Need 2Gb memory kit to run at 229MHz

5t3v0

Senior member
Dec 22, 2005
508
0
0
The time has come to upgrade. It's a bit sooner than I thought but after discovering last night that the stuttering in FEAR was down to my memory usage exceeding the 1GB of physical RAM, it's now an urgent quest.

So guys, your mission is to recommend me something that will get me going on FEAR again without upsetting my financial consultant (i.e. wife). Bear in mind though that I live in the UK so I pay in £ what you lot pay in $ and that my currency costs 1.8 of yours.

My CPU is running at 280HTT which means on a 166 (5:6) divider I get 229MHz RAM frequency. Because of the limitations of the board, the next divider down is 133 (2:3) which gives a RAM frequency of 180MHz. Now that's a waste of bandwidth imo & I might as well OC PC2700 ram to get that. So my preference is to run the 166 divider and get 229MHz out of the RAM. Alternatively I could spend £££ and get a 1:1 ratio but I'm reading that's not worth expense.

So my options are:

1. Add another 2x512 GEIL Value sticks same as I aready have.

I'm reading this is not good as it's not possible to run at 1T. Is this that drastic? I only noticed a marginal drop (c. 20) in my 3dmark05 score when testing this on my current setup. A bigger concern to me is whether I could OC all 4 sticks to the same as 2. Anyone know if this is an issue on nf4?

2. Buy some cheaper 2Gb RAM with slower timings (e.g. 3-4-4-8).

I'm reading that timings have a bigger effect on performance on AMD64 than speed. Not sure how significant this is. I hardly saw any improvement (26 3Dmarks) in 3Dmark05 score when increasing cas from 2.5 to 2.0. What are good overclockable modules in this category that would reach 229MHz?

3. Buy some Highish Performance PC3200 with faster timings

Modules I've looked at in this category are:
Corsair XMS3200C2PT @ £165
G.Skill Extreme ZX @ £159 *

*I'm not convinced G.Skill will go as high as 229 on default timings after some posts I've read.

So guys, what's the answer? Please bear in mind that I am a bang-for-buck sort of geezer rather than an all out speed freak.

cheers
Steve
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
Here's an update with some quick & dirty benchmark results on my current memory. I ran 1 pass of 3DMark05 1024x768 NoAA (all tests selected) to see what difference it had on the scores. Tests ran were with:
1. my max memory setup @ 229MHz 2-3-3-6 1T
2. memory speed lowered to 180MHz (all other settings same)
3. command rate lowered to 2T (all other settings same)
4. timings relaxed to 3-4-4-8 (all other settings same)
5. timings relaxed to 3-4-4-8 and command rate lowered to 2T (speed kept at 229MHz)

All cpu & graphics settings remained the same throughout. This is not a statistically valid test I know, but illustrative none the less.

Test MHz Timings CR 3DMarks CPUmarks
1.......229....2-3-3-6...1T....6478.........5297
2.......180....2-3-3-6...1T....6453.........4869
3.......229....2-3-3-6...2T....6451.........4939
4.......229....3-4-4-8...1T....6453.........4975
5.......229....3-4-4-8...2T....6437.........4849

What the 3DMarks results suggest is that memory speed, command rate or timings have very little effect on graphics dependent game performance. When all 3 are maximised there is a increase of only 25 marks. Nothing ground breaking in those findings. More interesting though is the effect on cpu scores. Clearly when all 3 settings are maximised, performance is significantly better but after that what these results show is that the reduction in performance is caused by: speed > command rate > timings. This contradicts what a lot of people are saying about timings being more important than speed on AMD64.

Based on these results (and I'm not saying I necessarily should), I would opt for memory that can overclock on looser timings rather than expensive high performance RAM. I would even speculate that looser timings and a 2T command rate might be acceptable as long as the 229MHz speed could be met. I'll test this combination out tomorrow as its way too late now. Anyone know if I can get 229MHz on 4x512 of the GEIL Value series? I see that 2 sticks are only £59 on OCUK.

