RAM: 1gb vs. 2gb

[Richard]

In this mini-review I try to investigate which benefits, if any, exist in upgrading from one to two gigabytes of main RAM when it comes to games. The loading times of the timedemo were recorded using a manual stopwatch. A graph of the benchmark performance is also present.

With 1gb of RAM Windows was set to run with a fixed size 1gb page file. With 2gb of RAM the Windows page file was disabled so both configurations have a total of 2gb of virtual memory.

Test System

CPU:

Intel Pentium 4 3.2C​

Motherboard:

Asus P4C800 Deluxe (Intel 875P)​

RAM:

1GB (2x512MB) PC3200 DDR400 RAM (Kingston)
2GB (4x512MB) PC3200 DDR400 RAM (Kingston)​

Video Card:

ATI Radeon X850 XT Platinum Edition 256MB AGP​

Video Drivers:

ATI Catalyst 6.2 WHQL​

Operating System:

Windows XP Professional SP2​

DirectX Runtime:

DirectX9.0c​

This system has a fairly tweaked installation of Windows XP. After boot the peak and current commit charges stand as follows:



There was some variation of the peak commit charge after each benchmark between 1gb of 2gb although they are probably due to Windows than anything else. Here is a comparison with all the benchmarks.



Only Battlefield showed any significant change (around 100mb more with 2gb RAM) and surprisingly Far Cry and Serious Sam 2 with uncompressed textures and lightmaps showed the 1gb RAM configuration with a higher peak commit charge.

What's most relevant here, however, is the doubling of the commit charge between uncompressed and optimal textures & lightmaps in Serious Sam 2.

Games Benchmarks - DirectX

F.E.A.R.
v1.03
1024x768
Maximum settings (SoftShadows OFF)
4xAA, 16xAF

Benchmark run:



In F.E.A.R. the actual in-game performance changed little between both configurations with the average fps being equal. What did change in a dramatic way were the loading times where it can be infered that with 1gb of RAM this game is dipping into the page file. Nearly halving the load times is quite an achievement, even if the worst loading time is not that severe.

[continues in next post]

 

[digitalwanderer]

More, MORE!!!!

 

[Richard]

Battlefield 2
v1.2
1024x768
Maximum settings
6xAA, 16xAF
Adaptive Anti-Aliasing, High Quality Anisotropic Filtering

Benchmark run:


Although the benchmark was run no results were recorded. BF2's demo playback is not very deterministic with explosions and physics never behaving exactly the same. The variance is pretty severe to invalidate any accurate results. The loadings remained the same which is surprising, especially considering BF2 has more than double the peak commit charge of F.E.A.R.

Serious Sam 2
1024x768
Maximum settings (Texture and Lightmaps set to Optimal)
4xAA, 16xAF

Benchmark run:



There is a slight performance drop in the benchmark with 2gb of RAM but not statisticaly relevant. The loading time improvements are more significant, especially when using uncompressed textures and lightmaps. I didn't record any benchmark results with 32-bit textures/lightmaps. With a 256mb video card the fps were in single digit numbers as the game was most likely saturating the AGP bus invalidating the results. This leads me to believe 2gb of RAM in SS2 is only recommended if you have a 512mb video card.

[continues in next post]

 

[Richard]

Far Cry
v1.32
1024x768
Maximum settings
4xAA, 8xAF
Adaptive Anti-Aliasing, High Quality Anisotropic Filtering

Benchmark run:



Like in F.E.A.R. there was no significant performance differential between both RAM configurations and only the first time loading showed an improvement which means this game isn't stretching 1gb of RAM.

Games Benchmarks - OpenGL

Quake 4
v1.0.0.5 beta 2
1024x768
Ultra quality
4xAA, 8xAF

Benchmark run:



Quake 4 is the only game in this suite to show an improvement in framerates with 2gb of RAM, and then only of a few fps. The graph also shows this configuration got consistently higher minimum framerates. Loading time improved too though Quake 4's internal asset manager is quite optimised which means if something is already in memory it does not load it thus the instantaneous second loading. Therefore 2gb of RAM will show even larger benefits after several different loadings.

Conclusion
In general there were no performance differences during benchmarks while loading times showed none to impressive improvements even though all games (except Serious Sam 2) ran at maximum quality settings. So, is the extra 1gb of RAM essential right now? Unqualified no. However, if you have a 512mb video card or if you exasperate when waiting for a level to load, it is recommended. An extra gigabyte of ram is also relatively inexpensive right now so why not?

 

[Cartoon Corpse]

do Boiling Point. everyone has said 2G is the sweetspot (i had 1G)...had lag.

anyway boiling point's process 'footprint' is 2.17G. and there are no loadzones.

 

[MuFu]

Nice work. Hope nobody was watching when you were bent over the PC with a stopwatch.

To what would you attribute the huge dip in the early part of the Far Cry 1GB plot?

 

[Richard]

Nice work. Hope nobody was watching when you were bent over the PC with a stopwatch.

If anybody did, they're used to it

To what would you attribute the huge dip in the early part of the Far Cry 1GB plot?

