL when the 256MB XT appears. The XL needs to be about $100 less - in the same way that the X800Pro was about $100 less than the X800XTPE.
I'm still not sure about XL clocks, either. Whether they should have been 10 or 15% higher, but ATI has compromised because of the dregs of the soft-ground problem.
Jawed
$50 for a 25% increase in core clock is nutty good value (let alone the increase in memory clock, too). No-one should be buying the XL when the 256MB XT appears. The XL needs to be about $100 less - in the same way that the X800Pro was about $100 less than the X800XTPE.
I'm still not sure about XL clocks, either. Whether they should have been 10 or 15% higher, but ATI has compromised because of the dregs of the soft-ground problem.
Jawed
Jawed
Legend
I'm not being sarcastic: is D3 a good test case? I ask because it seems like a game built for DX8 GPUs.RoOoBo said:
http://wiki.doom3reference.com/wiki/Texturing
Are diffuse, normal and specular maps all AF'd? I guess so.
Nice graph, as usual!
Jawed
Joe DeFuria
Legend
Dave Baumann said:
Yes, that's my "complaint" about RV530 from a market positioning standpoint. Hindsight is of course 20/20, and when these configurations were drawn up 18-24 months ago, I may have made the same decision that ATI did, guessing that there would be more apps to "show off" the RV530. As it stands, though, I'd say the situation is that ATI is probably a cycle or two ahead of its time.
SugarCoat said:
wireframe already answered, but I'll reiterate. It's all about "absolute" pixel/texel fill rate power interms of the perceived RV530 deficiency. So long as R580 is clocked in the same ballpark as R520, the R580 should indeed be a monster in all respects.
radeonic2 said:
aye just look at the 7800s they were 600$ and now they are under 500$ .I'm sure after the first 2 or 3 weeks we will see huge price drops and then fsaa and hdr is very important to me .
Whats also important is the future and I'm looking foward to playing games like fear not doom3
The Riddick game would do, if you configure it in the right way.RoOoBo said:And UT2004 is DX7. Now if you tell me about a game using DX9 level features in OpenGL ... and of course it can't use glSlang
http://www.3dgamers.com/games/chroniclesriddick/downloads/
Actually the beetle chassis is quite stiff, they dont run in to any chassis flex like say.. the mustang foxbody chassis does.wireframe said:
Also a porsche engine isn't worth it when you look at what engine builders are able to get from a standard 4 banger with bigger jugs and better flowing heads etc.
Just a tidbit from a beetle owner, with a stocker
So what' the time frame on the R580?
zeckensack said:
Riddick is the next after we get Doom3 to work without problems. In fact we already tested it and, as usual, not even the trace is properly captured and we were unable to reproduce it with GLPlayer. With so much functionality in the OpenGL API and so many different calls using different parameter types (the main problem are buffers) we only implement what we need. Riddick seems to be using PushAttrib/PopAttrib and we are implementing that in the library. We don't even have 3D textures implemented in the framework, so 3D texturing in the simulator, which is implemented, has never been tested and it's likely to fail, and that is a reason we have only used one of the Humus demos until now ...
After that there is IL-2, World of Warcraft, Serius Sam 2, the new DOOM3 engine based games (I hope they work with few problems after DOOM3 is completely fixed) ... Which is better than a year ago when I thought we would have to go just with UT and DOOM3 and I was lamenting we didn't choose D3D as the primary framework.
What strikes me somewhat that review's say 7800 GT is beating the X1800XL ... and it is true. But by how much ? A few percents ? Mostly in the 60-80 fps level. I'd say there is no practical difference between the two. The 7800 GT is better in OpenGL and heavy stencil shadow apps, the X800XL has a much smaller performance drop at 4x AA.
I'd say the only handicap is the price. But I would not take price into account in a hardware review. Custumers can do the math. Between two similar competitive products, they will choose what's cheaper.
I'd say the only handicap is the price. But I would not take price into account in a hardware review. Custumers can do the math. Between two similar competitive products, they will choose what's cheaper.
Similar threads
