• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

How important is memory speed REALLY?

TriggerHappy101

Golden Member
At a gamers standpoint how important is ram latency really? I was thinking of purchasing a gig of this: http://labs.anandtech.com/alllinks.php?pfilter=1799
CORSAIR ValueSelect 1GB (2 x 512MB)
It has a CAS of 2.5.

It will be going into the following (and soon to be made) system:
eVGA GeForce 7800GT 256MB
EPoX nForce4 Ultra (939) EP-9NPA+Ultra
AMD 64 3700+ 1MB San Diego

Will I really see much of increase in evenness of my FPS if I buy faster RAM? Or would it be smarter to just buy another gig (or half) of this kind?

Any articals\reviews of ram speed would be appericated. Thanks.
 
Originally posted by: JustAnAverageGuy
Link

Timings don't matter (much).

Command rate does though. So if you plan on getting 2GB, get it in the 2x1GB form, not 4x512MB.

This is something particular to AMD mobos, right? Do you know what it is called, or what it is referred to as? I've been trying to research it on google, but I haven't the slightest as to what to search for. (I have an Intel system, but I'm building my wife a A64 system, so I'd like to know, but never had to before).
 
Not particular to AMDs, but is significant to the K8/K9s because of the specifics of AMDs on-die mem controller. If you're going for AMD performance, avoid 2T at all costs.

Order of importance:
command rate (1 or 2 DIMMs only!)
DDR
timings
 
Originally posted by: Fullmetal Chocobo
This is something particular to AMD mobos, right? Do you know what it is called, or what it is referred to as?

Not the mobos. It's a limitation of the on-die memory controller. It can't address 4 double sided DIMMs with a 1T command rate.

It's called a "limitation". I don't know the technical name for it 😛
 
I would just like to point out that aside from memory bandwidth benchmarks, 2T's performance loss is GREATLY exaggerated. Its on the order of 2-5%.
 
Originally posted by: PingSpike
I would just like to point out that aside from memory bandwidth benchmarks, 2T's performance loss is GREATLY exaggerated. Its on the order of 2-5%.



Don't forget Super Pi
 
Originally posted by: PingSpike
I would just like to point out that aside from memory bandwidth benchmarks, 2T's performance loss is GREATLY exaggerated. Its on the order of 2-5%.

yep
 
I disagree STRONGLY that command rate doesn't have a huge impact on a64 performance. I did and can replicate testing with tighter timings at 2t, and losser times at 1t. The differences? The best example was 3dmark2001, where the vid card was not bottlenecking at all. The 1t setup scored 4500 points better each time ran, and framerates went up about 20%.

This has to do with the memory controller being on-die vs having the controller on the mobo.
 
Originally posted by: TriggerHappy101
>< well which one is it.

Alright, well, Ithink Ijust solved this... I checked out the thread about how it has nil impact, and they are all benchmarks of modern games that are gpu limited... I will repeat again, the reason that the 1t rate is faster isn't because the memory gets significantly faster (1-5%) but because it chokes the cpu... if the cpu isn't being used all the way anyways you get a lot less effect out of it.
 
heres one where the gpu isn't used at all:

SuperPI 8m 1T- 6:03m
SuperPI 8m 2T- 6:09m
1T is 2.4% faster

and heres your 3d01:

3DMark2001SE 1T- 22543
3DMark2001SE 2T- 21970
1T is 2.6% faster

and mine from that thread:

1t -- 2t
super pi 1.04 8M __ 6:10 ___ 6:27
pcmark04 __ 5143 ___ 4996
aquamark __66,903 ___ 65,615
3dmark01 __24,361 ___ 23,212
3dmark03 __11,710 ___ 11,592

 
Originally posted by: rise4310
heres one where the gpu isn't used at all:

SuperPI 8m 1T- 6:03m
SuperPI 8m 2T- 6:09m
1T is 2.4% faster

and heres your 3d01:

3DMark2001SE 1T- 22543
3DMark2001SE 2T- 21970
1T is 2.6% faster

and mine from that thread:

1t -- 2t
super pi 1.04 8M __ 6:10 ___ 6:27
pcmark04 __ 5143 ___ 4996
aquamark __66,903 ___ 65,615
3dmark01 __24,361 ___ 23,212
3dmark03 __11,710 ___ 11,592


hrm, weird... my experiance has definitely been different from yours on this. Was this done on the X2 in your sig? Maybe single vs dual core?

 
Originally posted by: rise4310
no that was my winnie 3200. are you sure you didn't change anything else?

Positive, I actually went back afterwrads and changed it again with the same results.

Maybe it's a matter of the extra cache I have vs your winnie? Any opteron users comment on this?
 
I'm in a similar situation. I have 2x512 ram now I want to move up to 2GB so got a 1x2GB kit. I will tell you it was not worth it and I'm sending the 2GB kit back.

I don't have the beachmarks with me (Im' at work).
I have a OP144 OC to 2250, For

I test with the 2x512, the 2xGB and 4x152 (2 different, matching sets one VS and one Kingmax)
I saw almost no diffenence (1-3% at most) in all my tests.
I even got rid of the over clock and thest was silimar.
My guess would be that T2 times are little slower (2%)
Also 2GB is almost never needed for now.

Now I don't have any FPS to test it on.
I used SuperPi and a game call Kohan II.
What I like about the Kohan 2 game for testing is it is a RTS with lots of units.
I ran a film at 1600% speed, and zoomed in a corner of the map with nothing happening. When I doing this video is never the limitation, it all CPU and memory.
This game when in a huge battle map would use almost 600MB just for the game.
And still 2 GB of memory is not needed.

Since I don't have FPS, I don't know if they use alot memory when in game (system memory not video card memory)

My suggestion is to get the 2x512 now and if need upgrade to any 2x512 for cheap later.
All these people in this forum will be selling them to get 2x1GB and you get it for cheap.




 
Back
Top