Radeon 9800 Pro preview

[aaXiom3D]

Dave, the issue is that the ATI driver internally seems to lower the R300 scores in order to show a bigger R350 advantage.

Has R300 performance gone down since the launch? No. So I believe your wording is poor.

Yep my english is poor (not my native language, I'm french ), but you probably understood what I said. The question is : why the R350 path is working with the R300 and get higher scores (compared to the R300 path) ?


Edit : typo

 

[Randell]

Are thsoe comaprisons between the same driver set on the same hardware? Just with the drivers recognising it in one run one as a 9700Pro and the other as a 9800Pro?

 

[Luminescent]

Believe it or not, R350 was tweaked a noticeable bit (timings, memory controller, hyper z, with some Ati magic if you know what I mean) from its original R300 form.

As for the drivers, I don't see how it would be possible to change anything more than Dave did with the 7.34 driver installation, but please post an explanation if you were successful in hacking through the system so those with an R350 board could know.

 

[Ichneumon]

Well, yes the 9800pro is of course "overclocked" from a 9700, but it got some significant enhancements... the stencil optimizations for HyperZ being of particular interest since that should help *clock-for-clock* get better performance in the oh-so-often-debated-lately DoomIII style rendering.

 

[Randell]

hmm the way I see it, Dave's figures still stand as his drivers reported his 9700Pro as a 9800.

Dave, did the control panel look teh same on the 9700pro, i.e. saying Smoothvision 2.1 etc?

 

[aaXiom3D]

Well, yes the 9800pro is of course "overclocked" from a 9700, but it got some significant enhancements... the stencil optimizations for HyperZ being of particular interest since that should help *clock-for-clock* get better performance in the oh-so-often-debated-lately DoomIII style rendering.

Agree with u

but still nobody did answer this :
why the R350 path is working with the R300 and get higher scores (compared to the R300 path) ?

 

[Dave Baumann]

I think the issue is that Tommi altered ati2mtag.sys file and got better scores for a 9700. I assume that what Tommi did was alter the sys file to make the R350 path available to R300, however essentially I did the opposite and make the drivers think that the R300 was an R350. With it reporting back as a "Radeon 9800 Series" I'd say that the drivers were installed and a 9800 (R350) and that the driver improvements are in place and the socres shown in the review are actually a reflection of the hardware increases.

 

[Althornin]

I think the issue is that Tommi altered ati2mtag.sys file and got better scores for a 9700. I assume that what Tommi did was alter the sys file to make the R350 path available to R300, however essentially I did the opposite and make the drivers think that the R300 was an R350. With it reporting back as a "Radeon 9800 Series" I'd say that the drivers were installed and a 9800 (R350) and that the driver improvements are in place and the socres shown in the review are actually a reflection of the hardware increases.

Which would also explain why you showed much less of a per clock performance increase (in most cases, other than pixel shaders) than AnteP did.

 

[MrB]

Here's how the Cat 3.2 RC4 inf looks like

"RADEON 9700 PRO" = ati2mtag_R300, PCI\VEN_1002&DEV_4E44
"RADEON 9700 PRO - Secondary" = ati2mtag_R300, PCI\VEN_1002&DEV_4E64
"RADEON 9800 PRO" = ati2mtag_R300, PCI\VEN_1002&DEV_4E48
"RADEON 9800 PRO - Secondary" = ati2mtag_R300, PCI\VEN_1002&DEV_4E68

they're basically calling the same path. Not sure how Thomas got improved results. When I get the chance I'll try this method and normal installation method and see what happens.

 

[Luminescent]

We believe you Dave. Now, what about testing the R350 with the current catalyst drivers released for the R300? Would this work? It would either make or break the "R350 is nothing but an overclocked R300, with some extra features" nonsense once and for all. I am almost completely sure that clock for clock, even without the hyper z, in pixel and vertex shader performance alone (when bandwith and the new memory controller are not a factor), the R350 is more optimized, like Dave concluded in the B3D review. I was told that timing tweaks were made to the R300 core for the R350, so its pipeline resources are probably put to use more efficiently.

This is aside from the major addition of the f-buffer which makes the R350 that much more flexible. It is capable of unlimited, unpenalized (no multipassing overhead) pixel instrutions per pass. Data-dependent and static branching in the pixel shader are now possible (although not entirely feasible).

 

[Brent]

Fair enough Dave. I'm not a current of any ATI product. And I'm thinking of getting a 9700 and using the "hacked" 9800 drivers (Thomas's) on it, plus overclocking it.