EDIT: Couldnt wait 'til tomorrow so just ran test 5. Results show slower timings and 2T command rate are only slightly worse than the slower speed with faster timings and 1T command rate. On a bang-for-buck system, timings and command rate could be safely sacrificed in favour of speed alone I reckon. Please feel free to disagree.
 

nealh

Diamond Member
Nov 21, 1999
7,078
1
0
it has been shown by several sources and you found the same thing..ram is not very important to a point..big $$$ on ram is wasted..Angry Games at DFI street(he is a mod) showed 1t vs 2t and memory speed has little performance gains..on AMD=CPU speed is king

I got 2 x 1gb of crucial valuram(not the binned ballistix that was around in summer for a while..I missed this) from newegg that has done very well

It does not like alot of voltage 2.8v anything more and it does worse

pc3200 that I got to 250mhz at 3-3-3-7-1t with my venice cpu...needs to set lower to work with my DC and maintain Prime stability

memtest 86 @250mhz was ok for 6hrs with no errors

I would think 230mhz should be a good bet on this ram but as always OC is YMMV
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
nealh, have you got a link for that DFI Street thread? Do you know if there are there any other issues with running 4x512 other than the 2T command rate? The cheapest option for me is to add 2 more sticks of what I currently have.
 

lopri

Elite Member
Jul 27, 2002
13,310
687
126
Just about any 1GB sticks rated PC3200 will do 230MHz. Of course timings will vary but then again as nealh says timings will make perfomance differences that no one will notice, unless your main app happens to be Sandra (or Everest).
And all these 1GB sticks use basically utilize 3 ICs: Micron, Infineon, Samsung UCCC. They do have different characteristics, yes, but in the end there is zero difference in real-life performance, unless, again, you happen to enjoy your games @800x600 screen resolution.

Go with the cheapest - but credible company's - sticks, is my recommendation.


 

5t3v0

Senior member
Dec 22, 2005
508
0
0
yeah, the cost concious approach always appeals to me. I'm not too worried about timings or command rate.

The dilema I have is whether to replace with 2 new modules of cheaper 1Gb ram and run at 1T but have lower timings, or to add 2 more 512Mb sticks of GEIL value and drop to 2T but have tighter timings? There's very little info about running AMD64 with 4 dimms in the forums, other than the 2T issue. Looks like not many people have tried it.
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
I found that thread on DFI Street, thanks. Good stuff & confirms what I was finding. I'm definately considering the upgrade to 4x512 as its the cheapest option for me. At least the tighter timings on the smaller modules, compared to value 1Gb modules, will compensate for the 2T command rate to some extent.

EDIT: While looking for GEIL Value prices on OCUK I noticed that the 2x1Gb kit was on offer at £117 & the 2x512Mb kit for £59. I'm now tempted to sell the 512s I already have on ebay & buy the 2x1Gb kit. Timings on 1Gb sticks are 3-4-4-8 but if its anything like the 512s I currently have there may be some improvement in that with a bit more voltage. Decisions, decisions. What would you guys do?
 

keldog7

Senior member
Dec 1, 2005
235
0
0
Glad you found the DFI thread - its a good read. I would hesitate to go with the 4x512 for the simple 2T issue. Some have reported performance hits of 25%...and as I recall, your upgrade was because FEAR was already stuttering.
I did *alot* of research into the Corsair 3500LL and 3200C2 ram - both will hit your 229MHz, and both should be able to do 2.5-3-3-7 1T timings at that speed without too much problem. But as you say, performace difference seen is likely to be small, so you might want to save your dough...can you get your hands on a friends RAM, and see how your rig performs with 4 sticks?
-A
 

ianmills

Member
Nov 19, 2002
50
1
61
There was another AWESOME post comparing different Memory speed and timings for A64 processors. It was much more in depth than just comparing 1t vs 2t. It also compared setting in different megahertz and other timings. It had benchmarks for at least 15 different programs (including many games).

Unfortunately, no matter how hard I search, I can't seem to find the post any more. But to the best of my memory, the poster's name was "Zim" and I think it was posted here at anandtech (or maybe at Tom's hardware? not sure...).

If anyone knows the post I'm talking about, it would be great if you could add the link here.

