Building a computer for a girl

[amjedm]

I asked her and she didn't mind the Akasa Zen - she said it was like an iPod.

I built my niece a computer in an Akasa Zen (black), I think she's happy with it

[kalniel]

You can have a single stick of DDR2 memory, but it will only run half as fast, nerfing the performance. You should get 2x512mb - it's usually only a couple of percent more expensive than a single stick.

Not quite true. The RAM will run at full speed just fine, even if you have a single stick. What you will have though is half the bandwidth. This isn't the same thing as speed at all, and argueably the AM2 already has more bandwidth than it knows what to do with, so you don't actually loose much in performance at all.

[my new pc is a]

That's a fairly pedantic point, as for all intents and purposes to the cpu the memory speed is the memory bandwidth speed. Of course it all depends how you interpret the word 'speed'.

Semantics aside, your point on AM2 not being much faster with a dual DDR2s interests me. Do you have any links/benchmarks for that?

[DanceswithUnix]

I run an AM2 3000+ box with a single stick of RAM. Performance is fine.

You get the same bandwidth with single DDR2 as an older S939 box in dual channel DDR mode, and that was not bandwidth starved. The "upgrade" to DDR2 when AMD went socket AM2 was only a few percent, so I suspect that is what you are giving up.

Edit: That is with an Asus 430 mobo, I think it was the M2NPV-VM (I wanted TV out and gigabit networking).

[kalniel]

That's a fairly pedantic point, as for all intents and purposes to the cpu the memory speed is the memory bandwidth speed. Of course it all depends how you interpret the word 'speed'.

Semantics aside, your point on AM2 not being much faster with a dual DDR2s interests me. Do you have any links/benchmarks for that?

I don't. But it's based on the logical facts of how bandwidth increases on the AM2 do not translate into any significant increase in FPS with games (for example using higher speed memory without adjusting the CPU speed). And yes there are benchmarks for that and no I can't be bothered to find them

Please feel free to show any contrasting benchmarks where an increase in bandwidth alone on the AM2 has lead to any significant increase in FPS and I will consider revising my view.

[my new pc is a]

I didn't think this was a focused gaming machine. At this price point the graphics card will usually be the limiting factor in terms of FPS. In Firefox or Office the higher speed memory will most definitely be useful.

Remember on an AM2 you'll probably be using DDR2 533. So in single channel mode you'll be running a dual core proc on "266mhz" (133mhz double pumped) memory. That's only PC2100 speeds. My cheapo Athlon XP 1700 from 5 years back had that!

2x512mb of DDR2 533 is just £46.99. By contrast the cheapest single stick DDR2 533 is £45.14. And the cheapest 1gb DDR1 400 is £45.81. Not really huge savings.

I guess this issue here is upgradability of the memory at a later date. If this is the case, then maybe consider the Gigabyte GA M61PM-S2. It's only 40 quid, uatx form factor, AND got 4 DDR2 slots. So you can run at the full rocketship speed from the get go, will keeping the ability to add a couple of 1gb DDR2 sticks later

[DratUK]

Dual channel vs Single Channel

http://www.devhardware.com/c/a/Memory/Dual-Channel/

Written in 2005 leave you to draw your own conclusions

[directhex]

Remember on an AM2 you'll probably be using DDR2 533. So in single channel mode you'll be running a dual core proc on "266mhz" (133mhz double pumped) memory. That's only PC2100 speeds. My cheapo Athlon XP 1700 from 5 years back had that!

your maths is suspect.

[kalniel]

And it's just not true. A single channel DDR2 533 stick runs at 266mhz real speed, 533mhz effective.

A dual channel pair of DDR2 533 sticks run at 266mhz real speed, 533mhz effective.

If FF or office are the applications of choice, then the difference in memory bandwidth is definitely not a vital concern - the most limiting factors in performance are either the internet, the user, or the hard drive.

And then, look at the super Pi result from the link posted a few posts up - single channel ram was only 2% slower, and that was for a relatively intensive process running at 100% cpu load. You aren't going to notice any difference running FF or office.

[DanceswithUnix]

Running the Linux benchmark "hdparm -T /dev/hda" to find out how fast the system can read from main memory hard disk cache:

Socket 939 (Venice) dual channel (4 sticks Geil value PC3200)...

[root@dot ~]# hdparm -T /dev/hda
/dev/hda:
Timing cached reads: 1908 MB in 2.00 seconds = 953.28 MB/sec

Socket AM2 single channel PC5400 (1xCorsair Value 512MB)

[root@gate ~]# hdparm -T /dev/hda
/dev/hda:
Timing cached reads: 1716 MB in 2.00 seconds = 856.74 MB/sec

So the single channel AM2 is 90% of the memory performance of my main S939 machine. This is not a fair fight though, the S939 machine is fully loaded and then overclocked, the AM2 is standard. Still, the point remains that there is no way I can see the AM2 memory bandwidth causing a problem.

[directhex]

there are applications which depend on memory bandwidth

you won't run them on a windows desktop