My turn to post a build I guess, wasted way too much time on this one…
First proper SLI board I get so wanted to play around a little bit since I got a couple 8800 GTS.
Everything done with random parts I had lying around.
Specs :
Cooling:
Had to cut some fins from the southbridge heatsink for the second graphics card to fit.
CPU heatsink was more of a pain, in the HP dc7700 it screws directly into the mainboard tray.
So I had to use a rear bracket (got one without heatsink for some reasons), but the nuts were too small.
Drilled and re-threaded them and now the heatsink screws do fit.
Also had to remove a piece of plastic from the heatsink which didn't fit.
Had to cut/adjust some fins of the northbridge heatsink as well.
At first I tried using some fancy Zalman heatsink but it was basically useless ( https://www.badcaps.net/forum/attach...1&d=1684603207 ). Also it had no PWM control so it was running at full speed (the board can only adjust voltage for chassis fan, CPU fan is controller with PWM only).
For the front fan, I cut an opening in the chassis since it was restricting airflow way too much, the front cover mesh is already pretty dense.
I was missing modular cables for PSU so I used Molex adapters instead for the second graphics card and the HDD, I absolutely hate them but they do the job…
One of the graphics card was dead when I got it. Basically heavy liquid damage, had to rework a lot of solder joints on the back, fix some traces, replace some caps, also reballed a VRAM chip. Did that a while ago, it was working fine but when I picked it again it didn't work anymore. Something was pulling the PCIe reset signal low, also preventing the motherboard from POSTing. In the end I couldn't really figure out what the problem was (no schematics/boardview for this card), it would come good again for a short time when injecting some voltage, so as a last resort I pushed the voltage above what I should and for now that took care of it (don't do this at home).
Motherboard also had several problems.
Very large ripple on CPU VCore ( https://www.badcaps.net/forum/attach...1&d=1684603334 ), caps were good, it was actually the buck controller (ADP3198) that wasn't regulating properly, so replaced that and that took care of this issue. I also took the opportunity to try to reduce Vdroop which can be quite large on this board, I observed more than 0.15V difference between idle and stress test, not that great for overclocking.
Bridged one of the resistor between CSSUM and CSCOMP, not the correct way to do it according to the datasheet, but much easier that adding a voltage divider on LLSET. Difference is less than 0.02V now.
Next issue was that the audio codec wasn't detected. Replaced the ALC883 with an ALC888 and that's fixed too.
Final issue was with Ethernet. Gigabit mostly didn't work even though it got uplink. 100M was mostly working but not 100% stable. It was both a hardware and a software issue. Software issue was the crappy driver from Nvidia, the Microsoft one works much better. Hardware issue was weird. Booting from PXE was extremely slow so for sure it wasn't just software. I replaced the Ethernet PHY IC (88E1116), didn't fix it. I replaced the ports stack (2xUSB+RJ45 which also contains the Ethernet transformer), didn't fix it either but at that point I realized that the issue was mostly on download using iperf3, so unlikely to be an issue on the physical link after the PHY, more likely it was a problem on the RGMII interface between PHY and MAC (Nvidia southbridge here). Started probing, touched up some solder joints and that fixed it. Couldn't get to the bottom of it here either…
All components came from donor boards, for once the crap I keep that should go to the trash bin has been helpful…
Now comes overclocking.
Voltages:
Frequencies:
3.32 GHz on the CPU is unstable, this is an FSB ceiling because it's the same when CPU multiplier and RAM ratio are decreased. Increasing CPU or northbridge voltage doesn't help.
RAM timings:
Other timings are set to auto. I'm not experienced with fine-tuning RAM timings, but this does just a little bit better than stock at least.
On the software side, sadly only Windows 7 works with SLI. 64-bit of course. Windows 10 crashes when driver loads for both graphics cards (if one is disabled it works). Windows 11 always crashes with random blue screens, SLI or not.
Nvidia drivers version 342.01.
CineBench R15 scores 360.
Unigine Tropics set to DirectX 10, 1280x720 window, 4x anisotropic, trilinear filtering ,shaders&textures to high, ambient occlusion&reflection enabled, no anti-aliasing.
Scores 1640, 65.1 FPS with SLI.
Scores 1034, 41.1 FPS without SLI.
Sorry for the wall of text, this was a great waste of time for a useless machine.
I have no use for it, it's stuck on Windows 7 for SLI, but I don't feel like just giving it away to someone random since I spent so much time on it…
First proper SLI board I get so wanted to play around a little bit since I got a couple 8800 GTS.
Everything done with random parts I had lying around.
Specs :
- CPU: Core 2 Quad Q9450
- RAM: 2x2GB KHX6400D2K2
- GPUs: 2x Nvidia GeForce 8800 GTS G80 320MB
- MB: P5N-E SLI
- PSU: Antec Basiq Power 550 Plus
- HDD: Hitachi Ultrastar A7K2000 1TB
- ODDs: DVD ROM, DVD RW
- Case: not sure
Cooling:
- 92mm rear fan
- 120mm front fan
- Additional heatsink on southbridge
- Additional heatsink on CPU VCore MOSFETs
- CPU heatsink from HP dc7700 MT
Had to cut some fins from the southbridge heatsink for the second graphics card to fit.
