Got my new A64 system up n running, with a socket A WB zip tied over the CPU I was wondering if there is a performance hit when running the memory on a divider? Like in the old days of Athlon XP`s everyone said running the memory 1:1 was best. And is there a app out yet which lets you change FSB etc. from within windows?
divdiers dont give a performance hit Make sure you run whole multipliers too. Download "clockgen" for voltage, fsb and multipliers and "A64 tweaker" for memory timings
eh? whats the deal with that? im running a multi of 8.5 here (doesn't like 9 for some reason..) and its fine. Its just like running a multi of 9 except you need a little more HTT. As a general rule, the half multis use the memory dividers of the multi above e.g. 289x9 = 2601mhz "200" mem divider = CPU / 9 = 289mhz 306x8.5 = 2601mhz "200" mem divider = CPU / 9 = 289mhz 289x9 = 2601mhz "166" mem divider = CPU / 11 = 236mhz 306x8.5 = 2601mhz "166" mem divider = CPU / 11 = 236mhz
The memory controller is integrated and as such unless the RAM can run at full CPU speed it will always be effectively running with a divider (even 200mhz DDR400 on 200mhz HT800/1000). There is aboslutely no need to keep the RAM synced with HT (eg 250mhz HT with 250mhz DDR500), the best contributors to perf is fast RAM timings inc the 1T Command Rate. Whether you can achieve the best timings depends upon your CPU (for the controller) and RAM (capabilities, size, number of sticks etc). 2x512MB seems the best but the Venice core is meant to improve upon this. Just to cover it I've seen lots of info and what you really want is lower timings rather than faster speeds. DDR400-PC3200 2-2-2 performs about the same as DDR580-PC4700 3-4-3. DDR580-PC4700 3-4-3 tends to be a touch slower than DDR474-PC3800 2.5-2-2. Also don't worry about running HT1000 as HT800 or lower doesn't seem to have any downside, so if you're o/c'ing the HT from 200mhz to 280mhz don't worry if you need to drop LDT from 5x (default HT1000) to 3x as 3x280mhz = 840HT. Perhaps it will matter more in the future, perhaps with Dual Core CPUs but for the moment the increased CPU speed is much more important than HT speed. Finally will CPUZ not give you stepping code info?
Thanks alot! I`ve already started torturing my venice 3000+, it`s at 2533mhz now, at stock voltage! It idles at 28* and when running prime it reaches 38*. I`m giving my ch-5 RAM 3Volts to chew on, and it`s stable at 230mhz 2-3-3-10. Is it "safe" to give ch-5 3Volts? (And by stepping i mean that batch-code that`s printed on the cpu)
so if ive understood you correctly a system running @ say 2.75 @1:1 with say 3 4 4 8 timings is not any better a system running a 2.75 on a divider with the lower speed ram @ 2 2 2 2 ? i know this is a crude example just trying to check i get your drift
With an AMD64 you're effectively always running on a divider because of the integrated memory controller, there is no need at all to keep the RAM running at the same speed as HT. For AMD the 4th RAM timing is rarely optimal when at its lowest, 2-3-3-11 is likely to be faster than 2-3-3-5, although we are talking minute diffs here. Anyway more critically DDR500-PC4000 3-4-4 will certainly be slower overall than DDR400-PC3200 2-2-2 because even DDR580-PC4700 3-4-3 isn't really any faster than DDR400-PC3200 2-2-2. Timings are easily the most important, forget HT and higher RAM speeds instead look for higher CPU speed and lower RAM timings. This is not to say 2.5-3-3 (very standard timing for DDR) is poor, it's only about 3% slower in most things so all those people upgrading from SktA systems with basic RAM needn't worry. I think a lot of this boils down to AMD64 not really needing greater memory bandwidth than DDR400-PC3200 and certainly needing tighter timings much more than higher speeds. This is why I find it really wierd AMD are reportedly moving over to DDR-II as the higher speeds with slow timings don't sit well with current AMD64. Perhaps it's just to do with the way DDR-II is improving and also falling in price, maybe a big redesign will see more benefit to faster RAM. Whatever the theories behind it the end results are what is most important. Benchmark and see what your particular setup performs best with, in particular test keeping the CPU speed the same but running RAM as fast as it can handle at each RAM timing (eg 2-2-2, 2.5-3-3, 3-4-4 & ideally others). Try to always maintain T1 Command Rate as its meant to be the most important timing. Don't fixate too much on synthetic tests such as 3Dmark, PiFast or Sandra. Instead test in what you actually use your PC for; games at your usual settings, compressing files or MP3/video encoding for example. Synthetic tests like Sandra often have little to do with real world benefits but can be useful, for example when determining which 4th timing is faster. None of this is set in stone, it's merely the picture I've pieced together when looking all over the Internet for relative information. A lot of people seemed to be wasting lots of cash plumping for PC4000 (or faster) but not realising the downside the necessary slower timings bring. Esp when factoring in cost they may have done better to stick with their old PC3200 RAM.
OK- did I miss it? Well sorry, i always have to be the one to cause trouble, but there is no FSB on an AMD64 system. I mean that goes with everything Austin has said, but hey lets make this clear.
No I have not heard something.... As slower is this timing the better. Try a memory benchmark and you see...
LOL - well, i am kinda serious, it is a an actual big difference, its more important than the thing being 64bit, but everyone is all over that little change. I mean, it does bug me when people look dead at a truck and call it a car, so - yeah. Im a nerd like that.
