OK, here is my 800XL over-clock results, and why I report them as I do.
First, I only am reporting the VPU and memory clock verses FPS against any benchmarks since FPS is subsystem related. The same VPU clock will perform FPS wise, more than likely, better on a newer PC than mine (an older ASUS P4PE with the 865 chipset). But, the VPU should over-clock the same between PC’s, or no? Do we have any comments on that statement?
RESULTS
X800XL 400/1000 (stock) 536/1000 (over-clock) – Sapphire 256 Meg retail part.
X800XT 500/1000 (stock)
The question is, can the 110nm XL part, under the same conditions, overlap the 130nm XT’s over-clock rates? Some members have done well with liquid cooling for just plain VP clock increases, but the question remains; can the 800XL overlap and exceed a higher rated parts over-clock?
To put it another way, and to clarify (Scottchen), if I buy a X800XT and an X800XL, will a reliable XL population overlap the over-clock distribution that the XT part achieves and what are the odds of buying a card that fits that standard deviation and overlap? This is a good question because the XL is a smaller core process than the XT, and it seems that it would clock better. But, the XL uses a poorer dielectric than the XT, so some of the expected performance is mitigated by the dielectric choice. Still, some parts are able to surpass rated specs pretty easily, and well into the realm of the “better” part. The old Celeron 300MHz CPU is a prime example. Most would run at 450MHz with just a jumper change. Yes, the higher rated part was just binned, and the CPU’s were the really same. Intel just got so good no real 300MHz parts were really made anymore. The X800 XL and XT VPU differences makes the gamble harder to quantify, however.
I do NOT have any data other than my card, and the 36MHz VPU core clock sure doesn’t seem like most any XL can exceed the XT in pure performance on a same-same basis.
Another good experiment would be to compile the data for over-volt biased cards (here we go again, Scottchen) to see if that will create the overlap that would signify that this card is a deal if over voltage bias is OK. Some issue may arise from how much voltage, however. What’s fair here? And no, I don’t want to include water-cooling VPU’s because systems are so different and expensive to buy that more people can benefit from the results of the air cooled tests. Or, we can add water cooling, too?
Some of the issues that might bias this table, depending on who wants to help fill it out, are what is considered an over-clock on the VPU?
I considered an over-clock successful if I can play Far Cry several hours with no lock-ups or visual anomalies. I used patch 1.1 and do have some questions;
1.0 I experienced a checkerboard pattern in Far Cry that wiped out rendering on every other square, alternating between each row, till I dropped into windows XP Home. Then, the rendering error was still stuck on 3D type desktop areas (title bars above programs and icons). If I rebooted, the VPU seemed to reset itself and looked fine in XP, till it would crash again in the game after minutes to hours.
1.1 I do not fully know what the error is from, but can guess that one side of the VPU feeding the double pumped memory is sending bit errors to the memory and the memory once corrupted, freezes. I rebooted, and dropped the clock back 6MHz until it was stable (see point number two) as long as I played the game.
1.2 I didn’t over-clock the memory because the Samsung memory chips seem almost too sensitive to reliably over-clock, and the benefits were pretty small on my benchmarks (10 mark increases on 3Dmark03 from 1000MHz to 1012MHz). And, the XT part is rated the same. Does it use the same memory?
2.0 I posted a question about rendering errors and got no responses. In the game, Far Cry, some objects like helicopters or jeeps would render bright yellow. I can’t fairly say that my card is running at 536MHz until I know that this is a software error, or a hardware error. I report the 536MHz number because it also does it in stock form making me thing that it is a driver related rendering error. But, I could have a bad card, doubt it, but I could. The checkerboard error in 1.0 above though, is definitely the card.
2.1 Is my over-clock real? Comments.
3.0 If we leave the memory speed alone, I’m assuming that the results will be easier to define, and test. I do not know if upping the memory really hurts the VPU clock side’s ultimate over-clock. But, if the memory is the same, it should still be a fair comparison.
4.0 Be careful of so called over-clock utilities like Rad Clocker. Mine tests WAY optimistic compared to freeze free gaming. It reported 545MHz, but my VPU would turn the monitor into a Betty Crocker picnic tablecloth checkerboard pattern faster than you can say, “here come the ants!” So use a harder game engine somehow.
If enough people fill in the table, statistics can reliably determine the distribution of XL and XT performance, and how big the overlap is. If the overlap were pretty large (like the 300MHz Celeron example) a good bet would be to buy the XL part. Or, you might see if your results significantly improve with more voltage against a similarly bias over-voltage XT part. Then again, it may say to get really better speeds, just get the XT part because the overlap is too small to expect truly bargain bin performance increases.
The two tables will have two sets of entries,
XL and XT with stock voltages and VPU speeds they achieve.
XL and XT with over-voltages and the VPU speeds they achieve.
From this, statistics can calculate the distribution, overlap, and percentage of the population that can be expected inside the overlap.
ATI already does this at marketed speeds. We just want to do it at over-clocked speeds.
Does anybody have a body of data to start with? Is the GreatApe able to improve out DOE? Scottchen Et all?
