Radeon 9600 PRO Overclocking
- Thread starter Dave Baumann
- Start date
OpenGL guy
Veteran
300 -> 340 is 13% memory overclock... 3D Mark 03 results consist of 4 tests. If the 4 tests are largely not memory bandwidth limited, then you won't get much benefit from the extra bandwidth.David G. said:WTF is this :
@537/300 gets 3816 3DMarks in 3DMark03
@537/340 gets 3860 3DMarks in 3DMArk03 ?!!??!
Only 44 3DMarks with almost 40% overclock ?
Why ??? :?
OpenGL guy
Veteran
No idea, I'm just giving informationgibz0r said:
But to put it in perspective: 9500 Pro at default clocks = 2.2 gigapixels/sec and 8.6 GB/s. 9600 Pro at 537/340 = 2.1 gigapixels/s and 10.9 GB/s. It's pretty clear that the 9500 Pro is a lot more bandwidth limited than the overclocked 9600 Pro.
Hier-Z does give it quite some alleviation on the BW side though.
OpenGL guy
Veteran
It depends completely on what application you are running...DaveBaumann said:
Good point. I'd really like some benchmarks with hyper-z enabled/disabled on the same card. GL_ext_reme and villagemark are nice, but what about real games? I think it's easy enough to switch off hierarchical-z with a tweaker so this could be tested easily. This would answer some questions, for instance if a radeon 8500 with hierarchical-z disabled is still faster than a 9000 at the same clock speeds (everybody just assumes it's faster because of the 2nd texture unit, but I have some doubts there), or if a 9500pro without hier-z would slow down almost to the level of a 9500 non-pro. But then again, maybe I'm the only one who is interested in thatOpenGL guy said:
Bull crap.David G. said:nelg said:I think some people might owe TSMC an apology.
Here's something I've red recently @ rage3d forums :
Here's what I've posted more than 2 weeks ago :
Link
Makes me look very sma*t errrr .... I mean intuistic , doesn't it ?
Hehe ... just joking ...
This has nothing to do with TSMC .
The fab has plenty to do with whether or not a chip works. More than plenty. They provide all the data for the sims you run and if they've got it wrong--then your chip doesn't work as advertized. I worked on a design that spent six months in limbo because the fab wouldn't own up to selling an SRAM cell that was never qualified in production.
My goodness, that's an impressive overclock for a card that's not using an external power connector. Perhaps power draw (in addition to profit margins) was the reason for ATi's "conservative" clocking?
Also, have you considered using Tenebrae as a (legal ) way to foresee Doom III performance?
The Zalman-cooled Sapphire 9600PUD) looks pretty swank, BTW. A 500/330 silent 9600PU is a nice alternative to a noisier, hungrier 9500P, particularly at the same price. I can't wait for 9600P's to begin to drop in price--the 9600 series may turn out to be an excellent and long-lived part for ATi, if Russ' info on TSMC's improved margins is correct.i
Also, have you considered using Tenebrae as a (legal
) way to foresee Doom III performance?The Zalman-cooled Sapphire 9600PU
D) looks pretty swank, BTW. A 500/330 silent 9600PU is a nice alternative to a noisier, hungrier 9500P, particularly at the same price. I can't wait for 9600P's to begin to drop in price--the 9600 series may turn out to be an excellent and long-lived part for ATi, if Russ' info on TSMC's improved margins is correct.iDave, that's an impressive o/c speed.
Can you tell me what the weather/temperature is like in London now? It's pretty hot and humid here in South East Asia (and other parts of the world), a growing market... and this kinda thing are primary disclaimers-inducing stuff when it comes to o/c'ing. While I have no doubt that o/c'ing the 9600PRO is a certainty no matter where you are (even in the Equator-zone countries), what can enthusiasts from such countries/zones gather from this thread?
Can you tell me what the weather/temperature is like in London now? It's pretty hot and humid here in South East Asia (and other parts of the world), a growing market... and this kinda thing are primary disclaimers-inducing stuff when it comes to o/c'ing. While I have no doubt that o/c'ing the 9600PRO is a certainty no matter where you are (even in the Equator-zone countries), what can enthusiasts from such countries/zones gather from this thread?
all I have to say is HOLY MOLY. Those cards are clocking @ >500 with a STOCK heatsink and no external power connecter. Where did nvidia go SOOO wrong with their .13 micron that they could not achieve these same results with any of the FX family members? Is this a sign of things to come with Ati's response to the NV35?
OpenGL guy said:300 -> 340 is 13% memory overclock... 3D Mark 03 results consist of 4 tests. If the 4 tests are largely not memory bandwidth limited, then you won't get much benefit from the extra bandwidth.David G. said:WTF is this :
@537/300 gets 3816 3DMarks in 3DMark03
@537/340 gets 3860 3DMarks in 3DMArk03 ?!!??!
Only 44 3DMarks with almost 40% overclock ?
Why ??? :?
I've pulled that number out of my a** . Sorry guys , I was thinking @ the VPU speed increase .
galperi1 said:all I have to say is HOLY MOLY. Those cards are clocking @ >500 with a STOCK heatsink and no external power connecter. Where did nvidia go SOOO wrong with their .13 micron that they could not achieve these same results with any of the FX family members? Is this a sign of things to come with Ati's response to the NV35?
From what little I know about processors, here's a possible (at least in my mind) reason. Some parts of the chip will be able to run much faster than other parts, but you can only clock it as fast as that slowest part can go. There's probably quite a bit (compared to most chips) of buggy sections of NV30 that limit its overall speed.
Maybe someone with an engineering background and actual knowledge can shed some light on my idea.
Reverend said:
Yesterday was the hottest day of the year! 25C+ where I live (which is pretty close to Dave in relation to the size of the known universe). Even the evening was pretty warm...
It was so nice I even went OUTSIDE for a bit.
You can see now why M10's "OVERDRIVE" is a marketable feature... the clock headroom is massive, even with very modest cooling.
MuFu.
Yeah, as MuFu says, it was unusually hot for April over here (hottest day in April since records began AFAIK) and the test room was also very toasty. If I were to run it now (morning, and the windows been open all night) there might be a little more to squeeze out of it.