DDR2 and Athlon 64 advice needed

[SixToes]

I'm building a new system to be used as a web/VNC/p2p server and media centre in one box. I figure the server tasks can be handled on one core and the media stuff on the other. I'm going for a bang-for-the-buck approach and planning to use these parts:

Case - Antec NSK2400
PSU - Seasonic S12 330
Mobo - Asus M2A-VM HDMI
CPU - AMD Athlon X2 3600+ (Brisbane core)
HD - Western Digital WD5000AAKS
RAM - ???

I'm planning to undervolt and/or use EIST and CnQ, whichever is more effective. I won't be overclocking.

I've not built a DDR2 based machine before and I'm having trouble finding information about how various RAM speed and timings affect real world performance. With DDR you bought the tightest timings you could afford and synched the speed with the FSB, but this doesn't seem to be the case with DDR2. My question is will buying more expensive DDR2-800 be worthwhile, or should I stick with slower (albeit possibly tighter timed) memory?

Any help is appreciated!

Thanks,
SixToes

[jaganath]

I'm having trouble finding information about how various RAM speed and timings affect real world performance.

http://www.anandtech.com/memory/showdoc.aspx?i=2800&p=9

This does not change the fact that the AM2 memory bandwidth is really greater than Core 2 Duo or the fact that AM2 scales better in memory, exhibiting a steeper slope in performance increase as memory speed increases than does Core 2 Duo.

http://www.anandtech.com/memory/showdoc.aspx?i=2800

Our AM2 launch reviews and the article First Look: AM2 DDR2 vs. 939 DDR Performance found that AM2 with DDR2-533 memory performed roughly the same as the older Socket 939 with fast DDR400 memory. Memory faster than DDR2-533, namely DDR2-667 and DDR2-800, brought slightly higher memory performance to AM2.

http://www.anandtech.com/cpuchipsets/sh ... i=2741&p=9

It appears AMD will succeed in launching a faster on-processor DDR2 memory controller. The latest AM2 pre-release samples are showing significant improvements over Socket 939 DDR in both memory bandwidth and latency. Unfortunately, the current AMD architecture running current applications and games doesn't appear to need the additional bandwidth or reduced latency.

http://www.anandtech.com/cpuchipsets/sh ... i=2741&p=3

Benchmark results were very interesting. From a broad perspective, memory bandwidth and latency of DDR400 was matched by DDR2 at a speed of just over DDR2-533. Put another way, DDR2-533 provides equivalent bandwidth and latency to DDR400 on this 4th spin of AM2. DDR2-667 provided a bit better performance, and DDR2-800 3-3-3 provided a 12% improvement (lower number) in Latency over DDR400 and a 13.3% to 28.6% improvement in bandwidth depending on the memory benchmark used.

...Unfortunately, the added memory bandwidth did not translate into the kinds of increases in real-world gaming performance that many might expect...Since increased memory bandwidth did not translate into similar increases in gaming performance we can only conclude that this iteration of AM2 is not particularly memory starved - a result that was really expected. This does open the door for future upgrades and revisions of AM2 that might make better use of the available memory bandwidth with fast DDR2 memory.

My question is will buying more expensive DDR2-800 be worthwhile, or should I stick with slower (albeit possibly tighter timed) memory?

Since you won't be overclocking, I can't imagine the small gains to be had from 800 vs 533 will have a good cost-benefit ratio.

I'm planning to undervolt and/or use EIST and CnQ, whichever is more effective

EIST is a feature of Intel CPUs only.

[SebRad]

Hi, sometime ago I spent a whole lot of time benchmarking an AM2 4200+ / GF 6150 system. The broad conclusion was that an increase in speed was worth more than a drop of timings, ie 667 5,5,5 was faster than 533 4,4,4. Dual channel was worth about a speed grade e.g. dual 533 ~ single 667 but varied depending on the test. As far as remember about 5-10% improvement from 533 to 800, I may have the file online somewhere but will email anyone who'd like it.
If the ATI chipset performs similarly and looking on Ebuyer.com 2GB of DDR2 533/667/800 is about £65/75/100 I would say that 667 looks like the best bet if budget is tight but if not go for the full 800 as isn't much money.
Seb

[EV10]

The broad conclusion was that an increase in speed was worth more than a drop of timings, ie 667 5,5,5 was faster than 533 4,4,4.

In effect, but not precisely so. The thing is that timings are bound to the frequency, determined in ticks - and so 5 ticks on 667 actually take exactly as much time as 4 on 533. Timings don't really suffer.

If the ATI chipset performs similarly and looking on Ebuyer.com 2GB of DDR2 533/667/800 is about £65/75/100 I would say that 667 looks like the best bet if budget is tight but if not go for the full 800 as isn't much money.

Hmm, it isn't the only vendor. Actually, at least somewhere, 533, 667 and 800 cost is very different. This is especially so for low-price brands.
I have the cheapest 1GB of DDR2-533 for $51, DDR2-667 for the same $51, DDR2-800 for $53.