Ian

edit: that's why I couldn't find it, It's "Zebo", not Zim :)
http://forums.anandtech.com/messageview.cfm?catid=28&threadid=1475190

Very good info!
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
I saw that post by zebo. Reaffirms the argument for value over high performance although he didnt specifically test the impact of 2T. My results above suggest that moving to 2T has the same efect as droping the memory timings, although when both timings and command rate are sacrificed together the game score drops a bit more, but only by 16 marks. Unfortunately dropping command rate to 2T doesnt allow me to increase my cpu overclock at this already high vcore & I'm reluctant to put in any more volts. So going to 4 modules @ 2T would be acceptable to me providing I could maintain reasonable memory timings (2/2.5-3-3-6) & get 229MHz. Its a gamble though. There's no way of knowing if the next pair of RAMs will do as well as these or if 4 will do as well as 2 when all put together.

EDIT: Perhaps I should buy both the 1Gb pair & 2Gb pair & see which does best & then sell the surplus on ebay.
 

MrSpadge

Member
Sep 29, 2003
100
6
0
1. 3DMark03 and 05 are generally GPU bound, so while they may reflect performance in games, they can't show you what happens when you're CPU limited. Their CPU benches give some clues, however, doing the actual graphics rendering on the CPU makes it totally bandwidth limited, without much effect of timings / latency (command rate increases latency).
In this sense the guy from DFI street is right, when he says that 1T/2T doesn't matter much for games. Sure, with fast memory you can get a few percent higher performance. But when you're CPU-limited in a game, this may bring you up from 15fps to 15.5fps. Great, isn't it?

2. Going away from games to benches like Super Pi or Apps like the seti@home caching clients, things become different. Here a few percent do matter in the sense that you'd need 50 to 200MHz more CPU-MHz to reach the same performance, of course depending on the application. If you consider the efforts of some people to get additional CPU-MHz, then paying attention to fast RAM (while not spending a lot on it!) may well be worth it.

3. In your case I'd say you can easily live with 2T command rate, if it's games that you use your machine for. I can't help you much with the decision 4 x 512 vs 2 x 1024 though. But considering that you'll normally always loose some money while reselling, i'd rather go for the 2 additional modules. Overclocking with 4 modules is more stressful for the memory controller and won't get as high as with 2 of them, but you're only aiming for 229MHz - that *should* be fine.

4. The guy at DFI says that 2T is great to get higher CPU speeds. For me he looses most of his credibility in memory questions by saying that. Sure, 9 x 275 = 2475MHz @ 1T will be slower than 9 x 305 = 2745MHz @ 2T, but 2745MHz CPU and memory divider 10 / 183MHz (accessible via A64 tweaker) -> 275MHz mem @ 1T and tighter timings will be quite a bit faster than the 2T option.

5. Personally I'm upgrading from 2 x 512MB Crucial Ballistix PC400 @ 227MHz 2.5-2-2-1T to 2 x 1024MB MDT (great value at ~160? in Germany), which makes 2.5-3-3-1T at least up to 218MHz, hopefully up to 227Mhz as well. To bad my second set of Ballistix died, both after 2 - 3 months of 24/7 crunching at their specified voltage of 2.8V...

Hope this helps.
Regards, MrS
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
Thanks for the reply MrS. Games are certainly the most demanding resource intensive applications I tend to use so the difference between 1T & 2T will be slight if at all noticeable on this graphics card (I hope). I have had these modules running at 237MHz 1T on 3.0v with the default CL2.5 which was 8 hours memtest stable. I'm sure they will do more seeing as they like so much voltage but I have no idea where the limits is as I've had no need to try to max them out. So I'm hopefull the 229MHz threshold can be reached with 4 modules. If not then I'll have 4 of the b@stards to sell instead of 2 unless the new ones can be RMA'd!
 

MrSpadge

Member
Sep 29, 2003
100
6
0
If they don't reach the 229MHz you could alternatively use the divider for 150MHz with the A64 tweaker (-> 210MHz). This will work on your board, but you'd have to either do it after every boot or set it to automatically apply the settings.

MrS
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
Nice app. If its that easy why dont the mobo mfrs update their BIOSes with more dividers? The basic 4 (1:2, 2:3, 5:6, 1:1) that most supply are so limiting.