That probably occurred in the 2gb configuration too. It just so happens that FRAPS doesn't quite tell you when the actual benchmark begins; it automatically starts recording once it detects a OGL/D3D context, which for some games is right at the main menu, so you can see in some of the graphs both plots are offset a bit and do not quite match up. In that FC graph it means the initial part of the 2gb graph was left out. I had to resort to manually timing it to know how much seconds I should ignore at the beginning and end of the FRAPS log - which is not ideal. (I'm using the last free version - don't know if the latest - paid - version fixes this) It also only records the framerate at every second which is far more problematic since a game could access the page file in between these FRAPS readings and they wouldn't show up as a huge drop in framerate. Anyway, the average framerates still show both configurations stayed pretty much the same.

I originally meant to post my subjective experience with both configurations but here it is: I only noticed an improvement with 2gb in Quake 4 and Battlefield 2, especially after playing for some time, loading several levels, etc.

do Boiling Point. everyone has said 2G is the sweetspot (i had 1G)...had lag.

anyway boiling point's process 'footprint' is 2.17G. and there are no loadzones.

I don't have Boiling Point, and isn't that game supposed to be really buggy and slow? I don't want to use unoptimised games; they may stress the system but they probably aren't representative of most games. If I didn't feel this way I could have benchmarked with Visual Studio, Office, ClimatePrediction, etc. opened at the same time. That would definitely show 2gb as being much, much faster but how many people play games with those kinds of programs in the background?

I should note that 2gb will probably be essential for gaming in Windows Vista. As I mention in the first post, XP for me is using ~100mb after boot. The latest beta of Vista is using >500mb after boot (and that's after I disabled a couple of things).

 

[aaronbond]

just wondering, but why only 1024x768 on an x850? Monitor size, personal preference, what? Thanks.

 

[Richard]

just wondering, but why only 1024x768 on an x850? Monitor size, personal preference, what? Thanks.

My monitor can go higher but I didn't want to overly stress the pixel fill-rate for these benches.

 

[Andy]

I'm interested to know if there are any significant differences if you used 2x 1GB DDR Sticks or even 1x 2GB DDR stick instead of 4x 512MB DDR, also if you used DDR2 instead of DDR.

 

[Tahir2]

I'm interested to know if there are any significant differences if you used 2x 1GB DDR Sticks or even 1x 2GB DDR stick instead of 4x 512MB DDR, also if you used DDR2 instead of DDR.

That would involve too many variables to be useful.

4x 512MB would probably only work at DDR2700 speeds.
1x 2GB stick would be single channel as opposed to dual channel.

In both cases the performance would be lower I would think.

DDR to DDR2 would mean a different CPU and motherboard. CPU could stay the same but motherboard would need different chipsets IIRC.

 

[Rys]

4 x 512MiB is doable without dropping speed on Athlon 64 and P4. On A64, all you sacrifice is 1T command rate and therefore a bit of bandwidth.

From a price point of view, and curiousity's sake, 4 x 512MiB vs 1 x 2GiB vs 2 x 1GiB would be a great piece to read.

I myself had to go 4 x 512MiB simply because I bought my memory in 2 stick lots, over time. So I'd like to see what selling the first two and going straight for 2 x 1GiB would have bought me.

 

[Tahir2]

Rys you want that tested on an NForce4? I don't think it is doable on a P4 for me and I would be limited to free 3D benchmarks (like 3DMark).

From my limited experience going from 1T to 2T doesn't impact real world nearly as much as synthetic benchmarks state but dropping speeds (and I have seen it necessary on certain chipsets) does impact performance. I hope ATI have fixed this in the Xpress 200 replacement...

 

[Richard]

4 x 512MiB is doable without dropping speed on Athlon 64 and P4.

Yep. I'm using 4x512 and still running at PC3200.

 

[Tahir2]

Well there you go then..

 

[BRiT]

Yep. I'm using 4x512 and still running at PC3200.

And for those into overclocking 4x512[2t] and 2x1024[1t] have both hit and exceeded 300Mhz FSB in the AMD realm.

 

[MuFu]

Repost from other thread:

On my system (Asus A8V, Opteron 144) anything above 9x260HTT fails MT86+, while with 2 DIMMs 9x300MHz is rock solid, even at 1T.

Pretty much killed my overclock. Should have bought a 146!

 

[Tahir2]

I think the fact that you have an ASUS A8V is hampering your efforts.

You wanna overclock - get a DFI!

 

[MuFu]

I think the fact that you have an ASUS A8V is hampering your efforts.

That could well be the case, but I would have thought that the A64's mem controller is a likely point of failure when running 4 DIMMs (since it obviously struggles with 1T at moderate speeds). For me, boosting the core voltage helped the O/C in memory/HTT-limited situations; something that wasn't true with 2 DIMMs.

This is the highest HTT speed I've seen on an A8V, incidentally. Was expecting about 275, max.

 

[Tahir2]

Probably because everyone else gave up and threw theirs in the bin.

Erm...