Jump to content
Corsair Community

Looking for optimal memory on i875P using 4:5-Divider


Recommended Posts

Hello guys, well, I hope you don't think I'm coming around asking dumb question or kind of that, I just had a lot of problems finding "good" RAM for my machine. Since I'm running a Pentium4 2.4C at 283 MHz FSB on an Asus P4C800 deluxe - I guess it's obvious that I have to use the 4:5-divider, since there's no RAM on earth being capable of doing this 1:1. ;) I've been using some good Infineon PC333s, but they didn't do that very well. Then I came around with some Twinmos PC3200 (CH-5) which became a horror... they just "hate" the 4:5-divider, will mean, they do not run 200 MHz when the FSB is set to 250 MHz (4:5). So well, I finally decided to switch back (!) again to Corsair memory. I heard something about new revisions floating around that sort of "reject" fast timings. So, is there a special set of TwinX (1024 MB) RAMs that allow me to use [b]227[/b] MHz with good timings? I heard about the PC3700 / PC4000 that they won't run fast timings. Is there any solution for me? Thanks a lot in advance (and please don't worry about my English, I'm from Germany, eh? ;) )! BadFred.
Link to comment
Share on other sites

  • 2 weeks later...
If you want Corsair and tight timings and to overclock, I'd get the xms3500 sticks. However, the Intel 875 based boards perform GREAT with the PC4000 sticks. I have had mine at FSB267/DDR534 at 2.5,3,4,7 timings. It gives fantastic benchmarks at those settings and tests stable with Memtest. Mike.
Link to comment
Share on other sites

Ah,now that you mentioned it, specmike, could you give us some Sandra results ? Preferably the unbuffered memory benchmark (with MMX,enhanced MMX,SSE,SSE2,buffering boxes unchecked, 9 boxes unchecked in total) . Thanks :)
Link to comment
Share on other sites

[QUOTE][i]Originally posted by Therion [/i] [B]Ah,now that you mentioned it, specmike, could you give us some Sandra results ? Preferably the unbuffered memory benchmark (with MMX,enhanced MMX,SSE,SSE2,buffering boxes unchecked, 9 boxes unchecked in total) . Thanks :) [/B][/QUOTE] Sure, here are 4. I am not going to push any harder than this today. I tried a FSB270/DDR540 w/PAM+Turbo. I had Sandra, CPU-Z, and Photoshop all open at the same time capturing the image. Photoshop crashed and took Windows with it:( Since this is our editing=money making puter too, I can't afford to do that too often. I did break about 3550/3620 the other day, the JPEG is on another forum though. 1st run is box stock as far as the RAM, SPD@250/500. Vcore is 1.55v and Vdimm is 2.75v. No turbo or PAM. Mike
Link to comment
Share on other sites

[QUOTE][i]Originally posted by Therion [/i] [B]Did you run the Sandra benchie with all the boxes unchecked ? (the ones i mentioned b4 ) ? The results sure look nice ! [/B][/QUOTE] Yes, one way to tell if someone is fudging on Sandra as to being buff or unbuff is to look at the entries on the red and blue lines. It will either say Int ALU Buff 3429 or Float FPU BUff 3425 if they are allowing the buffering prefetches to be used. If it does not have Buff on the line, it is off. I assume you mean the 9 boxes in this pic?
Link to comment
Share on other sites

Yes specmike, those are the ones,thanks. I asked because the buff/unbuff thingy in the sandra red and blue stripes, also leaves if one unchecks JUST the buffering checkbox if i recall. thanks for the confirmation again :)
Link to comment
Share on other sites

I would say Definatly use XMS 3500, Check my sig file... I have seen people mention this on various forums, and have verified it in my own testing, pretty much at any FSB, 5:4 ram divider at 2-2-2, beats 1:1 at 2.5-3-3, and totally blows away 1:1 at 3-4-4 (not necesaily just mem benchies, but actual game benchies) The newer 3700 and 4000 cant seem to run low latency at any speed. long story short... you are better of with 5:4 at low latency, then getting the latest pc4000 at high latency with a 1:1 div. :D
Link to comment
Share on other sites