EDIT: Used the tweaker to set the 150 divider (mem speed = 210) & command rate to 2T & ran 3DMark05 again. I've added as test 6 to the previous tests for comparison:

Test MHz Timings CR 3DMarks CPUmarks
1.......229....2-3-3-6...1T....6478.........5297
2.......180....2-3-3-6...1T....6453.........4869
3.......229....2-3-3-6...2T....6451.........4939
4.......229....3-4-4-8...1T....6453.........4975
5.......229....3-4-4-8...2T....6437.........4849
6.......210....2-3-3-6...2T....6448.........4760

Shows a very small reduction (3 marks) in game performance from but the drop in cpumarks is massive. Bit worrying. Delving a bit deeper, both cpu tests dropped 0.1 fps from the same setup at 229MHz. Probably not a big deal in the grand scheme of things.
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
Originally posted by: keldog7
can you get your hands on a friends RAM, and see how your rig performs with 4 sticks?

This is probably a dumb question but do the modules have to be the same size in each dual memory pair? I just remembered I have a celeron box sitting here with 2x256Mb of PC2700. Could I run 2x512 in 1 pair & 2x256 in the other pair as a test?
 

Stoneburner

Diamond Member
May 29, 2003
3,491
0
76
So let's say i get a pair of mushkin redline sticks and run them at 290 mhz at fairly tight timings, it wont be a significant improvement over my pc3200 corsair memory which is running at a 3:2 divider?
 

keldog7

Senior member
Dec 1, 2005
235
0
0
i think you can put different sized pairs in the two banks...never tried it though. Also the timings will have to be manually adjusted to the slowest pair...PC2700 might not perform terribly well, but at the price you're getting them for ($0), you've nothing to lose but a few hours...
-A
 

MrSpadge

Member
Sep 29, 2003
100
6
0
Mhhh, guys, what about the general down-clocking with 4 modules? C-steppings could not offically do 4 Modules with 2T @ DDR400, E-stepping can do DDR400 officially with single sided modules. Too bad there's not much information on running 4 modules...
With 4 single sided modules it should be fine, but 4 double sided (intern organization) will be stressful for the memory controller.
Sell 2 x 512MB and buy 2 x 1GB the safer bet? Although I wouldn't choose GEIL for the 1GB. 3-4-4 @ DDR400 is pervert!

MrS
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
Originally posted by: Stoneburner
So let's say i get a pair of mushkin redline sticks and run them at 290 mhz at fairly tight timings, it wont be a significant improvement over my pc3200 corsair memory which is running at a 3:2 divider?

Not much in games although cpu bound apps might. The increased bandwidth from running 1:1 instead of 3:2 will have a noticable effect on performance I suspect, but the difference between tight timings & loose timings will be virtually unnoticable except in Everest or Sandra.

Originally posted by: keldog7
i think you can put different sized pairs in the two banks...never tried it though. Also the timings will have to be manually adjusted to the slowest pair...PC2700 might not perform terribly well, but at the price you're getting them for ($0), you've nothing to lose but a few hours...

I did mean for testing purposes only - I already have some PC2700 spare. I wasnt proposing to run at DDR333 permanently. Even if I do put all 4 together, it doesnt really help much because, as MrSpadge says, there's no guarantee that 4 modules will run @ DDR400, and I wont be able to test this without getting hold of 4 PC3200 dimms.

Originally posted by: MrSpadge
Mhhh, guys, what about the general down-clocking with 4 modules? C-steppings could not offically do 4 Modules with 2T @ DDR400, E-stepping can do DDR400 officially with single sided modules. Too bad there's not much information on running 4 modules...
With 4 single sided modules it should be fine, but 4 double sided (intern organization) will be stressful for the memory controller.
Sell 2 x 512MB and buy 2 x 1GB the safer bet? Although I wouldn't choose GEIL for the 1GB. 3-4-4 @ DDR400 is pervert!

