Is 9500 + Soft9700 faster than 9700 PRO?

blcjr

Golden Member
Oct 28, 1999
1,010
0
0
I'm one of those fortunate ones whose 9500 could be successfully patched with Soft9700. At (then) less than half the cost of a 9700 PRO, I figured it as a marvelous bang for the buck, even if it could not quite match a true 9700 PRO as far as core/memory speeds are concerned (especially the memory speed). But I'm seeing such dramatic 3dMark2001SE scores relative to what others are posting, taking into account other system differences, that is has me wondering if my mod didn't turn out to produce something faster than a 9700 PRO.

To understand the context of my puzzlement, I've been comparing my 3dMark2001SE scores with chizow's. With an Athlon XP 2100+ T-bred B o'clocked to 2401 Mhz, chizow reports 15622 3d marks. With my more modest overclock of the same chip to 2090 Mhz, I'm getting 16141 3d marks. With so many other components of the equation so similar -- nForce2 chipset mobos, almost identical FSB's -- how is that I can have a score higher than chizow's with over 300 Mhz less of CPU speed? Puzzling, to say the least.

In viewing the details of the 3dMark2001SE scores since I did the Soft9700, one thing consistently stands out: the reported multitexture fill rate. I'm no Johnny Guru when it comes to understanding these things, but supposedly the absolute maximum is 2500 MTexels/s. But I'm consistently seeing numbers greater than 2700 MTexels/s since doing the Soft9700 mod. Would this give anybody who does understand these things a clue as to why I seem to be seeing better performance with a 9500 + Soft9700 than others are getting with a true 9700 PRO.

I don't know if this is the explanation or not, but for what it's worth, I'm overclocking the 9500 to 351/297. That's a little better than the default 9700 PRO for the core speed, but a little less for the memory speed. Could that higher core speed account for the higher multitexture fill rate? It has been a while since I fiddled with the core/memory speeds on the 9500, but I don't recall the overclock making a noticeable difference in 3d marks.

I also note that chizow is running DX 9.0, and I'm running 8.1, and that he's enabled 8x AGP while I'm using 4x AGP. Could there be something about 3DMarks2001SE being optimized for DX 8.1 that would cause scores to be lower with DX 9.0, all other things equal?

All things considered, I keep coming back to the notion that my 9500 + Soft9700 mod yielded something faster than a default 9700 PRO. Is that possible?

Finally, lest anybody is wondering, the score is not a fluke. I just ran it again, and got 16105. And earlier this morning I bumped the multiplier up to 11.5, and got 16504. But any time I go above 11x190FSB, I run into Prime95 errors, so I'm not going to "publish" any such scores.

 

blcjr

Golden Member
Oct 28, 1999
1,010
0
0
Originally posted by: SickBeast
to answer your question: NO.

the 9700pro uses faster memory.
So what's your explanation for the faster fill rate? Is it simply a function of me overclocking the core to 351?

-Baz

 

blcjr

Golden Member
Oct 28, 1999
1,010
0
0
Originally posted by: Killrose
Maybe DX9 slows down in 3DMark200se because it is a DX8.1 test?
In another thread, chizow says that 9.0 3d marks are about 1k lower than 8.1, so that's probably the main difference.

 

AnAndAustin

Platinum Member
Apr 15, 2002
2,112
0
0
;) The diff could be what's installed in Windows and what's running in the background. Otherwise I'd seriously expect his mod of the 9500 (if that is what he/chizow is using) simply didn;t enable all 8 pipes, something which can be hard to identify. Obviously don't forget a Rad9700PRO will easily o/c better than a Rad9500 .... added to which a Rad9500PRO (8 pipes guaranteed) yields Rad9700 perf when o/c'ed, even with its 128bit DDR.
 

SickBeast

Lifer
Jul 21, 2000
14,377
18
81
So what's your explanation for the faster fill rate? Is it simply a function of me overclocking the core to 351?
Yes, that would seem to be the only possible explanation in my books.

Congrats on the mod by the way...I've heard that the success rate isn't all that great.
 

Viper96720

Diamond Member
Jul 15, 2002
4,390
0
0
The only scores that 3dmark uses are the game scores anything else won't affect the score. You would get similar scores running all the tests and just the game ones.
 

Rand

Lifer
Oct 11, 1999
11,071
1
81
Originally posted by: blcjr
I'm no Johnny Guru when it comes to understanding these things, but supposedly the absolute maximum is 2500 MTexels/s. But I'm consistently seeing numbers greater than 2700 MTexels/s since doing the Soft9700 mod. Would this give anybody who does understand these things a clue as to why I seem to be seeing better performance with a 9500 + Soft9700 than others are getting with a true 9700 PRO.
.

The peak theoretical Multitexturing fill rate of a stock ATi Radeon 9700 Pro is 2600MT/s, or 2.6GT/s.

Given that your card is clocked your at 351/297, your reported scores of 2700MT/s+ are not at all unusual. The 8x1 R300 at 351MHz would have a peak MultiTextured fillrate of 2808MT/s, so seeing scores of 2700MT/s in synthetic benchmarks shouldnt be unexpected at all.
 

hdeck

Lifer
Sep 26, 2002
14,530
1
0
Originally posted by: AnAndAustin
;) The diff could be what's installed in Windows and what's running in the background. Otherwise I'd seriously expect his mod of the 9500 (if that is what he/chizow is using) simply didn;t enable all 8 pipes, something which can be hard to identify. Obviously don't forget a Rad9700PRO will easily o/c better than a Rad9500 .... added to which a Rad9500PRO (8 pipes guaranteed) yields Rad9700 perf when o/c'ed, even with its 128bit DDR.
chiz is using a 9700 pro, not a modded card.
 

ASK THE COMMUNITY

TRENDING THREADS