[QUOTE][i]Originally posted by retrospooty [/i] [B]I would say Definatly use XMS 3500, Check my sig file... I have seen people mention this on various forums, and have verified it in my own testing, pretty much at any FSB, 5:4 ram divider at 2-2-2, beats 1:1 at 2.5-3-3, and totally blows away 1:1 at 3-4-4 (not necesaily just mem benchies, but actual game benchies) The newer 3700 and 4000 cant seem to run low latency at any speed. long story short... you are better of with 5:4 at low latency, then getting the latest pc4000 at high latency with a 1:1 div. :D [/B][/QUOTE] Any FSB+RAM performance at 1:1 will outperform the same FSB+RAM at 5:4. Think about it, FSB200/DDR400 at 1:1 is better than FSB200/DDR333 at 5:4. That's WAY more RAM bandwidth. PC4000 on a Canterwood or Springdale board doesn't need low latencies. Benchmarks and actual real world testing that I have done and seen others do verifies this. The 5:4 divider is made to compensate for lack of ability of the RAM to clock higher . Would you care to compare Sandra bandwidth scores? As far as game benches clock speed is usually the single greatest factor. This is especially true in 3D games/benches. But memory is important also So, why would you want to run 5:4 when you can run 1:1? If you can hit the same FSB at 1:1 as you can at 5:4 (I can) then using 5:4 only limits your memory bandwidth. You'd have to OC approximately 20% higher at 5:4 to overcome the advantages of 1:1. I don't think there are many 2.4c CPUs that will OC to FSB300+@5:4 which is what it would take to exceed what my system is giving. So, what benchmarks are you using/seeing that tells you that 1. PC4000 is not as good as PC3500? 2. 5:4 is outperforming 1:1? Mike
Link to comment
Share on other sites

1:1 outperforms 5:4 at the same latency settings for sure... but 5:4 at 2-2-2 will outperform 1:1 at 3-4-4 any day... I didn't believe it myself until I tested it. What I am saying is that the newest pc4000 rams from the various vendors are all rated at 3-4-4, and dont run low latencies very well at any speed. Given that fact, in most situations (depending on your setup and max FSB that your mobo/cpu can achieve) pc3500 at 2-2-2 will give better performance than pc4000 at 3-4-4 Try this... (assuming you have some ram that will run at 2-2-2) Put your ram at 1:1 3-4-4 and the set it at highest FSB you can run stable at (for me its 285, but I can repeat the same results at 220) Try several tests like SSandra unbuffered and 3dm2001 and one of your favorite games (and any other test you would like) , and write down the results. Then , at the same FSB, change the divider to 5:4 and the latency to 2-2-2, and run the same tests. This way the FSB and CPU remain at the same speed and the only change between tests is the ram freq. and latency. Post back your results . :D I am at work now (hardly working) I will post some SSandra benchies later
Link to comment
Share on other sites

I have several JPEGs posted in this thread where my scores at 250FSB equal or exceed yours at 280. My CPU won't even do 280 yet my RAM makes up the difference. You have the awesome Winbond BH5 chips on your XMS3500 but even those are being slowly but surely left behind. Also, it's RARE to find any 875/865 board that will run 2,2,2, timings. The RAS to CAS delay is usually limited to 3 regardless of MOBO brand and the average person cannot run a 2 there. I'd say you have an exceptional MOBO AND CPU there. But, you cannot keep up with the mighty PC4000. Here's a link and 2 more JPEGs for you. PC4000 at less FSB is better than 5:4. Nice rig though, you've definitely got a screamer. Mike.
Link to comment
Share on other sites

[URL=http://www.abxzone.com/forums/showthread.php?s=&threadid=52042]Here's the LINK.[/URL] And here is the back breaker, my finishing move if this was a pro-wrestling match. :p Mike PS: Buffered at the same settings was over 6400/6400 but I don't have the pic anymore. I don't have PC Mark but I guess I can dig it up if I need to. Although, if it brings the AGP card into it, you'll kill me as I use a 32mb dual head WS card.
Link to comment
Share on other sites

Nice scores... However, I am pretty sure the Asus board you use is simply faster mhz per mhz and you are doing better then 3-3-4 latencies, alot of peeps cant. Also worth noting, Sandra is not really a great measure of memory, especially when comparing 2 different systems. When you get time, try what I suggested earlier ... and run some other tests, like your favorite game, and some 3dmark2001, and compare the results. -------------------------------------------------------- (assuming you have some ram that will run at 2-2-2) Put your ram at 1:1 3-4-4 and the set it at highest FSB you can run stable at (for me its 285, but I can repeat the same results at 220) Try several tests like SSandra unbuffered and 3dm2001 and one of your favorite games (and any other test you would like) , and write down the results. Then , at the same FSB, change the divider to 5:4 and the latency to 2-2-2, and run the same tests. This way the FSB and CPU remain at the same speed and the only change between tests is the ram freq. and latency. -----------------------------------------------------------
Link to comment
Share on other sites

