peter said:
Hi
I have question of the possible upgrade within my old mobo and cpu. I use
the pc for photography (Photoshop, RAW conversion) not gaming.
Would I benefit more from getting more memory or changing to AMD Athlon XP
3000 ?
And how much improvement should I expect?
My pc:
Processor
Model : AMD Athlon(tm) XP 2000+
Speed : 1.67GHz
Mainboard : ASUSTeK Computer INC. A7N8X
Total Memory : 1GB DDR
Chipset
Model : ASUS nForce2 AGP Controller
Front Side Bus Speed : 2x 134MHz (268MHz)
Total Memory : 1GB DDR
Memory Bus Speed : 2x 134MHz (268MHz)
Memory Module 1
Type : 512MB DDR
Speed : PC3200U DDR-200
Standard Timings : 3.0-3-3-8 2-11-0-0
Set Timing @ 200MHz : 3.0-3-3-8 2-11-0-0
Set Timing @ 167MHz : 2.5-3-3-7 2-9-0-0
Memory Module 2
Type : 512MB DDR
Speed : PC2100U DDR-133
Standard Timings : 2.5-3-3-6 2-0-0-0
Set Timing @ 133MHz : 2.5-3-3-6 2-0-0-0
Set Timing @ 100MHz : 2.0-2-2-5 2-0-0-0
Video System
Adapter : NVIDIA GeForce 6200 (256MB DDR3, AGP 3.00, PS 3.0, VS 3.0)
tia
peter
From a processing core perspective, you'd expect a 3000+ to be 50%
faster than a 2000+ (if you believe the AMD P.R. rating). The 3000+
has twice the cache, but the core clock runs at 2100 or 2167,
depending on which 3000+ we're talking about. So the core clock is
going to be 2100/1667 = 1.26, and the rest of the improvement is
presumably due to doubled cache and higher FSB.
Barton 2100 (3000+) OPGA 200 512 10.5x 1.65V 85oC 53.7W
Barton 2167 (3000+) OPGA 166 512 13x 1.65V 85oC 58.4W
An old rule of thumb for Photoshop, was that if the undo buffer
was set to 1 buffer, then Photoshop needed 5x as much memory
as the size of the image being processed. Plus a chunk for the
OS as well. 1GB might give you room for say a 150MB image.
If you routinely exceed that image size, in processing, then you're
hitting the scratch disk. The memory bandwidth might be, say 900MB/sec,
while a hard drive might be 60MB/sec. So hitting the scratch disk
is a big penalty. If you worked on a 300MB image, the disk light
might be coming on a lot. You can judge that for yourself, and
test various sized images, to see if your version of Photoshop
follows those rules or not. If you have your "undo" setting
higher than 1, then the multiplier would be larger than 5X.
You can comfortably double the memory to 2GB. The chipset doesn't
work properly, for any memory quantity over 2GB. So you can forget
about 3x1GB as a configuration. Your best config would be
2x1GB, leaving the third slot empty. That gives dual channel
operation. (Even if you try 2x1GB plus 512MB, I'm told it doesn't
work. Even 2x1GB plus 256MB doesn't work. I have personally
tested 3x512MB, and three sticks do run stable. So the issue
seems to be total memory quantity, rather than number of sticks
as such.)
The modules should not come from Ebay, as the DDR modules
there use x4 chips. The 1GB modules you'd purchase, should use
(16) 64Mx8 chips, which usually takes the form of branded
RAM (Crucial, Kingston, OCZ, Corsair, Samsung, Micron). Ebay
has unbranded modules with Samsung chips (and the chip name is
not to be confused with Samsung making the whole module - there
is no brand name on the Ebay module itself). The Ebay modules may be
cheaper than others, but they're not recommended from a resale
perspective (if you want to sell them, or give them to a friend
later).
For example, this kit would make a nice 2x1GB solution. Always
read the reviews, before buying any memory...
http://www.newegg.com/Product/Product.aspx?Item=N82E16820227210
1) The memory upgrade helps, if you're hitting the scratch disk
a lot. But the best you can do, is double the memory. If
the images are so big, that you're still hitting the scratch
disk, even after moving to 2GB memory, then a memory upgrade
might not be a big improvement.
2) Say the processing speed was somewhere between 26% and 50% faster,
with a new processor. Some of the time is spend processing,
and some of the time is spent sloshing files back and forth
to disk (or memory). The sloshing part could currently be
dominating what is going on.
Perhaps you can judge for yourself, based on your own experiments,
what to expect. You could, for example, slightly overclock
the processor, and measure the improvement in batch processing
time. And maybe that would give an idea how much improvement
processing alone makes. You could use a small overclock,
so the machine doesn't get too annoyed with you (like 5%).
(Change CPU clock from 133 to 140MHz. I hope you're familiar
with how to program memory timings, because to do the testing
properly, you want to try to preserve the memory settings
if you can. And that means doing some manual programming,
copying the currently used timing values seen in Windows,
down into the BIOS screens.)
If overclocking or fiddling with memory timings or clock, always
test with memtest86+ first. Don't immediately boot into Windows.
I use a Linux LiveCD for further testing, after a couple passes
of memtest, but that isn't everyone's cup of tea. The Linux LiveCD
is nice, because you don't need any hard drive connected in order
to use it. Thus your hard drive doesn't get corrupted, while doing
initial overclocking testing. Either Knoppix or Ubuntu would do,
and they're a 700MB download, plus burning the ISO9660 to a
CD.
Another alternative might be to slow the CPU clock down by 5%,
and do your percentage improvement testing that way. Maybe that
would be less risky and less fiddling to do. The purpose of
trying 5%, is to estimate how much difference that 26% or
50% number might make.
Paul
