Faster Memory - What good is it?
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
Faster Memory - What good is it?
Optimizing Performance - Fast memory?
1. What hardware governs how quickly I can flit from window to window. Does the speed of the memory affect that?
2. What attributes of performance does fast memory help? What does it not help?
1. What hardware governs how quickly I can flit from window to window. Does the speed of the memory affect that?
2. What attributes of performance does fast memory help? What does it not help?
Re: Faster Memory - What good is it?
Well, someone did a benchmark on PCSX2.ces wrote:2. What attributes of performance does fast memory help? What does it not help?
39.02 FPS - SLUS 20672 - Intel Pentium Dual-Core E2160 - 2.75 GHz OC (920Mhz DDR2) - Shadow Lady
36.12 FPS - SLUS 20672 - Intel Pentium Dual-Core E2160 - 2.75 GHz OC (800MHz DDR2) - Shadow Lady
I guess that's one application where it helps. However, I think few others would find the memory to be a bottleneck.
Faster memory can benefit overclockers.ces wrote:Faster Memory - What good is it?
Intel motherboards clock the RAM at some ratio of the FSB speed, aka the memory divider. Many boards are limited to 3 standard memory dividers for FSB200 CPU's: 1:2, 3:5, and 3:4. CPU's with other FSB's (266, 333, etc) will have slightly different dividers.
You calculate your RAM clock like this:
For divider X:Y
(FSB * Y) / X
Example for FSB200 and 1:2 divider: (200 x 2) / 1 = 400 MHz RAM clock
Completing the table for 200 MHz FSB CPU's:
Code: Select all
DIV FSB:MEM EFF MEM SPD
--- --- --- ------------
1:2 = 200:400 (= DDR2 800)
3:5 = 200:333 (= DDR2 667)
3:4 = 200:267 (= DDR2 533)
So with a FSB of 200 MHz, and the standard memory divider 1:2, this RAM would run at 400 MHz.
That's fine if you're not overclocking. If you intend to overclock by increasing FSB, you either have to slow down your RAM clock to keep it under 400MHz, or buy faster RAM.
You slow down the RAM clock by picking different memory dividers.
@ 266 FSB you need the 3:4 divider to keep the RAM within spec:
Code: Select all
DIV FSB:MEM EFF MEM SPD
--- --- --- ------------
3:4 = 266:355 (= DDR2 709)
Code: Select all
DIV FSB:MEM EFF MEM SPD
--- --- --- ------------
3:4 = 300:400 (= DDR2 800)
Code: Select all
DIV FSB:MEM EFF MEM SPD
--- --- --- ------------
1:1 = 333:333 (= DDR2 667)
5:6 = 333:400 (= DDR2 800)
So I don't do that. It's easier and less risky to just buy RAM rated for higher clock speeds and use standard memory dividers.
Moving to DDR2-1066, you can clock up it to 533 MHz.
Completing the table for a 200 MHz FSB at various FSB overclocks:
Code: Select all
DIV FSB:MEM EFF MEM SPD
--- --- --- ------------
1:2 = 240:480 (= DDR2 960)
1:2 = 266:533 (= DDR2 1066) *PERFECT*
1:2 = 333:666 (= DDR2 1333) *Unstable
3:5 = 240:400 (= DDR2 800)
3:5 = 266:443 (= DDR2 887)
3:5 = 300:500 (= DDR2 1000) *Stable, but slower than RAM's rated speed
3:5 = 333:555 (= DDR2 1110) *22 MHz RAM overclock = probably stable
3:4 = 266:355 (= DDR2 710)
3:4 = 300:400 (= DDR2 800)
3:4 = 333:444 (= DDR2 888)
3:4 = 400:533 (= DDR2 1066)
Re: Faster Memory - What good is it?
What is PCSX2?ilovejedd wrote: Well, someone did a benchmark on PCSX2.
Re: Faster Memory - What good is it?
www.google.comces wrote:What is PCSX2?ilovejedd wrote: Well, someone did a benchmark on PCSX2.
It's a PS2 emulator. But the answer was in the following post. Fast memory is required for overclocking headroom so you're not limited by your RAM when overclocking. And of course benchmarking.
...Or for compressing DivX
From this [url]thttp://www.tomshardware.com/reviews/ram-speed- ... 07-15.html[/url]page on Tom's Hardware:
From this [url]thttp://www.tomshardware.com/reviews/ram-speed- ... 07-15.html[/url]page on Tom's Hardware:
So, if you care about memory throughput-benchmarks you should get fast memory. Oh, the heatspreaders on memory seem to be only for marketing as well, because the surface area is already very high for such a low wattage and the heatspreader only adds material for the heat to go through, so the temps will not change a lot.The results must look disappointing for the memory vendors, as the largest performance differences we found amount to 7-8% with DivX and WinRAR, while almost all other benchmarks and applications perform alike: a 1-3% performance delta cannot be noticed at all. Some games showed several per cent performance difference between low-latency high-speed memory and conventional high-latency average speed DIMMs. The synthetic benchmarks on the memory revealed even more differences, but these clearly aren’t very relevant in everyday life.
Or you can underclock the memory while overclocking the CPU? Is that correct?psyopper wrote:Other than a 1-2 % difference in most normal, non-synthetic tasks - not much. As pointed out, it's really handy for overclockers.
To take an i3-530 to 4 Ghz (common 30% overclock on stock voltage and air) you need 1200 Mhz RAM at the minimum dividers of most motherboards.
-
- Posts: 404
- Joined: Sun Jul 04, 2004 9:18 pm
- Location: Illinois, USA
Jay, ilovejedd, Scoop and psyopper,
Do you have any opinions on this:
More Cache - What good is it?
viewtopic.php?t=58469
Do you have any opinions on this:
More Cache - What good is it?
viewtopic.php?t=58469
On my Gigabyte board the memory clock is a multiple of the BCLK. My options are 6, 7, 8.ces wrote:Or you can underclock the memory while overclocking the CPU? Is that correct?psyopper wrote:Other than a 1-2 % difference in most normal, non-synthetic tasks - not much. As pointed out, it's really handy for overclockers.
To take an i3-530 to 4 Ghz (common 30% overclock on stock voltage and air) you need 1200 Mhz RAM at the minimum dividers of most motherboards.
6 x 200 BCLK came out to 1200 Mhz. for a 20 x 200 overclock. The max I would be able to get and keep my 1066 RAM clock is a 177 BCLK. The 530 is a 22x processor which would get me 3.89 Ghz.
I tried loosening the memory timings to his 180 x 22 to get me 3.96 but that put my RAM at 1080 Mhz and it just couldn't hold it, no matter how loose I set the timings, up to a 1.8v voltage.
That being said, I only paid $45 for 2 GB of the stuff... And if you are not overclocking, a real apples to apples comparison of different RAM only netted about a 3% difference in synthetic benchmarks.