[Eug]

Anybody try comparing performance yet between various grades of RAM in the G5?

It seems that Crucial.com simply specs CL3 PC3200 RAM for the Apple G5, which leads me to believe that that's all Apple is supplying. There are several sites now offering shorter latency RAM now.

I know it wouldn't make a humungous difference (eg. ~5%), but I'm interested to know if it makes ANY difference. ie. Does the G5 automatically run at the faster speed if there is faster RAM?

IOW, has anyone REMOVED their stock RAM and then put in faster RAM, and compared performance with the stock RAM?

[MisterRogers]

Goodmorning Eng. I bought 5 sticks of 512 DDR 400 xms low latency cas 2 Crucial ram for my Dual. Stock is cas 3. Under profiler, the G5 reports the cas, but who knows if it does anything with it. When the G5 arrived, I installed all, and found out that I had one bad stick. RMA'd the bad, installed 2 along side my stock 512 (1 Gig wasn't enough :-) When my rma'd stick returns next week, I'll be putting in the either Crucial pair, and removing the stock ram. I then plan to run some benchmarks to give some cas 2 numbers for comparison. My box has the 9800, so I'll do PSBench, UT2003, what else? I'm not prepared to go so far as to re-image my system to be sure everything's optimal, but I'll do what I can. BTW, I'm running B74 Panther, and will likely be running something new by then. Come to think of it, If I keep up with my seeds on the G5, I might as well benchmark right after a clean installation. Anywho, numbers will be forthcoming.

[Eug]

Looking forward to the benches.

I bought 5 sticks of 512 DDR 400 xms low latency cas 2 Crucial ram for my Dual.

BTW, is it right to assume you didn't buy the RAM at the www.Crucial.com store? The Crucial website only specs CL3 RAM.

[MisterRogers]

Correct. I bought it from NewEgg.

[Cadaver]

I have two sticks (2x256MB) of Corsair XMS PC3200 low-latency TwinX RAM (2-2-2-6-1T).
Works great in my PC. Isn't recognized at all in my G5.

[Severed Hand of Skywalker]

Sweet Jesus! 6:40 in the morning and you are worrying about 5% speed differences in RAM!

[Eug]

Originally posted by Cadaver:
I have two sticks (2x256MB) of Corsair XMS PC3200 low-latency TwinX RAM (2-2-2-6-1T).
Works great in my PC. Isn't recognized at all in my G5.

Weird. I would have assumed it would work, but I guess not.

[Hydra]

Originally posted by Eug:
Anybody try comparing performance yet between various grades of RAM in the G5?

I read something over at the forums at XLr8 that the dual G5 was OK with PNY's PC3500 DDR433 CL2.5 from COMPUSA. I found it for $120 per 512MB. I installed it and ran the Apple Hardware Test disc (the test took like 50 minutes) and it reports it as CL2.5 as opposed to Apple's OEM stuff that showed it as CL3.0.

I pulled the OEM CL3 512MB RAM and put in only the 1024MB of CL2.5 PNY RAM. I fully realize that having double the RAM might in and of itself improve performance over the OEM stuff but in all the benches I did it seemed to make little or no difference.

Here are a few tests w/ apps in identical preferences/configs on the same machine,
dual G5/Radeon 9800:

CINEBENCH 2003 v1

***********************************************
W/512BM CL3
***********************************************

Rendering (Single CPU): 220 CB-CPU
Rendering (Multiple CPU): 400 CB-CPU

Multiprocessor Speedup: 1.81

Shading (CINEMA 4D) : 268 CB-GFX
Shading (OpenGL Software Lighting) : 713 CB-GFX
Shading (OpenGL Hardware Lighting) : 1381 CB-GFX

OpenGL Speedup: 5.15

***********************************************
W/1024MB CL2.5
***********************************************

Rendering (Single CPU): 220 CB-CPU
Rendering (Multiple CPU): 401 CB-CPU

Multiprocessor Speedup: 1.82

Shading (CINEMA 4D) : 268 CB-GFX
Shading (OpenGL Software Lighting) : 712 CB-GFX
Shading (OpenGL Hardware Lighting) : 1400 CB-GFX

OpenGL Speedup: 5.22
***********************************************
***********************************************

Quake 3 Highest settings @ 1680x1050

Demo4 CL3/CL2.5 = 187.7/188.1

Crusher CL3/CL2.5 = 133.4/133.6

***********************************************
***********************************************


UT20031280*1024 highest settings:

flyby antulus

cl3.0

/Score: 79.283043
Min FPS: 41.052471
Average FPS: 89.883400
Max FPS: 214.650925

cl2.5

/Score: 79.023087
Min FPS: 42.518604
Average FPS: 89.531830
Max FPS: 218.034561

botmatch antalus

cl3.0

/Score: 34.575390
Min FPS: 12.976949
Average FPS: 34.576248
Max FPS: 76.790764

cl2.5

/Score: 35.444725
Min FPS: 15.279066
Average FPS: 35.444778
Max FPS: 72.729271


botmatch asbestos

cl3.0

/Score: 54.080051
Min FPS: 16.886076
Average FPS: 54.353783
Max FPS: 122.748871

cl2.5

/Score: 55.415985
Min FPS: 26.154280
Average FPS: 55.743450
Max FPS: 122.104843

***********************************************
***********************************************

I'd have to say that if there is a difference between the two speeds of RAM, it is very small. These results are far from scientific but I really don't think anyone will find any significant difference. The other question is if the PNY DDR433 PC3500 spec (which is claimed by PNY to be completley backwards compatible with PC3200/2700/2100) has any issues if run at the slower 400 MHz speed. Apple's hardware test disk did report it as cl2.5 so I wonder if Apple has the G5 just run all cl2/2.5/3.0 at 3.0 speed anyway.

-Jerry C.

[Eug]

Wow thanks Hydra! That was a lot of work.

Yeah, I think your conclusion is right. Memory speed makes minimal difference here at best. Perhaps it's partially because memory bandwidth is already so high, and it is not the limiting factor in these benches.

I would have suggested running Xbench too, but the results are likely too variable to compare minor memory speed differences.

I'm not sure how useful those Quake III numbers are though. If anything I'd expect a more representative test would be to run it at 640x480. (No need though, since we've got the other tests.)

P.S. It's too bad that UT2003 isn't better dual-optimized. UT2003 botmatches could really stand to get another jolt of CPU speed to calculate all that bot AI.

[Hydra]

Originally posted by Eug:

P.S. It's too bad that UT2003 isn't better dual-optimized. UT2003 botmatches could really stand to get another jolt of CPU speed to calculate all that bot AI.

I agree, but the developers of UT2003 said they will be rolling in some nice G5 improvements into it in the near future. I don't know if this means a patch for UT2003 and/or UT2004 will include these. So I think we may see a nice boost. I really haven't played UT2003 at 1680x1050 and thought to myself it was choppy - it seems pretty smooth almost everywhere with all the eye-candy on in wide-screen.

These tests were all done with sound on by the way (testing different RAM in the same machine means it's the only variable, so sound will hurt both RAM's performance equally). The beauty of the dual G5 is that I can play Spearhead or UT2003 while doing audio iChat at the same time and the performance barely suffers. I would like to see a PC bench for UT2003 doing the same I don't consider myself a hardcore gamer by any means but I do find myself enjoying the full resolution 20" CD goodness my dual G5 is giving me.

-Jerry C.