Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

mac57mac57

macrumors 6502a
Original poster
Aug 2, 2024
640
387
Myrtle Beach, SC
I have always wondered whether equipping your Power Mac G5 Quad with PC2-4200U-444 RAM would make for a faster machine than if the same machine were equipped with PC2-3200U-288 RAM. It seems obvious that the faster RAM ought to make for a faster machine, but as we shall see in a moment, it doesn't really make a whole lot of difference!

Since I have just resurrected my LCS-cooled Quad, which has been sporting 8 GB of PC2-4200U-444 RAM up until now, and I have been totally unsuccessful selling the 16 GB of PC2-3200U-288 RAM that came with it originally, I sensed an opportunity to finally answer this question.

So, using XBench 1.3's Memory test, I tested first with the PC2-4200U RAM and then swapped out that RAM for the PC2-3200U RAM and tested again. The results are presented below, side by side, with the PC2-3200U results on the left and the PC2-4200U results on the right:

PC2-3800U RAM vs. PC2-4200U RAM, PMG5 Quad.jpg


As you can see, the results are virtually identical within a small tolerance. Some things are faster with the 3200U RAM and some are faster with the 4200U RAM, but the bulk are more or less the same.

So, worried about the RAM that came with your machine? Should you upgrade PC2-3200U RAM to PC2-4200U RAM to get more speed? My advise would be not to bother... it does not seem to make much difference.

This is with XBench 1.3. Are there any known weaknesses of this test suite that might skew these results?
 
Last edited:
  • Like
Reactions: NikolaPPC
It should probably be noted that stock configurations shipped with the faster PC2-4200U memory, and while any competent machine will downshift all of the memory to the same speed, it's still best practice to make sure that all of the memory in a system is the same speed.

Also, if you are looking to max your system out, 2GB DIMMs do not come in the slower 3200 speed anyway outside of server memory of the ECC variety, which A1117s can use, though YMMV on that one.
 
Largely agreed @Doq, but 2GB sticks do come in the slower PC2-3200U speed. My LCS Quad now has a full complement of 16 GB of said RAM... works like a champ! I cannot say whether it is ECC or not, but it is from a mixture of manufacturers, and so I doubt it.

However, like you, I always make sure that all the RAM in any given machine is of the same speed. Else you just don't know what speed any of the RAM is being accessed at.
 
  • Like
Reactions: Doq
According to manual Quad has 533 MHz memory controller, so fastest continuous data stream has PC-4200 ( 533 MHz) modules.
Bigger speed of lower-frequency modules is caused by different CAS latency.

Btw, what is PC2-3200U-288 ? Is it CAS latency or something else?
If it is, it probably should be -222-8 in more standard way, where 8 is ns. But who knows.

In your example second modules are PC4200-444-12, i.e. 12 ns latency. So if reading from RAM is not continual, it takes 50% more time to read data from new place.

If you are search for speed, I recommend PC2-4200 CL 333 modules. It should be fastest.
 
You can use Open Firmware to read the SPD info from the DIMMs. I think the i2c addresses are 0x50 to 0x57 (0xA0-0xAE for reading).
Connect to another computer using telnet to capture the results from Open Firmware.

You need a utility to parse the SPD info. You can convert the hex for each to a binary file and use a Windows or Linux utility.
Linux has i2c-tools which includes decode-dimms.

The parsed info should include timings and latencies.
 

Attachments

  • read-i2c.txt
    10.9 KB · Views: 6
Maybe it is lack of my knowledge, but:
Modules PC2-4200U-444: I assume that here is 444=CL. For 533 MHz ( MT/s) module it means 12ns.
And sometimes such modules are marked PC4200-444-12.

And number 8 (8ns) was my pure speculation of meaning of very strange PC2-3200U-288 code. What is -288?
 
And number 8 (8ns) was my pure speculation of meaning of very strange PC2-3200U-288 code. What is -288?

Both "PC2-4200U-444" and "PC2-3200U-288" are how Mac OS X Sorbet Leopard's System Information app shows the RAM in its "Memory" tab. I am not knowledgeable of the intricacies of RAM naming, so I honestly don't know what the "444" and "288" actually stand for.

You mention "PC2-4200 CL 333" RAM. Since I don't know how to decode the numbering (other than the "4200" part) I am not sure what this implies with respect to CAS latency. Does anyone reading this have their G5 equipped with this class of RAM? Is it in fact observably faster (or at least shows up as faster in XBench results)?

Here is what Google's AI Summary had to say about which was faster, the "444" RAM or the "333" RAM (the emphasis throughout is theirs, not mine):

The main speed determinant for PC2-4200U memory modules is the stated clock speed, which for PC2-4200 is 533MHz, not the trailing numbers like 444 or 333.
The "444" and "333" likely refer to the memory timings (latency), such as the CAS Latency (CL), tRCD, and tRP.
Lower timings (like 3-3-3 compared to 4-4-4) generally indicate lower latency, meaning the memory can respond faster to requests. So, PC2-4200U-333 might be slightly faster in terms of latency compared to PC2-4200U-444, as it suggests lower timings.
However, the overall performance difference between these two modules is likely to be negligible. While lower latency is desirable, real-world performance gains might not be substantial, especially compared to the impact of the memory's clock speed (533MHz in this case).
In summary:

  • PC2-4200U-333 might have slightly lower latency.
  • Both PC2-4200U-444 and PC2-4200U-333 are rated at the same 533MHz speed.
  • The overall speed difference between them is likely minimal.
Therefore, while PC2-4200U-333 might theoretically be slightly faster due to lower latency, the practical impact on performance is likely insignificant.
 
Last edited:
If I simplify memory speed parameters:

Speed is the main factor. On this depends max. speed of continuous data stream. Higher speeds than speed of controller has no direct effect.
Speed has main role for example with stream benchmark and audio+video processing.

Latency is important when accessing memory randomly. I.e. reaction time to read from location which was not previously accessed.
Lower latency (3 vs 4) has only little stream speed impact, upto 5% ( Tom´s HW. )
Latency has important role for example with databases, networking and also for servers.

If I look to your benchmark pictures, there are also stream results, and PC2-3200U-288 is faster than PC2-4200U-444.
It is strange. Are you sure, that "PC2-3200U-288" is really PC-3200 module? I am suspicious, that you have in real faster module and only OSX Profiler shows this value. If you want, you can look on label and chip types on memory module.

In any case, most important are real world results, so if PC2-3200 gives you better results than PC2-4200, keep it.
 
Interesting comment, @sailorMH. TWO of the eight 2GB sticks in the machine are labeled as PC2-4200U but recognized by my Power Mac G5 as PC2-3200U. The other six sticks in the machine are both labeled and recognized as PC2-3200U.

I complained to the seller (bought them on eBay) and they ultimately gave me a nice discount on the sticks, as compensation for the "lower than advertised" speed.

There is undoubtedly some variation in the SPDs, since all six of the sticks that I bought came from the same place and ALL were labeled as PC2-4200U. Four were recognized as PC2-4200U while 2 were recognized as PC2-3200U. I have not bothered trying to decode the SPD values - at some point the law of diminishing returns takes over.
 
  • Like
Reactions: sailorMH
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.