chewie198

Distinguished
Apr 28, 2005
1
0
18,510
Hi everyone,

I'm looking at building a cheap gaming computer for my brother, and I've settled on just about every component other than the RAM. Basically, he wants the fastest he can get for the least money, and since I have an old Athlon 1.4 Thunderbird system lying around with 2 sticks of DDR266 memory, I was thinking about having him use those to cut costs. With a new Nforce4 board he should be able to run them in Dual Channel; however, I was wondering what kind of performance impact could be expected from the lower bandwidth. After looking at some benchmarks here on THG and my own system, I've ended up a little puzzled. On my own system I'm running a Athlon 64 3500+ with a gig of Corsair XMS Pro, and I tried underclocking my RAM to 266 to see how much it would affect FPS. The max performance differential I've experienced was 6%, with the most being 5 frames per second in UT2004. In Doom 3 the difference is barely measureable, around 2-3%. Yet, after looking at various posts in this forum, people are claiming that DDR2100 is 55% slower, and as such real performance gaps will be around 15-20% or so. Of course, if I was running a more memory intensive game such as Quake III, I'm sure I could expect larger gaps, maybe up to 25-30 FPS. The only thing is, these games are already running in the hundreds of frames per seconds anyways, so it doesn't even matter. From anyone who's actually upgraded their RAM in a modern game like Half-Life 2, what is the max performance difference you've ever experienced, and is it worth not going with the "free" sticks I already have and buying completely new ones instead? Remember, the goal here is buying the most performance for the money, and no money at all means a basically infinite performance/price ratio, assuming the DDR266 runs fast enough in the first place.
 

BrentUnitedMem

Distinguished
Oct 8, 2004
693
0
18,980
Hey Chewie .. "ARRRRRRRRRHG"

okay sorry that was lame.

The maximum possible difference in DDR266 (PC-2100) and DDR-400 (PC-3200) at the same module latency is 52% calculated like so:
_________________________________
How long does it take to send 3200MB ?

PC-3200 = 1 second
pc-2700 = 1.185 seconds, 18.5% performance difference
pc-2100 = 1.52 seconds, 52% performance difference
_________________________________

At zero latency, PC-3200 is 52% faster than PC-2100.
For real world applications zero latency does not exist; there is a latency difference. Normally, the BIOS will run PC-2100 at a better latency than PC-3200 as follows:
PC-2100 standard latency CL2.5 or CL2
PC-3200 standard latency CL3

The theoretical perfomance gap of 52% is significantly narrowed by:
-The fact that PC-2100 runs at a better latency
-noise, physical resistance on wires
-other weak hardware links
-etc

And so the real world performance gain going from PC-2100 to PC-3200 is about <b>14%-19%</b> at best, possibly 6% is an average rating in your case.

<font color=green>*****
"Memory with lifetime warranty? So, whose lifetime is that?"
<A HREF="http://www.brentcrowley.com/" target="_new">homepage</A>
<font color=red>AIM BrentUnitedMem<P ID="edit"><FONT SIZE=-1><EM>Edited by BrentUnitedMem on 04/28/05 11:49 AM.</EM></FONT></P>
 

jammydodger

Distinguished
Sep 12, 2001
2,416
0
19,780
Bear in mind that although it has a clockspeed increase of 52%, clock speed increases are never proportional to performance increase's. Also bear in mind that it is only the memory subsystem being upgraded and not the entire computer.

In a memory intensive application I would expect PC3200 to be at most 15% faster than PC2100. The memory results you are seeing in games seem about right.