- Joined
- Jul 25, 2001
- Location
- Rovaniemi, Finland
Yesterday I was asking if I could gain anything from OC'n my Geforce DDR. I got some spare time so I tested it.
First I built better cooler from old 486 sink and the original fan from my ORB (6000rpm). This blob was enough to take up 2 PCI slots... Temperatures reported by the ASUS smartdoctor set below 50C @ default 120Mhz core speed. At this time of year, the ambient room temp is close to 20C (not measured). The tests were run with open case.
For testing I used 3Dmark2001 poly troughput and fillrate tests. Poly tests were run at 640x480x16 to stress the core and fillrate test at 1024x768x32 to stress the mem. Also, mem OC'n was done with core @150. Test system is 1.2Ghz Tbird, 256Mb, MSI K7T Turbo, Windoze98, nvidia ref. driver 6.50.
I got the core up to 150Mhz, where 155Mhz produced plenty artifacts. The sink felt cool when finger tested, and smartdoctor reported 55C under bencmarking, so I presume that Id need better thermal goo or even a TEC to get temps lover than that. I was able to benchmark the memory up to 370Mhz, where it produced some artifacts. 360 was stable but produced slight artifacts only in the game3 @1024x768x32, but ran it perfectly at 800x800x32. Will investigate tomorrow.
I dont have access to any web space so I cant post any graphs. If you want the whole .xls sheet, mail me.
I recorded all data at 10Mhz increases. 130, 140, 150 for core and 310, 320, 330, 340, 350, 360, 370 for mem. I also calculated the theoretical values from direct Mhz increase. The results were good.
Core OC gain was measured with 3Dmark2001 polygon troughtput test.
120Mhz: 8,3Mpolys/sec for 1 light and 1,8Mpolys/sec for 8 lights
130Mhz: 8,7 and 1,9
140Mhz: 9,1 and 2,1
150Mhz: 9,4 and 2,2
150Mhz theoretical (Mhz vise): 10,9 and 2,4
The charts show that slight OC to 130Mhz gains only some 5% for both, 1 and 8 lights, where OC to 140 and 150Mhz gain 10%1L, 16%8L and 13%1L, 22%8L. Mhz vise the 150Mhz core should gain 31% - 32%.
Mem OC gain was measured with 3Dmark2001 fillrate test.
300Mhz: 244,1Mpix/sec for single texturing and 441,2 for double texturing.
310Mhz: 249,2 and 446,3
320Mhz: 256.4 and 480.7
330Mhz: 263.9 and 493.8
340Mhz: 271 and 505.7
350Mhz: 279.2 and 518.1
360Mhz: 285.1 and 525.8
370Mhz: 293.8 and 536.1 *artifacts*
360Mhz theoretical (Mhz vise): 292.9 and 529.4
Looking at the charts I found out that OC'n the memory proved much more efficient; all speeds were within 3% from Mhz vise values and up to 350Mhz double texturing was producing more gain than Mhz vise values! If the 360Mhz proves stable the gain will be 17% and 19%. At 370Mhz gains were 21% and 22%. Also, unlike core OC'n, mem OC'n produced clear linear rise in performance all the way from 310Mhz to 370Mhz, single texturing raising slightly faster.
Clearly OC'n the old geforce gives expected results. Specially with memory the fillrate grows just as expected. The core scalability starts to fade @150Mhz but still produces clear benefit from OC'n. So my conclusion is that it is worth it as I got some 20% increase in specific performances. The 3Dmark2001 overall score grew 11%, not as much as specific tests, likely because of the many tests that use CPU, like the vertex shader perf. (software emulation) and sprite rendering (soft.emu. also). My current score at default settings is now 2780.
First I built better cooler from old 486 sink and the original fan from my ORB (6000rpm). This blob was enough to take up 2 PCI slots... Temperatures reported by the ASUS smartdoctor set below 50C @ default 120Mhz core speed. At this time of year, the ambient room temp is close to 20C (not measured). The tests were run with open case.
For testing I used 3Dmark2001 poly troughput and fillrate tests. Poly tests were run at 640x480x16 to stress the core and fillrate test at 1024x768x32 to stress the mem. Also, mem OC'n was done with core @150. Test system is 1.2Ghz Tbird, 256Mb, MSI K7T Turbo, Windoze98, nvidia ref. driver 6.50.
I got the core up to 150Mhz, where 155Mhz produced plenty artifacts. The sink felt cool when finger tested, and smartdoctor reported 55C under bencmarking, so I presume that Id need better thermal goo or even a TEC to get temps lover than that. I was able to benchmark the memory up to 370Mhz, where it produced some artifacts. 360 was stable but produced slight artifacts only in the game3 @1024x768x32, but ran it perfectly at 800x800x32. Will investigate tomorrow.
I dont have access to any web space so I cant post any graphs. If you want the whole .xls sheet, mail me.
I recorded all data at 10Mhz increases. 130, 140, 150 for core and 310, 320, 330, 340, 350, 360, 370 for mem. I also calculated the theoretical values from direct Mhz increase. The results were good.
Core OC gain was measured with 3Dmark2001 polygon troughtput test.
120Mhz: 8,3Mpolys/sec for 1 light and 1,8Mpolys/sec for 8 lights
130Mhz: 8,7 and 1,9
140Mhz: 9,1 and 2,1
150Mhz: 9,4 and 2,2
150Mhz theoretical (Mhz vise): 10,9 and 2,4
The charts show that slight OC to 130Mhz gains only some 5% for both, 1 and 8 lights, where OC to 140 and 150Mhz gain 10%1L, 16%8L and 13%1L, 22%8L. Mhz vise the 150Mhz core should gain 31% - 32%.
Mem OC gain was measured with 3Dmark2001 fillrate test.
300Mhz: 244,1Mpix/sec for single texturing and 441,2 for double texturing.
310Mhz: 249,2 and 446,3
320Mhz: 256.4 and 480.7
330Mhz: 263.9 and 493.8
340Mhz: 271 and 505.7
350Mhz: 279.2 and 518.1
360Mhz: 285.1 and 525.8
370Mhz: 293.8 and 536.1 *artifacts*
360Mhz theoretical (Mhz vise): 292.9 and 529.4
Looking at the charts I found out that OC'n the memory proved much more efficient; all speeds were within 3% from Mhz vise values and up to 350Mhz double texturing was producing more gain than Mhz vise values! If the 360Mhz proves stable the gain will be 17% and 19%. At 370Mhz gains were 21% and 22%. Also, unlike core OC'n, mem OC'n produced clear linear rise in performance all the way from 310Mhz to 370Mhz, single texturing raising slightly faster.
Clearly OC'n the old geforce gives expected results. Specially with memory the fillrate grows just as expected. The core scalability starts to fade @150Mhz but still produces clear benefit from OC'n. So my conclusion is that it is worth it as I got some 20% increase in specific performances. The 3Dmark2001 overall score grew 11%, not as much as specific tests, likely because of the many tests that use CPU, like the vertex shader perf. (software emulation) and sprite rendering (soft.emu. also). My current score at default settings is now 2780.
.