The trick to procrastination is to just not do it very long.<P ID="edit"><FONT SIZE=-1><EM>Edited by rower30 on 08/01/05 09:24 PM.</EM></FONT></P>
First, I only am reporting the VPU and memory clock verses FPS against any benchmarks since FPS is subsystem related. The same VPU clock will perform FPS wise, more than likely, better on a newer PC than mine (an older ASUS P4PE with the 865 chipset). But, the VPU should over-clock the same between PC’s, or no? Do we have any comments on that statement?
RESULTS
X800XL 400/1000 (stock) 536/1000 (over-clock) – Sapphire 256 Meg retail part.
X800XT 500/1000 (stock)
The question is, can the 110nm XL part, under the same conditions, overlap the 130nm XT’s over-clock rates? Some members have done well with liquid cooling for just plain VP clock increases, but the question remains; can the 800XL overlap and exceed a higher rated parts over-clock?
To put it another way, and to clarify (Scottchen), if I buy a X800XT and an X800XL, will a reliable XL population overlap the over-clock distribution that the XT part achieves and what are the odds of buying a card that fits that standard deviation and overlap? This is a good question because the XL is a smaller core process than the XT, and it seems that it would clock better. But, the XL uses a poorer dielectric than the XT, so some of the expected performance is mitigated by the dielectric choice. Still, some parts are able to surpass rated specs pretty easily, and well into the realm of the “better” part. The old Celeron 300MHz CPU is a prime example. Most would run at 450MHz with just a jumper change. Yes, the higher rated part was just binned, and the CPU’s were the really same. Intel just got so good no real 300MHz parts were really made anymore. The X800 XL and XT VPU differences makes the gamble harder to quantify, however.
I do NOT have any data other than my card, and the 36MHz VPU core clock sure doesn’t seem like most any XL can exceed the XT in pure performance on a same-same basis.
Another good experiment would be to compile the data for over-volt biased cards (here we go again, Scottchen) to see if that will create the overlap that would signify that this card is a deal if over voltage bias is OK. Some issue may arise from how much voltage, however. What’s fair here? And no, I don’t want to include water-cooling VPU’s because systems are so different and expensive to buy that more people can benefit from the results of the air cooled tests. Or, we can add water cooling, too?
Some of the issues that might bias this table, depending on who wants to help fill it out, are what is considered an over-clock on the VPU?
I considered an over-clock successful if I can play Far Cry several hours with no lock-ups or visual anomalies. I used patch 1.1 and do have some questions;
1.0 I experienced a checkerboard pattern in Far Cry that wiped out rendering on every other square, alternating between each row, till I dropped into windows XP Home. Then, the rendering error was still stuck on 3D type desktop areas (title bars above programs and icons). If I rebooted, the VPU seemed to reset itself and looked fine in XP, till it would crash again in the game after minutes to hours.
1.1 I do not fully know what the error is from, but can guess that one side of the VPU feeding the double pumped memory is sending bit errors to the memory and the memory once corrupted, freezes. I rebooted, and dropped the clock back 6MHz until it was stable (see point number two) as long as I played the game.
1.2 I didn’t over-clock the memory because the Samsung memory chips seem almost too sensitive to reliably over-clock, and the benefits were pretty small on my benchmarks (10 mark increases on 3Dmark03 from 1000MHz to 1012MHz). And, the XT part is rated the same. Does it use the same memory?
2.0 I posted a question about rendering errors and got no responses. In the game, Far Cry, some objects like helicopters or jeeps would render bright yellow. I can’t fairly say that my card is running at 536MHz until I know that this is a software error, or a hardware error. I report the 536MHz number because it also does it in stock form making me thing that it is a driver related rendering error. But, I could have a bad card, doubt it, but I could. The checkerboard error in 1.0 above though, is definitely the card.
2.1 Is my over-clock real? Comments.
3.0 If we leave the memory speed alone, I’m assuming that the results will be easier to define, and test. I do not know if upping the memory really hurts the VPU clock side’s ultimate over-clock. But, if the memory is the same, it should still be a fair comparison.
4.0 Be careful of so called over-clock utilities like Rad Clocker. Mine tests WAY optimistic compared to freeze free gaming. It reported 545MHz, but my VPU would turn the monitor into a Betty Crocker picnic tablecloth checkerboard pattern faster than you can say, “here come the ants!” So use a harder game engine somehow.
If enough people fill in the table, statistics can reliably determine the distribution of XL and XT performance, and how big the overlap is. If the overlap were pretty large (like the 300MHz Celeron example) a good bet would be to buy the XL part. Or, you might see if your results significantly improve with more voltage against a similarly bias over-voltage XT part. Then again, it may say to get really better speeds, just get the XT part because the overlap is too small to expect truly bargain bin performance increases.
The two tables will have two sets of entries,
XL and XT with stock voltages and VPU speeds they achieve.
XL and XT with over-voltages and the VPU speeds they achieve.
From this, statistics can calculate the distribution, overlap, and percentage of the population that can be expected inside the overlap.
ATI already does this at marketed speeds. We just want to do it at over-clocked speeds.
Does anybody have a body of data to start with? Is the GreatApe able to improve out DOE? Scottchen Et all?
The trick to procrastination is to just not do it very long.<P ID="edit"><FONT SIZE=-1><EM>Edited by rower30 on 08/01/05 09:24 PM.</EM></FONT></P>