For decent but functional brands like Samsung Original it's the same: 533 and 667 are precisely the same, 800 a little bit more. Might be because the prices go right from the vendor with standard margin, and 533 are just old stock. Seems that manufacturers shipped all three at the same cost, only 667 and 800 starting later.

[SixToes]

Thanks for all the feedback, it's really helpful. The Anand article was exactly what I was looking for and failing to find, and it's great to get feedback from SebRad and EV10, you guys seem to know what you're talking about.

I've had a look through eBuyer and these are the modules I'm considering - they are all Corsair, but only because they are the cheapest brand name on eBuyer. I've had problems with generic modules before so don't want to use them this time.

£79.99 - Corsair (VS2GBKIT667D2) 2 x 1 GB DDR2 667 MHz 2x128Mx64non-ECC 2x240 DIMM Unbuffered CL5
£86.99 - Corsair 2GB Kit (2x1GB) DDR2 PC2-5300/5400 (667/675MHz) (TWIN2X2048-5400C4) CL4-4-4-12 with Heat Spreader
£104.99 - Corsair (TWIN2X2048-6400) XMS2-6400 2048MB 5-5-5-12 2X240 DIMM Black XMS2 Heat Spreader

Based on feedback above the two more expensive sets will perform about the same so I may as well save £18 and go for 667 CL4 over the 800 CL5, is that correct?

Thanks again,
SixToes

[Arvo]

For [stable] AMD systems [at my work], built lately, I've choosen TWIN2X1024-5400C4 - best price/performance ratio IMHO (for Corsair brand, of course); thereby TWIN2X2048 of same spec chips should be enough usable for you. These chips are compatible with most of boards (running on normal voltage, unlike some overclockers RAM modules).

Of course if you want minimum power usage, you should underclock RAM too

Generic (or less-known brand) modules usually are running well enough either - I've not seen big problems with these for some years. I've mostly used various Apacer modules; sometimes on some boards they needed little relaxing of timings (CL2.5>3.0 for DDR400 on Abit AV8 for example), but nothing serious. On one local hardware forum people are using AData chips - the cheapest ones here in Estonia; no troubles heard so far (for about a year or so).

[SebRad]

Hi, the benchmarking I did can be found here.
(Viewed in "notepad" with Comic Sans MS font the alignment works out)
I did lot's of speed / timing tests with Dual Channel config and as it's rather time consuming only a few in single channel config. Have a look and draw your own conclusions where you want to be with speed vs timing vs cost but my tests showed 800 5,5,5 to be just slightly better than 667 4,4,4.
Seb

[EV10]

I've had a look through eBuyer and these are the modules I'm considering - they are all Corsair, but only because they are the cheapest brand name on eBuyer.

Check less cool brands like Samsung - it's where many best chips come from. Actually there are a few less known, but good brands.
Generic chips are bad only if either a) they are really bad (a one-day come-and-go brand), b) you mix them up. Brands' main advantage is that you'll be able to get compatible chips later.

Of course if you want minimum power usage, you should underclock RAM too

Well, RAM actually takes very little power. Heatsinks are put on high-cost modules just for looks. The most they can possibly need is a small aluminum plate covering the unit, and even that just for overvoltage overclocking above 1000MHz, and not necessary. Touch the chips in work - they're cold or at most a little warm.

[autoboy]

What is funny is that I have actually seen people over at tomshardware upgrade their power supplies because they want to stuff in more RAM. "I wanted to upgrade to 4GB but this 500W power supply was holding me back. I upgraded to a 700W unit and I was all set."

Pretty funny stuff. Way to upgrade that power supply for a 5W increase in power needs.

[jaganath]

I have actually seen people over at tomshardware upgrade their power supplies because they want to stuff in more RAM. "I wanted to upgrade to 4GB but this 500W power supply was holding me back. I upgraded to a 700W unit and I was all set."

OMG that is frickin' hilarious! (and also a bit disturbing, how can anyone be that retarded?)

[EV10]

Most people seem to forget the energy conservation law - that power consumed in a PC (most of it for AC) goes into heat and not somewhere.

It's OK not to know that DDR2 consumes less than DDR, and that it takes overvoltage and overclock to get just 5W out of a module, but I guess all overclocking enthusiasts have at least once touched working memory.

Of course, heatsinks do a lot to create an impression that it is hungry, but still. BTW, it's likely that these heatsinks actually increase memory operating temperature, by working as insulators. Pretty much like a bull barrel is hotter on the inside than a normal one under constant use, despite larger surface and mass - because heat transfer is slower and gradient appears.

And THG is one of these sites seemingly taking advertisement deeper than a banner.

[SixToes]

Thanks again guys.

On one local hardware forum people are using AData chips - the cheapest ones here in Estonia; no troubles heard so far (for about a year or so).

I've seen this memory at good prices in the UK, so maybe I'll give that a try.

my tests showed 800 5,5,5 to be just slightly better than 667 4,4,4.

Great stuff, exactly what I was after. Thanks for putting the results up.

Check less cool brands like Samsung

Heh - actually (along with Crucial) my first choice - but I couldn't find any good prices right now. I think I'll try AData or Kingston Value this time though.

Thanks again for all your help!