Memory latency- waste of money and time.

[SpeedZealot369]

Hi, I've always been conchise of my memory timings, making sure I'm getting the most performance I can, but I just saw some benchmarks that indicate how there is virtually no difference between valueram and crazy expensive ram (performance wise) and even if there is, not big enough to justify some people spending so much money on "fast" memory.

 

[orangat]

Your question was given an in-depth analysis by an AT article I believe.

Latency gives a relatively better boost in performance compared to bandwidth for A64 systems.
Some apps like vid encoding are less affected by memory and are more cpu bound.

 

[Hacp]

You won't notice much of a difference between 2.2 and 2.4 either

 

[Zap]

Well, there is a difference in performance between cheap and expensive RAM... you just have to decide whether paying 50% more $$$ is worth only 2% more performance.

 

[warzer]

Wouldn't you still need premium ram to hit the 250-300+ fsb mark that people are doing to o/c there opty's and x2's?

 

[AkumaX]

Originally posted by: warzer
Wouldn't you still need premium ram to hit the 250-300+ fsb mark that people are doing to o/c there opty's and x2's?

using dividers to keep their memory in spec

 

[imported_miahallen]

warzer is correct that many people buy expensive RAM with fast timings, but they usually run them at lower timings in an attempt for a good OC. If I purchased value PC3200 RAM with 3-4-4-8 timings and wanted to OC, I'd have to use a divider. But if I purchased PC3200 RAM with 2-2-2-5 timings, there I good chance I could run it at a slower speed (ie 3-4-4-8) and OC the crap out of it, running it in sync with my FSB speed.

Take a Venice 3000 (stock at 2.0GHz)
Value RAM @ 3-4-4-8 with a 4:3 divider
now OC the proc to 2.4GHz, this will put your RAM at PC2880 speeds
Your RAM is limiting your memory bandwidth to ~5.8GB/s

Take the same Venice 3000
Premium RAM @ 3-4-4-8 but with no divider
now OC the proc to 2.4GHz (same speed)
But this time your RAM is being OCed as well...to PC3840 speeds
expanding your memory bandwidth to ~7.7GB/s (a 33% increase in this example)

This will provide much more than a 2% increase in speed,
anywhere from 5-30% depending on the application.

 

[imported_miahallen]

However the article is correct, if you never plan on OCing.

 

[Zebo]

I made a thread on this awile back with some tests and stuff but basically it's a waste of money - far better things to buy like video card - nicer case to get the chicks - hardrive - whatever - LCD's also get the chicks BTW but anyway memory should be last thing to pay a premium for IMO.

 

[Zap]

Originally posted by: miahallen
This will provide much more than a 2% increase in speed,
anywhere from 5-30% depending on the application.

Yeah, if your "application" is Sandra.

Buying "premium" memory is like buying a P4EE or A64 FX, you can get "near" it for a lot less. You don't even have to buy the Celeron/Sempron of memory either, just that super expensive stuff is just super expensive. Ever heard of point of diminishing returns?

FAST 2GB DDR Kits - Part 2
At same 2.4GHz, running DDR400 and DDR533 (huge difference) I'll just list top scorer, regardless of brand. This roundup was for "enthusiast" memory = usually costly stuff but which can clock high and/or run low latencies. Still, some brands could not overclock to the maximum DDR533 that was tested. I've listed all five tests.

Quake3 545.2 - 558 - 2.3%
Sandra 6070 - 6972 - 14.9%
Sandra2 2679 - 3146 - 17.4%
Super PI 82 - 80 - 2.5% (lower is faster on this one)
RTCW 118.9 - 121.9 - 2.5%

There ya go. Everything shows a real and measurable gain, but "most" show only in the 2% range, the exception being Sandra which averages about 16%. BTW Sandra is known for being very sensitive to memory speeds and of course very unlikely that few "real" program that someone will sit down and use (versus just benchmark) will show more than the 2% range in gain, and probably almost none will show the gains that Sandra shows.

 

[PingSpike]

High end memory is an all around bad buy unless you've a benchmark fiend with money to burn. As has been said, its performance for the dollar is pretty insignifigant in most real world applications. Best to blow your cash on other components, particularly the video card in regards to gaming.

And there's no real reason to fear running on a divider on the A64. I believe that technically, you're always running on a divider on A64...since ram speed is based on cpu speed and a divider, not cpu/htt and a divider.

 

[xtknight]

If you don't overclock, there is essentially absolutely no difference.

 

[hclarkjr]

similar question asked in general hardware

 

[kmmatney]

You can easily overclock with Value RAM. My RAM can't go any higher than maybe 208 MHz, but I still achieved a 50% overclock using a memory divider which keeps the memory at a perfect 200 MHz for PC3200.

 

[Throwmeabone]

Originally posted by: kmmatney
You can easily overclock with Value RAM. My RAM can't go any higher than maybe 208 MHz, but I still achieved a 50% overclock using a memory divider which keeps the memory at a perfect 200 MHz for PC3200.

Does that make the system slower than if you used RAM that did not need a divider?

 

[RallyMaster]

Nope. I'm on ValueSelect and I see a noticeable improvement when loading in Call of Duty 2. It even eliminates the lag that I had before. Awesome.