Advice on Memory Upgrade

[Overkill0001]

Hi, this is my first post.

I have a dilemma about RAM. Here is my PC:

Processor: AMD Athlon 64 X2 4200+ 2.2G
Mainboard: ASUS M2N-E AM2 NVIDIA nForce 570U
Memory: CORSAIR XMS2 (2 x 512MB) DDR2 800

When I want to upgrade memory, what route do you advise I take?

I read that filling all slots will cause it to run in 2T command rate. I was having trouble deciding because I wasn't sure if using 4 sticks would cause it to slow down.

Then I read a post:

If you add 4 sticks of DDR2, you will be using 2T timings. If you are using 2 sticks of DDR2 at DDR2-800 or higher, you will also be forced to use 2T timings.

Is that true?

If so, this would mean I shouln't worry about being forced to 2T anyway because I will be using 2 sticks of DDR2 800 in the first place?

(I can't currently check if my memory is in 1T or 2T because I don't have a graphics card now).

I have been considering either just getting another 2x512MB, for 4x512MB in dual chanel. Or even save up and get 2x1GB for 3GB in dual channel. Or just get a 1GB stick and leave the other open, but not have dual channel.

I'm not sure if I want to go with 4x512MB because in the future 2GB might not be enough. I'm mostly trying to be more futureproof.

Please give me advice on what you would do in my situation.

BTW, it's mostly for games.

 

[Rob Williams]

Welcome to the forums!

The quote from me is true... it's usually how it works. Higher than DDR2-800 with 1T is possible, but you will need much more voltage. At that point though, higher frequencies will offset the loss from 1T.

If you don't want to overclock, which I don't assume you do, then I would recommend 2 * 1GB of PC2-6400 stuff. That will allow you 1T timings without an issue. To be honest though, I've been playing around with AM2 all weekend, and the difference from 1T to 2T seems to be far lower than on 939/940/775.

For an example, when I ran 2T instead of 1T at a specific setting (I think it was DDR2-800), I lost 200MB/s on average in SANDRA. 1T gave me something like 8800/8800... so 200MB/s lower than that is nothing to worry over. At that point, more bandwidth is not going to really make a difference except in benchmarks.

So to be honest, I would just go with 2 * 1GB with 1T. You could try 2 * 1GB + your 512MB * 2, but that would force 2T. You could just test both configurations and decide which you like better. I am willing to bet 2 * 1GB will be the preferred route though.

 

[Overkill0001]

Thanks for the fast reply! Right, I wasn't planning on overclocking. My motherboard can only go to 1.95V on the memory.

http://www.techreport.com/etc/2005q4/mem-latency/index.x?pg=8

According to that, the difference between 1T and 2T is nothing on higher resolutions. Also Sandra seems to exagerate the difference between 1T and 2T.

Also I am confused about you saying I should go with 2 * 1GB with 1T, when earlier you said DDR2 800 forces 2T.

 

[Rob Williams]

Yes, the difference between 1T and 2T in gaming is minimal. If you'd ever notice the difference, it would likely be when you have a lot going on at once, or multimedia production. Or similar.

I am not sure about SANDRAs exagerations, but Everest reports similar figures also. Like I said though, 2T and 1T don't seem to matter -as- much with AM2... at all. The differences are minor, really.

As for the confusion, I didn't word things properly. It will likely depend on the kit, but DDR2-800 1T -is- possible. It's when you go beyond that it becomes increasingly more difficult.

 

[Overkill0001]

Actually I just read a couple reviews on my RAM where people could NOT get it to run at 1T. So I might as well just fil up all my slots anyway right, since it's probably 2t in the first place.

BTW, I truly appreciate your help.

 

[Rob Williams]

Yes... it seems it's definitely a 'per kit' basis. I have run three kits so far today at DDR2-800 with 1T reliably, with only 1.9v. It's luck more than likely.

