PDA

View Full Version : Ram Latency and Gaming


Blacklash
01-23-05, 03:22 PM
Dated and yet interesting article here. It also tends to support the theory that benchmarks are entertaining and useful in some regards yet do not always reflect actual gaming performance.

Tight latency settings show nice gains in actual gaming yet small in benches like 3dmark, particularly for AMD.

"The Halo scores in particular are worth pointing out. Our P4 system gains over 4% going from CAS 3.0 to 2.5, and over 10% going to 2.0. The Athlon 64 does even better, earning 7.5% moving from CAS 3.0 to 2.5 and 13.6% going to 2.0. In both cases, that's an absolutely enormous speed increase, but we were able to verify it with repeated testing. Something about the way Halo is coded makes it highly sensitive to RAM latency, and we wouldn't be surprised if there were other games out there with similar properties."

Source:

http://www.extremetech.com/article2/0,1558,1637769,00.asp

So tighten them up people, and yes outside of gaming it is almost inconsequential.

superklye
01-23-05, 03:30 PM
There were a couple of pretty big threads about this a week or two ago, but yeah...besides benchies, lower latencies really have almost no effect on performance at all.

keith33
01-23-05, 04:56 PM
The difference between 1T and 2T though..... HUGE!!! Relax timings and go down to 1T if you can. It's crazy.

superklye
01-23-05, 05:17 PM
What does the 1T/2T mean?

jolle
01-24-05, 07:13 AM
is extremetech.com down? cause I havent been able to get onto that site for days..
Wondering if this is some screwup on my ISPs part, if the site doesnt allow certain countries to connect or something like that..

1T = command per clock enabled..
I think it means that the CPU waits 1 cycle after sending to memory for response, instead of 2 cycles.. not sure tho..

nIghtorius
01-24-05, 09:59 AM
What does the 1T/2T mean?

It's an Athlon64 thang :D

It's the number of cycles before issueing a command to the memory.. 1T = 1 cycle / 2T = 2 cycles.

What effects does it have:..

memory bandwidth efficiency.. with 1T you'll get 92~94% of the theoritical bandwidth.. while 2T only delivers 80~85%.. in others words.. 1T command rate makes Athlon64's fly!! :afro:

Razorman
01-24-05, 01:55 PM
It's an Athlon64 thang :D

It's the number of cycles before issueing a command to the memory.. 1T = 1 cycle / 2T = 2 cycles.

What effects does it have:..

memory bandwidth efficiency.. with 1T you'll get 92~94% of the theoritical bandwidth.. while 2T only delivers 80~85%.. in others words.. 1T command rate makes Athlon64's fly!! :afro:


It's not just an Athlon 64 "Thang" it affects all systems & cpus.

Lowering the latency can give you huge performance increases.

Tr1cK
01-24-05, 04:22 PM
I only get like 86-85% effenciency (sp?) with mine set to 1t like I have it now in sig. When I clock back to like 236fsb and run cas 2 I get 91%. When I clock cas 2 back up to 240fsb I get back down to 86%. Anybody know what gives? Why does 4mhz make the difference between 91% and 86%?

Does SANDRA have a bug with the K8N Neo2? It shows my RAM as being set at 2.5-3-3-3 not 2.5-3-3-11. It also shows 2-bank interleaving and not 4.

ynnek
01-24-05, 05:29 PM
yea.... but if you look at all the other game benchmarks, painkiller didn't even budge, and d3 didn't really gain much at all..

So IMO, I agree with this article's conclusion... If your on a budget, the extra $ can probably spent better elsewhere..

but if your prepared to spend alot of $ anyways on everything, then go right ahead..

Mr_LoL
01-24-05, 05:50 PM
mine is set at 2T. According to sandra my bandwith efficency is 71% :thumbdwn:

jAkUp
01-24-05, 06:08 PM
Just lowered mine to 1T and it seems to be perfectly stable... we will see though :)

Blacklash
01-24-05, 06:28 PM
Just wanted add, yes, 1T for teh win!!

Filibuster
01-25-05, 10:03 PM
I only get like 86-85% effenciency (sp?) with mine set to 1t like I have it now in sig. When I clock back to like 236fsb and run cas 2 I get 91%. When I clock cas 2 back up to 240fsb I get back down to 86%. Anybody know what gives? Why does 4mhz make the difference between 91% and 86%?

Does SANDRA have a bug with the K8N Neo2? It shows my RAM as being set at 2.5-3-3-3 not 2.5-3-3-11. It also shows 2-bank interleaving and not 4.

Sandra shows my memory timings all stupid also.
I don't remember what it reports but its way off.

JoKeRr
01-26-05, 05:21 PM
can u have a 0T instead, no wait, just execute?? wouldn't that be more efficient?

Viral
01-26-05, 05:49 PM
Just lowered mine to 1T and it seems to be perfectly stable... we will see though :)

You mean all those SLi benchies were done with a 2T command rate?? :eek:

PSYCHODAD
01-27-05, 03:00 AM
Just wanted add, yes, 1T for teh win!!

Agreed, I've ran my A64 system at 1T since day one and it works smashingly.

CMav
01-27-05, 01:26 PM
I am a noob at overclocking and such. How do you change the 1T/2T settings? I have the Asus sli-deluxe board.

MustangSVT
01-27-05, 01:36 PM
Hrm, those graphs are at 640x480. My current memory is OCZ Premier Series PC3200. I'll run some tests at 3.0-4-4-8, then 2.5-3-3-7, and then 1T and 2T on the 2.5 and come back with results. Let's see how realistic this whole latency thing is.

My prediction is that if you haven't realized yet, difference between memory timings at the realistic settings I play at with my 6800GT like 1152x864 4xAA 8xAF should be between 0 to 1%. I'll see how it goes.

PSYCHODAD
01-27-05, 02:49 PM
I am a noob at overclocking and such. How do you change the 1T/2T settings? I have the Asus sli-deluxe board.
In the bios.

jolle
01-27-05, 03:33 PM
Hrm, those graphs are at 640x480. My current memory is OCZ Premier Series PC3200. I'll run some tests at 3.0-4-4-8, then 2.5-3-3-7, and then 1T and 2T on the 2.5 and come back with results. Let's see how realistic this whole latency thing is.

My prediction is that if you haven't realized yet, difference between memory timings at the realistic settings I play at with my 6800GT like 1152x864 4xAA 8xAF should be between 0 to 1%. I'll see how it goes.

Ofcource it all depends on where the bottleneck is..
what you do now, is to shift the load, from very low on the video card, to high, which could potentially mask the pure benefits of RAM and CPU settings, such as timings..
Results will vary from game to game, some games are heavier on CPU then others, in some the load on the GPU will mask the CPU performance..

So testing at 640x480 shows the pure benefits, if you run it at 1152x864 4xAA 8xAF, and get no improvments, all you prove is that your setup, in that game, is GPU bottlenecked..