AMD with out-of-sync ram?
ThatFatCat5
Minn-eee-soda
what would happen if someone attempted running a http://www.newegg.com/Product/Product.asp?Item=N82E16819103547 with DDR266 ram, would it run incredibly slow, not work at all?
and should it have DDR500 ram?
and should it have DDR500 ram?
0
Comments
It doesn't really matter what memory you use as long as it is DDR, it could be PC10,000 for all I care your motherboard will try to run it at whatever it supports, if it's PC1600 it will just run the ram as fast as it can which will probably be at PC1600 levels.
If your mobo's BIOS is up to date and you drop a current revision CPU in, you'll see the option appear under the memory settings....
That being said your PC will run fine with PC2100. The A64 Integrated Memory Controller will mask most of the performance loss of the slower ram
That is very interesting Omega, I have yet to hear anything about that. I know that they were contemplating a move to DDR500 instead of DDR2, however I didn't know there was any 'official' support for it today. Is this just a mainboard thing or is actually the IMC?
Have any links to some more info? I'm just surprised that AMD would support something that requires the reference clock and HTT bus to be run outside of spec.
ex: Venice "E6" 3200+ 10X 200mhz = 2000mhz (HTT 5x 200mhz = 1000)
or
Venice "E6" 3200+ 8x 250mhz = 2000mhz (HTT 4X 250mhz = 1000)
The news is several months old, and it didn't make much of a splash...
Finally found some old info...
Anandtech: Athlon 64 Revision E: Unofficial DDR500 Support New Memory Speeds
OK the support may be Unofficial but it's there
Thanks for the link
I'm surprised there wasn't more press on this I'm going to have to check this out. From the looks of that article, it's not even a multi drop and a reference clock increase, it appears to be async memory dividers that allow the memory to run faster than the reference clock, i.e. > 1:1. Very interesting..
The multiplier part is correct, however the memory frequency is not determined from the reference clock (it sometimes works mathematically, but it sometimes does not). The memory frequency is based on an IMC selected 'divisor' of the CPU's operating frequency, not the reference clock.
I.e. take an overclock of 8x300MHz (2400MHz) and a 5/6 divider.. should work out to 250MHz if it was indeed determined from the reference clock. If you try that combination, it actually comes out as 240MHz. Memory controller used [CPU Clock/10] not [300MHz*5/6]
ClockSpeed/Divider= Memory Speed
Examples:
FX-55 13x Multi HTT200 2600mhz
DDR400 (200mhz) (Mem Ratio 1:1 (Ideal)) Memory Divider 13
2600 / 13 = 200mhz
DDR333 (166mhz) (Mem Ratio 6:5 (Ideal)) Memory Divider ((13)*6/5) = 15.6 = 16
2600 / 16 = 162.50mhz
DDR 266 (133mhz) (Mem Ratio 6:4 (Ideal)) Memory Divider ((13)*6/4) = 19.4 = 20
2600 / 20 = 130.00mhz
3200+ 10x Multi HTT200 2000mhz
DDR400 (200mhz) (Mem Ratio 1:1 (Ideal)) Memory Divider 10
2000 / 10 = 200mhz
DDR333 (166mhz) (Mem Ratio 6:5 (Ideal)) Memory Divider ((10)*6/5) = 12
2000 / 12 = 166.67mhz
DDR266 (133mhz) (Mem Ratio 6:4 (Ideal)) Memory Divider ((10)*6/4) = 15
2000 / 15 = 133.33
3000+ 9X Multi HTT200 1800mhz
DDR400 (200mhz) (Mem Ratio 1:1 (Ideal)) Memory Divider ((9)*1/1) = 9
1800 / 9 = 200mhz
DDR333 (166mhz) (Mem Ratio 6:5 (Ideal)) Memory Divider ((9)*6/5) = 10.8 = 11
1800 / 11 = 163.63
DDR266 (133mhz) (Mem Ratio 6:4 (Ideal)) Memory Divider ((9)*6/4) = 13.4 = 14
1800 / 14 = 128.57
Pics Below
Opty 165 (9x) HTT278 with Memory set at 333mhz (6:5) ratio
2502.7 / 11 = 227.51mhz Memory
Yup, used to confuse the heck out of me till I finally wrapped my head around a64 dividers.. Same reason why half multipliers do funny things with the memory frequencies too. Oscar Wu made a good reference table with all of the actual DRAM frequencies a while back up to 5GHz..