I believe this would negate any need for a 9800 for me at least.

the 9800 Pro can clock to 440Mhz/370Mhz stock

9700 pro can't

http://www.hardocp.com/article.html?art=NDM5

we might start to see more improvements in games that use the stencil buffer for things like shadow volumes etc... ala doom3, on the 9800 pro vs. 9700 pro

only time will tell there, plus the improved memory controller and f-buffer

 

[Slides]

I believe this would negate any need for a 9800 for me at least.

the 9800 Pro can clock to 440Mhz/370Mhz stock

9700 pro can't

I never said it could.

But the performance per $ for a 9700 is FAR greater than the 9800pro. Those who can afford the 9800, good for them, but not I.

 

[Nagorak]

Of course the R9700 Pro is a better deal. The R9800 is in the new overpriced card range.

 

[demonic]

the 9800 Pro can clock to 440Mhz/370Mhz stock

Bah, thats nothing

The guy who did the Hexus.net review...

This is what he wrote on a forum I goto... (few posts pieced together)

Just been having a further play with the 9800 Pro. This time with a fan blowing across it. The core seems to hit a ceiling at about 425 now, but the memory is running strong at 800MHz

Just done it at 1280x960, where the benchmark is even more card limited.

UT2K3 1280x960x32 Flyby 4xAA, 8xAF:

R350 @ 420/770 = 125.2
R350 @ 380/680 = 111.8
R300 @ 380/680 = 101.6
R300 @ 325/620 = 89.8
FX @ 500/1000 = 86.6

That's a gain of:

12% over a stock 9800 Pro
23% over an overclocked 9700 Pro (380/680)
40% over a stock 9700 Pro
45% over a stock FX Ultra.

I just wish those ultras that come with +400Mhz Core and +400Mhz Ram, only came with 128Mb... I don't think we need 256Mb cards just yet, and also it makes then v. expensive

 

[Himself]

Not to step on anybody's pet, but inf level mods or forced installs do not prevent a driver from reading the device id and running different bits of code as a result. The device id is used for things like preventing overclocking and such in ATI's drivers already from what I understand. The inf stuff is for plug and play detection at install time, that's it, hacking the inf or forcing the install is no different. Your enum registry will probably still show the real deviceid in both cases, and even if it didn't the drivers could read the PCI config data to get at it.

I suspect TB is hacking the driver so that it reads the device id as being 350, that's if his results are repeatable.

 

[Ante P]

Not to step on anybody's pet, but inf level mods or forced installs do not prevent a driver from reading the device id and running different bits of code as a result. The device id is used for things like preventing overclocking and such in ATI's drivers already from what I understand. The inf stuff is for plug and play detection at install time, that's it, hacking the inf or forcing the install is no different. Your enum registry will probably still show the real deviceid in both cases, and even if it didn't the drivers could read the PCI config data to get at it.

I suspect TB is hacking the driver so that it reads the device id as being 350, that's if his results are repeatable.

That's why he needs to make it public so that us "previewers" can run the tests again.

 

[tb]

Hi!

Here are the changes:

C:\>comp ati2mtag.sys modr98k_ati2mtag.sys

//crc signature
OFFSET 148
Datei1 = 76
Datei2 = 7E

//asic id's from 444e to 484e
OFFSET 672C4
Datei1 = 44
Datei2 = 48
OFFSET 67318
Datei1 = 44
Datei2 = 48
OFFSET 6E864
Datei1 = 44
Datei2 = 48
OFFSET 6E8DC
Datei1 = 44
Datei2 = 48
OFFSET 6E99C
Datei1 = 44
Datei2 = 48

//asic id's from 484e to 444e
OFFSET 67390
Datei1 = 48
Datei2 = 44
OFFSET 6EE34
Datei1 = 48
Datei2 = 44
OFFSET 6EEF4
Datei1 = 48
Datei2 = 44

File size of my file is 576128 bytes.

Copy this changed file over the one in the windows\system32\drivers directory an restart.

Thomas

 

[Ante P]

Hi!

Here are the changes:

C:\>comp ati2mtag.sys modr98k_ati2mtag.sys

---

So what about uploading the modified file for stupid people like me who won't know what to do with the info you just posted?

 

[THe_KELRaTH]

I suppose the other option would be to change the 9700Pro Bios to show it as a 9800Pro.

 

[sabeehali]

Hi!

Here are the changes:

C:\>comp ati2mtag.sys modr98k_ati2mtag.sys

---

So what about uploading the modified file for stupid people like me who won't know what to do with the info you just posted?

and please do it fast so antep can re-preview the 9800 (using the new 9700) and we can see if the results are different from B3d.
Thanks


/me waiting impatiently