I think you are right m8. The risk is too high. I dont want to be the one posting in a week's time groaning that I can only get DDR333 out of my 4 modules. So what do you suggest I get instead of GEIL? Bear in mind I only need to get 229MHz & dont want to spend a fortune.
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
OK. What about this 2Gb kit: G.Skill HZ PC4000 (F1-4000USU2-2GBHZ) which will cost £161 delivered? Its only 3-4-4-8 but it will mean I can run the 183 divider using the A64tweaker which will give a ram freq of 252MHz so well worth the low timings. Alternatives are the G.Skill PC3200 ZX with CL2 is £7 more expensive & the G.Skill PC3200 HS CL2.5 is the same price as the PC4000. I doubt the faster timings on these modules will compensate for the lower MHz and there's no guarantee they will overclock to 229MHz. So the choices are:

G.Skill HZ PC4000 @ £161
G.Skill ZX PC3200 @ £167
G.Skill HS PC3200 @ £161
 

MrSpadge

Member
Sep 29, 2003
100
6
0
I really like these MDTs for value, though I'm not sure whether you can buy them in the UK. I just tested the one I currently have in my system: 230MHz 2.5-3-3-1T @ auto voltage (should be 2.6V) gave me an error in S&M mem test after 2 passes, ~18min. Not bad i'd say, and ~2.7V may be enough to pass that test. Alternatively there should still be some headroom with looser timings. I don't want to push it though, because it's not my one :D
Too bad they seem to have increased in price over the last couple of weeks (160? -> 200?) and shipping from Germany to the UK is at least 30? - not really worth it, isn't it? 153£ for memory for which I can practically guarantee you 2.5-3-3-1T @ 210MHz, maybe even at 229MHz...

The GSkill look alright, though a bit expensive for my taste. OK, so since I need to choose a 2GB kit myself soon I did some research for proper information on value RAM, namely the ordinary Crucial @ 126£ for the 2GB (didn't even dare to ask for a review of the MDT). I'm sure this memory would have the same characteristics of the Ballistix (good timings and frequency headroom at 2.5-3-3), I couldn't find any proper information on that. I'd surely have to read forums for that, which takes ages *rolling eyes*.
The G.Skill HZ PC4000 might be an easy choice, or the GEIL 2 x 1GB which you mentioned before. You'd possibly have to run it at 210MHz 1T though.

... enough for today ;)
MrS

Edit: actually it took less than 5 min to find the right thread (never mind the German, there're enough numbers there :D )
http://www.forumdeluxx.de/forum/showthread.php?t=177162
- corsair value select seem to do it at 2.85V
- Crucial goes a bit higher, though I'd hesitate now to give them 2.8V
- GEIL value looks pretty bad
- G.Skill ZX seems really sweet, but only 2 entries
- Hynix orig: looks good, not sure you can get exactly that one
- Infineon orig: nice!
- MDT TSOP: maybe not as sure a shot as I thought, based on the 4 samples ;)
 

5t3v0

Senior member
Dec 22, 2005
508
0
0
Danke schon for that link. Interesting. I read a review about the GEIL value where they also said it was a poor overclocker ~210 max. Pity after the 512Mb sticks do so well. That offer for the GEIL Value 2Gb kit on OCUK has ended anyway, so they're back up to £130 for the pair which is £138 delivered. Not worth it. For £23 I can get the G.Skill. No contest. I've never seen MDT over here. As for Crucial, I've had their bog standard stuff in the past & its great for stock speeds but doesnt overclock much in my experience. I'm starting to think that if I'm going to spend £130 for a kit of ordinary ram, I might as well spend £160 for PC4000 or PC3200 with faster timings. Goes totally against what I started out saying but I want a bit of future proofing too. Fvck it, its only money.
 

MrSpadge

Member
Sep 29, 2003
100
6
0
I think I'll go with the G.Skill ZX. OK, so that's the price of 3 MDTs for 2 of them, but so what... it's only the same price as 2 x 512MB Ballistix 1 year ago! And I need to feed my X2 24/7 cruncher with data ;)
So either go with the G.Skill or take a shot at the ordinary Infineon or Samsung (UCCC, TCCC). Infineon isn't much cheaper though.

MrS
 

BOLt

Diamond Member
Dec 11, 2004
7,380
0
0
i've got 1gb dimms of corsair valueselect that can do 229MHz @ 1T. see the link in my sig. LMK!