CPU heatsink was more of a pain, in the HP dc7700 it screws directly into the mainboard tray.
So I had to use a rear bracket (got one without heatsink for some reasons), but the nuts were too small.
Drilled and re-threaded them and now the heatsink screws do fit.
Also had to remove a piece of plastic from the heatsink which didn't fit.
Had to cut/adjust some fins of the northbridge heatsink as well.
At first I tried using some fancy Zalman heatsink but it was basically useless ( https://www.badcaps.net/forum/attach...1&d=1684603207 ). Also it had no PWM control so it was running at full speed (the board can only adjust voltage for chassis fan, CPU fan is controller with PWM only).
For the front fan, I cut an opening in the chassis since it was restricting airflow way too much, the front cover mesh is already pretty dense.
I was missing modular cables for PSU so I used Molex adapters instead for the second graphics card and the HDD, I absolutely hate them but they do the job…
One of the graphics card was dead when I got it. Basically heavy liquid damage, had to rework a lot of solder joints on the back, fix some traces, replace some caps, also reballed a VRAM chip. Did that a while ago, it was working fine but when I picked it again it didn't work anymore. Something was pulling the PCIe reset signal low, also preventing the motherboard from POSTing. In the end I couldn't really figure out what the problem was (no schematics/boardview for this card), it would come good again for a short time when injecting some voltage, so as a last resort I pushed the voltage above what I should and for now that took care of it (don't do this at home).
Motherboard also had several problems.
Very large ripple on CPU VCore ( https://www.badcaps.net/forum/attach...1&d=1684603334 ), caps were good, it was actually the buck controller (ADP3198) that wasn't regulating properly, so replaced that and that took care of this issue. I also took the opportunity to try to reduce Vdroop which can be quite large on this board, I observed more than 0.15V difference between idle and stress test, not that great for overclocking.
Bridged one of the resistor between CSSUM and CSCOMP, not the correct way to do it according to the datasheet, but much easier that adding a voltage divider on LLSET. Difference is less than 0.02V now.
Next issue was that the audio codec wasn't detected. Replaced the ALC883 with an ALC888 and that's fixed too.
Final issue was with Ethernet. Gigabit mostly didn't work even though it got uplink. 100M was mostly working but not 100% stable. It was both a hardware and a software issue. Software issue was the crappy driver from Nvidia, the Microsoft one works much better. Hardware issue was weird. Booting from PXE was extremely slow so for sure it wasn't just software. I replaced the Ethernet PHY IC (88E1116), didn't fix it. I replaced the ports stack (2xUSB+RJ45 which also contains the Ethernet transformer), didn't fix it either but at that point I realized that the issue was mostly on download using iperf3, so unlikely to be an issue on the physical link after the PHY, more likely it was a problem on the RGMII interface between PHY and MAC (Nvidia southbridge here). Started probing, touched up some solder joints and that fixed it. Couldn't get to the bottom of it here either…
All components came from donor boards, for once the crap I keep that should go to the trash bin has been helpful…
Now comes overclocking.
Voltages:
- VCore: 1.30000 V (setpoint, actual is around 1.33V-1.34V)
- Memory: 2.085 V
- Northbridge: 1.563 V
Frequencies:
- FSB: 1650 MHz
- Memory: 825 MHz (linked)
- CPU: 3300 MHz (locked multiplier x8, 412.5 MHz BCLK)
3.32 GHz on the CPU is unstable, this is an FSB ceiling because it's the same when CPU multiplier and RAM ratio are decreased. Increasing CPU or northbridge voltage doesn't help.
RAM timings:
- CL: 5
- RCD: 4
- RP: 4
- RAS: 12
- RC: 16
- CMD: 2 cycles
Other timings are set to auto. I'm not experienced with fine-tuning RAM timings, but this does just a little bit better than stock at least.
On the software side, sadly only Windows 7 works with SLI. 64-bit of course. Windows 10 crashes when driver loads for both graphics cards (if one is disabled it works). Windows 11 always crashes with random blue screens, SLI or not.
Nvidia drivers version 342.01.
CineBench R15 scores 360.
Unigine Tropics set to DirectX 10, 1280x720 window, 4x anisotropic, trilinear filtering ,shaders&textures to high, ambient occlusion&reflection enabled, no anti-aliasing.
Scores 1640, 65.1 FPS with SLI.
Scores 1034, 41.1 FPS without SLI.
Sorry for the wall of text, this was a great waste of time for a useless machine.
I have no use for it, it's stuck on Windows 7 for SLI, but I don't feel like just giving it away to someone random since I spent so much time on it…
Comment