Memory Bandwidth Tests... Any Real Differences (part 2)

[graysky]

About 7 months ago I posted data comparing two memory dividers (1:1 and 3:5 @ 333 MHz) on my then Q6600/P965 based system and concluded that for the 67 % increase in memory bandwidth, the marginal gains in actual performance weren't worth the extra voltage/heat.

 

Since then I've upgraded my hardware to an X3360/P35 setup and wanted to revisit this issue. Again, two dividers were looked at: one pair running 8.5x333=2.83 GHz, and another running @ 8.5x400=3.40 GHz:

 

333 MHz FSB:

1:1 a.k.a. PC2-5300 (667 MHz)

5:8 a.k.a. PC2-8500 (1,067 MHz)

 

400 MHz FSB:

1:1 a.k.a. PC2-6400 (800 MHz)

4:5 a.k.a. PC2-8000 (1,000 MHz)

 

I figured there would be a much greater difference in the 333 FSB case since the memory bandwidth increased by 60 % vs. 25 % in the 400 MHz FSB case. All other BIOS settings were held constant with the exception of the divider (and the strap) and the given FSB. Subtimings were set to auto and as such could vary as managed by the board which I found out, was required since manually settings some of the subtimings lead to either an incomplete POST, or an unstable system.

 

The benchmarks were broken down into three categories:

1) "Real-World" Applications

2) 3D Games

3) Synthetic Benchmarks

 

The following "real-world" apps were chosen: x264, winrar, and the trial version of Photohop CS3. All were run on a freshly installed version of Windows XP Pro x64 SP2 w/ all relevant hotfixes. The 3D games were just Doom3 (an older game) and Crysis (a newer game). Finally, I threw in some synthetic benchmarks consisting of the Winrar self test, Super Pi-mod, and Everest's synthetic memory benchmark. Here is an explanation of the specifics:

 

Trial of Photoshop CS3

Edited May 11, 2008 by graysky

[red930]

some comparisons/info that you might find interesting (haha so we don't have to work on this anymore and can work on newer stuff and looks like you are on top of this)

 

tight timings? loose timings? dividers? 1T or 2T? 754 or 939? (read here!)

 

Socket AM2 vs Socket 939 Comparison

 

 

(we will be moving these threads here eventually but if you wish to run with the information we have and make a complete comparison chart or guide please feel free to use this information as it would be newer than our info and probably better )

Edited May 11, 2008 by Angry_Games

[fire_storm]

Well done. :thumbs-up:

 

I've red both of your guides and it seems like your overall conclusion is that really fast memory does not equal a big performance increase.

 

But would your conclusion be different if you compared DDR2 to DDR3?

 

I wonder what the difference between PC2 8500 and PC3 16000 would be.

[ccokeman]

Not what you would think. Yet!

[Waco]

This is why I advocate buying a lot of cheap memory over buying any kind of "performance" memory.

[ccokeman]

I have done both with mixed results. I have bought volue ram that was incredible( Had DT-43 Hynix chips DDR Pc3200) that did 250MHz. I have bought some supposedly high performance ram DDR3200 still that just stunk. 1 set had BH-5 chips and the other I have no idea since ididnt pull the heat spreader. If you do your research you can find some gems for not a whole lot of coin. But sometimes you get burned.

[six]

I have done both with mixed results.

 

I think the point was that the difference in memory speed and available memory bandwidth between high speed "high-performance" memory and whatever the "standard" stuff is has little practical value in "real world" application. I think they are trying to say that the difference between running whatever cheap DDR2 533 or DDR2 800 and going to some high speed stuff at or above DDR2 1000 won't gain you enough performance increase for applications you might run to justify running high-performance memory.

 

I don't happen to completely agree with that assessment, but I'm pretty sure that is what is being said.

 

My own opinion is that I/O speed holds back more "real world" application of software than does anything else. For one thing, the faster you can get data off your hard drive, the more important your available memory bandwidth becomes. If we compare the same benchmarks used (though maybe not the games... another story for another day) running from a clean system with small buffer, average speed drives, and do the same thing from faster drives with larger buffers, we might certainly see larger performance gaps when changing from lower speed to higher speed memory. The fact does remain that I/O is the weak spot in data intense applications.

[Waco]

The fact does remain that I/O is the weak spot in data intense applications.

I agree, computers are certainly I/O bound these days.

 

However, even with SSDs I can't see high speed memory mattering much at all. Memory just doesn't play much part in computers these days.

[six]

I agree, computers are certainly I/O bound these days.

 

However, even with SSDs I can't see high speed memory mattering much at all. Memory just doesn't play much part in computers these days.

 

That's the thing right there, but the other way around. Nearly a decade ago, I saw a RAMDrive for a server application that was 5 Gb of PC-66. They had it connected to a SCSI BUS. It was raging fast, but of course, volatile, and therefore never caught on. RAMDrives (both virtual and actual) have limited application and we are still a long way from non-volatile SSDs that can change the landscape.

 

These new SSDs are starting backwards. First of all, none of them are really all that fast yet, but disregarding that: What needs to happen is the elimination of the IDE/SATA BUS and any SCSI BUS. Future SSDs need to be on the memory BUS where the CPU can have direct access. I don't think we can escape the necessity of RAM, but we can combine storage and cache on the same device when it becomes possible to attach it to a BUS that can handle both and when SSDs can be capable of achieving the same bandwidth that system memory currently gives us.

 

Until then, the slower the hard drive, the less bandwidth is required from the system memory. My own tests tell me that an application like the Winrar as mentioned by the OP can indeed benefit measurably from faster memory. Packing a 4.5 Gb DVD image into 50 meg chunks from a Stripe 0 pair of Raptors to a high density single WD 640 showed me a difference of a couple of minutes between DDR2 800 and DDR2 1200. Due to the fact that CPU utilization was still very low, I'm still convinced it is the I/O speed that is the hold up, but never the less, the difference was measurable.

[graysky]

I edited the first post switching the highest 400 MHz FSB run from 5:6 to 4:5 (960 MHz vs. 1,000 MHz) and included some info about subtimings to make things more clear.

[Waco]

In some cases, yes, memory bandwidth is essential. Those only pop up on applications that spend a large amount of time manipulating data in memory (like compression). Other things aren't necessarily I/O bound, they just don't suffer much from lower bandwidth.

 

Anyway, buying "cheap" DDR2-1066 is still much more cost effective than the performance stuff that'll run at the same speed but with slightly better timings.

 

High bandwidth is good, I'm not saying people shouldn't go for it. It's just not all that important for most stuff.