If you are not going to be overclocking, there should be no issue with the 3GB of ram, unless you can manage 1T with your 2GB kit.

No problem about the help, it's the reason we have a forum

 

[Overkill0001]

My RAM defaulted to 2T like I thought. I have yet to try it on 1T. I think I will at least update the BIOS before trying that.

 

[Rob Williams]

Yeah, ram will always default to 2T. It's up to you to enable it. Let me know how you make out with it.

 

[Overkill0001]

Okay, I set it to 1T. It's fine and stable but I haven't noticed any difference. My 3Dmark06 score went from 4371 to 4365 so it actually dropped an insignificant amount. Where would I be more likely to tell the difference between the 1T and 2T command rate? Game loading times?

 

[Rob Williams]

More likely with game loading times, yes. To see some result, try downloading SANDRA: http://www.sisoftware.net/

Then run the memory bandwidth benchmark, and you will get a buffered result. You -should- notice a difference between 1T and 2T. But like I said, it seems less important on AM2 than it was on 939.

 

[Overkill0001]

I timed the loading times in HL2, Quake 4, and CounterStrike Source (starting up the game, and loading a saved file [in CSS just loading it up]). I used CPUZ to check and make sure the CR was right after changing the BIOS.

Interestingly, the times were EXACTLY the same. After finding this out, I think I'm not going to worry about filling up all four slots.

 

[Rob Williams]

Well, the thing is on AM2, the bandwidth is -so- good to begin with, that the more you add is going to be useless. People who get 13,000/13,000 in SANDRA... it's just for bragging rights really.

Perhaps with hardcore multimedia you may notice a difference, or games that hog a ton of ram. Still, if you want four slots, go for it

After you do that though, I'd do some quick tests yourself to make sure that it's not taking away from any performance. I haven't had a weird combo like that in my machine, so I haven't tested it out.

 

[Overkill0001]

I went ahead and tried out Sandra for the heck of it. I don't know exactly what to go by but anyway;

@1T

RAM Bandwidth Int Buff'd iSSE2 : 6850 MB/s
RAM Bandwidth Float Buff'd iSSE2 : 6841 MB/s

@2T

RAM Bandwidth Int Buff'd iSSE2 : 6685 MB/s
RAM Bandwidth Float Buff'd iSSE2 : 6651 MB/s

even in Sandra the difference is only 2.5%. In current games and loading times it's zilch. Even in something that really needs ram it seems like I wouldn't actually notice a differece.

I'm glad I looked into this, some people I asked made it seem like going from 1T to 2T would slow it down so much that it would be better to just take out the 2x512MBs and only use 2x1GB. I think just keeping 3GB will be better, in the future that's what apps will need anyway so I might as well get it now. I just hope 2x512MB and 2x1GB will be okay, I have a feeling it will, but to be safe I'm gonna use the exact same brand and timings.

I'll do some 1T vs 2T tests on my Athlon 2800+, Socket A and I bet there will be a bigger difference.

 

[madmat]

I timed the loading times in HL2, Quake 4, and CounterStrike Source (starting up the game, and loading a saved file [in CSS just loading it up]). I used CPUZ to check and make sure the CR was right after changing the BIOS.

Interestingly, the times were EXACTLY the same. After finding this out, I think I'm not going to worry about filling up all four slots.

I think you'll find that the stumbling block on load times (once you get past the point that the game is saturating the RAM at any rate) is the CPU and the speed at which it can decompress and load levels, textures and what have you.

 

[Rob Williams]

Yes, what Matt said. And also, I like I said, AM2 doesn't make much of a difference. 2T is slightly more loose latency wise over 1T... it really depends on what you are doing when you will see a difference. Maybe if you were converting video overnight... 1T may speed it up by 30 minutes. It's hard to say really.

This is a graph using 939... note the differences between 1T and 2T here. This is a platform where you may see differences. AM2 is a wicked platform for having great bandwidth though, so no real need to worry either way.