I don't have any games and the computer in question is an editing computer. The 3DMark scores or any scores that include AGP performance are useless here as my AGP card cannot come close to what yours can. And, a 3D based benchmark is going to be based primarily on the card capabilities and not as much on the RAM. It will benefit from a higher CPU frequency which will mean more FPS. So, back to my original point, if you can run the same FSB at 1:1 as you can at 5:4, why throttle the RAM down? Can you imagine if you could do that 280FSB you get with some PC4000 at 1:1? That would be DDR560! It's really up to the original poster here. If they want to run a 5:4, the PC3500 is definitely the best because of the latencies for sure. I just don't know why'd anyone would pick 5:4 to begin with. I also wanted to dispel the wrong idea that 5:4 (asynchronus operation) is better than 1:1 (synchronus). That holds true on any computer, any platform, any RAM, anywhere, any time. Running asynch in and of itself introduces latencies into the system as it has to compensate and get the memory back into synch with the CPU at some point in the NB and this slows things down. Also, where have you heard that Sandra is not good for comparing different systems? That's the point of any benchmarking tool. And, are you saying that a person would be better off buying an Asus board instead of Abit since the Asus is faster cycle per cycle? If so, why'd you buy Abit? And, how many people have you seen that cannot tighten the latencies with the PC4000? I'd like to see those posts. Most are not able to tighten the PC3700 timings (Samsung chips) but it will still clock very well. The PC4000 is a Hynix chip and it is VERY FAST! Mike.
Link to comment
Share on other sites

OK, 3dmark is testing the V-Card as well, but I wasnt suggesting you compare to my 9800, I was suggesting you compare to yourself at 1:1 3-4-4-8 vs yourself at 5:4 2-2-2-5 in more than just Sandra memory test. Sandra is only one test, and not a very good one at that. 3dmark is heavily affected by CPU and Memory speed, this is why its a good test to run (the CPU and Vcard are the same, only your mem speed aand timings are changed. In my previous tests with this motherboard going from 1:1 to 5:4 dropped about 1-2 % in performance on various benchmarks (3dm2k1,3dm2k3,Q3,UT2k3,Sandra,and pcmark and sciencemark) Going from 2-2-2-5 to 3-4-4-8 dropped 5-7% in those same tests. REad in 5-7% drop is more thatn 1-2... In short, at ANY FSB, 5:4 at 2-2-2-5 is faster than 1:1 at 3-4-4-8, in most cases its faster than 1:1 at 2.5-3-3. Unless, the possibility exists that your mobo does not do 5:4 properly. As far as why I bought the Abit board... well, I wasn't aware of the info at the time :D Here is a good thread to read through.... [url]http://www.xtremesystems.org/forums/showthread.php?s=&threadid=16191[/url] At the time I was thinking what you thought, until I actually tried what Zroc asked me to try on the 52nd post. He was right.
Link to comment
Share on other sites

[QUOTE]In short, at ANY FSB, 5:4 at 2-2-2-5 is faster than 1:1 at 3-4-4-8, in most cases its faster than 1:1 at 2.5-3-3. Unless, the possibility exists that your mobo does not do 5:4 properly. [/QUOTE] Last point for me and I'll agree that we disagree. Are you saying this is at the same FSB for each ratio or any FSB. For example, are you saying that your system at 280 using 5:4 is giving greater memory bandwidth than mine at 250 using 1:1? Or did you mean equal FSB on each? I still don't understand the logic and it bears out in every test I have seen and I will do more but...... FSB200/DDR400 (1:1) FSB200/DDR320 (5:4) or FSB300/DDR600 (1:1) FSB300/DDR480 (5:4).....:confused: The CPU speed is the same and the memory bandwidth drops by 80mhz or 120mhz respectively. How is that better? Mike
Link to comment
Share on other sites

OK, I am not saying we can compare my 280 5:4 to your 250... the most fair test is to test your system, against your system at both settings. We know you can easily run at 250mhz right ? So try to run several tests (not just Sandra) at 250 1:1 3-4-4-8 , then run them again at 250 5:4 2-2-2-5 and compare yourself to yourself without changing anything at all except the Ram ratio, and latency. What I am saying is 250mhz @ 5:4 2-2-2-5 is faster than 250@1:1 3-4-4-8 in real applications, not just looking at Sandra's memory test. So... in essence 250@ 5:4 is 200, so, 200mhz ram, at 2-2-2-5 is faster than 250 at 3-4-4-8 in almost every single application, game or benchmark. I know that takes a bit of effort to go through, so if you don't wanna try it its cool, we can agree to disagree. My main point is the higher latency is more of a penalty than most people realize. :D
Link to comment
Share on other sites

I will do some testing with what I have but I have no RAM that will run 2,2,2 timings. My PC3200C2 will do 2,3,3,6 on my P4C800 but it is nowhere close to my PC4000. [QUOTE]So... in essence 250@ 5:4 is 200, so, 200mhz ram, at 2-2-2-5 is faster than 250 at 3-4-4-8 in almost every single application, game or benchmark[/QUOTE] Show me one, I have disproven this in Sandra, the most applicable test, since I did a memory test albeit 2 different MOBOs, yours vs mine. I will find you some more links, that use other benches than Sandra. :sillygrin Mike
Link to comment
Share on other sites

Archived

This topic is now archived and is closed to further replies.

×
×
  • Create New...