Hi All,
I was pointed to Zebo's Basic OC guide some months ago as I was just setting out on the overclock road. I have since done a lot of investigation, as the new generation of Dual x 16 PCIE boards seem to have some important differences to the "older" 2 x 8 channel boards.
So to add a few extra bits to Zebo's "industry benchmark" guide, specific to the now very poular A8N32 SLI - Deluxe board.
1. The addition of the second x16 channel for the second PCIe slot has meant that there are now two HTT links.
NVidia's previous (i.e: 'pre NForce4 SLIx16') chipsets for AMD 8th-Gen platforms were all single-chip designs using a Hypertransport-link to communicate with the processor.
The original NForce4 SLI chip doesn't care whether it's 'talking' to the processor or yet another NVidia-chip which is necessary to provide additional PCI-e lanes.
In the case of the A8N32, the second chip which serves as the Northbridge in the NF4 SLIx16 chipset is called NForce 4 SPP 100 (aka C51D).
This means that there are two values that appear in BIOS K8>NB which is the "CPU Multi" and this has a corresponding bandwidth that should be set to 16/16. The second set of values are SB>NB and these are the second HTT link which amongst other things controls the second x16 PCIe slot and these should be set manually to 200 x 5 with 16/16 bandwidth for all SLI set-up's, or problems can ensue.
2. The Memory Timings calculation is completely different as I found out with a mild OC.
Problem was I had my 4800+ set at 235 x 11 = 2585Mhz with Memory divider at 183, so should have had memory at 215Mhz or DDR 430, but CPUz & SiSoft showed memory at 198.8Mhz!!!
Having finally got some sense out of a technical guy at Asus and I am afraid this throws whole calculation of effective memory speed upside down, as the calculation method they gave me does give the 198.8Mhz (DDR398) figure on the OC I was running!
It is NOT a simple "mem limit" divided by "base clock" (200) x CPU FSB/HTT = "effective Mem Speed"!!!!!
The calculation involves calculating two new values:
1. "Effective BASE CPU Default Speed" - This is basically what the CPU would run at if the CPU FSB were set at the default 200. (i.e. if you have K8>NB multi at 11x this = 200 x 11 = 2200Mhz, if K8>NB Multi is at 10x = 200 x 10 = 2000Mhz etc......)
2. "Effective CPU/RAM Ratio" - This is the value of "Effective BASE CPU Default Speed" divided by the Mem Limit value you have set, but the value is then rounded up to the nearest whole number (no decimal places). i.e. if Effective BASE CPU Speed is 11 x 200 = 2200 and Mem Limit is at 183, the Effective CPU/RAM Ratio is 2200/183 = 12.02186, which rounded up to nearest whole number is 13.
OK now we have the two "new" values needed to work out effective mem speed, as Asus say the board works it out as follows:
Effective Mem Speed = (CPU Multiplier divided by CPU RAM Ratio) multiplied by CPU FSB.
In this example: CPU FSB - 235, CPU Multiplier x11, Mem Limit 183.
Effective BASE CPU Default Speed = 200 x 11 = 2200
Effective CPU/RAM Ratio = 2200/183 = 12.02186 rounded up to 13.
Therefore Mem Speed = (11/13) = 0.846154 x 235 = 198.8Mhz or DDR397!!!!
This, though I hate to say it matches exactly the results from CPUz and SiSoft!!
I have posted this separately on Asus forum and have made an Excel spreadsheet that calculates for you, but I don't know how or where to post this, so can email if anyone wants.
Hope you can find some way to add this to your guide, as without this calculation the mem speeds are all way off.
Ridesy
I was pointed to Zebo's Basic OC guide some months ago as I was just setting out on the overclock road. I have since done a lot of investigation, as the new generation of Dual x 16 PCIE boards seem to have some important differences to the "older" 2 x 8 channel boards.
So to add a few extra bits to Zebo's "industry benchmark" guide, specific to the now very poular A8N32 SLI - Deluxe board.
1. The addition of the second x16 channel for the second PCIe slot has meant that there are now two HTT links.
NVidia's previous (i.e: 'pre NForce4 SLIx16') chipsets for AMD 8th-Gen platforms were all single-chip designs using a Hypertransport-link to communicate with the processor.
The original NForce4 SLI chip doesn't care whether it's 'talking' to the processor or yet another NVidia-chip which is necessary to provide additional PCI-e lanes.
In the case of the A8N32, the second chip which serves as the Northbridge in the NF4 SLIx16 chipset is called NForce 4 SPP 100 (aka C51D).
This means that there are two values that appear in BIOS K8>NB which is the "CPU Multi" and this has a corresponding bandwidth that should be set to 16/16. The second set of values are SB>NB and these are the second HTT link which amongst other things controls the second x16 PCIe slot and these should be set manually to 200 x 5 with 16/16 bandwidth for all SLI set-up's, or problems can ensue.
2. The Memory Timings calculation is completely different as I found out with a mild OC.
Problem was I had my 4800+ set at 235 x 11 = 2585Mhz with Memory divider at 183, so should have had memory at 215Mhz or DDR 430, but CPUz & SiSoft showed memory at 198.8Mhz!!!
Having finally got some sense out of a technical guy at Asus and I am afraid this throws whole calculation of effective memory speed upside down, as the calculation method they gave me does give the 198.8Mhz (DDR398) figure on the OC I was running!
It is NOT a simple "mem limit" divided by "base clock" (200) x CPU FSB/HTT = "effective Mem Speed"!!!!!
The calculation involves calculating two new values:
1. "Effective BASE CPU Default Speed" - This is basically what the CPU would run at if the CPU FSB were set at the default 200. (i.e. if you have K8>NB multi at 11x this = 200 x 11 = 2200Mhz, if K8>NB Multi is at 10x = 200 x 10 = 2000Mhz etc......)
2. "Effective CPU/RAM Ratio" - This is the value of "Effective BASE CPU Default Speed" divided by the Mem Limit value you have set, but the value is then rounded up to the nearest whole number (no decimal places). i.e. if Effective BASE CPU Speed is 11 x 200 = 2200 and Mem Limit is at 183, the Effective CPU/RAM Ratio is 2200/183 = 12.02186, which rounded up to nearest whole number is 13.
OK now we have the two "new" values needed to work out effective mem speed, as Asus say the board works it out as follows:
Effective Mem Speed = (CPU Multiplier divided by CPU RAM Ratio) multiplied by CPU FSB.
In this example: CPU FSB - 235, CPU Multiplier x11, Mem Limit 183.
Effective BASE CPU Default Speed = 200 x 11 = 2200
Effective CPU/RAM Ratio = 2200/183 = 12.02186 rounded up to 13.
Therefore Mem Speed = (11/13) = 0.846154 x 235 = 198.8Mhz or DDR397!!!!
This, though I hate to say it matches exactly the results from CPUz and SiSoft!!
I have posted this separately on Asus forum and have made an Excel spreadsheet that calculates for you, but I don't know how or where to post this, so can email if anyone wants.
Hope you can find some way to add this to your guide, as without this calculation the mem speeds are all way off.
Ridesy
Facebook
Twitter