Jump to content
Corsair Community

P4c CPUs with Corsair PC3700&PC4000 RAM


Recommended Posts

This may get a bit lengthy or boring :zzzz: but I did not force you to click this link;) With the release of the Corsair PC3700 and PC4000 DIMMs, many are asking about proper combinations of RAM+CPU. Normally you would think that the fastest RAM going and the fastest CPU going would be the BEST combination, cost not withstanding. But, it seems that this is not the case. To fully utilize PC3700 RAM on current P4c CPUs and on the current Intel 875/865 chipsets, you must run the RAM at a 1:1 ratio according to the benchmarks I have seen. Some debate the validity of synthetic benchmarks and say they prove nothing. If that is your belief, then don't buy any Corsair RAM above DDR400. The 5:4 ratio is not as good as the 1:1 as far as overall memory bandwidth which is why we buy good fast RAM. Back to the point, the FSB for PC3700 (DDR466) at a 1:1 ratio is FSB233 and FSB250 for PC4000 (DDR500). This means that you have to be able to overclock (OC) your CPU to run the RAM at its stock speed. To OC the RAM past stock, you REALLY have to OC your CPU. Well Hawk, a frequent poster at 2 other forums I visit who is a reviewer [URL=www.bleedinedge.com]HERE[/URL] has tested a retail sample of all 5 issues of the P4c CPUs. He has found the following: P4 CPU Speed Stock MAX FSB 2.4c 288FSB 2.6c 270FSB 2.8c 266FSB 3.0c 247FSB 3.2c 238FSB His results are exceptional and he knows how to get speed out of a rig. One of his nic names is "Overclocking Junkie" and it fits. The reason I am posting all of this is to show that you can't necessarily utilize the DDR466 and DDR500 RAM with EVERY CPU available. So, for any of you that are considering a new rig with Corsair RAM, consider your CPU choices. Hawk's results are not etched in stone however so buy whatever CPU floats yer boat. For example, the Goog stated recently that he has run 2 3.0s. One would go over 250FSB, one would not so it is a hit and miss. It is always hit and miss with OCing but so far, the 2.4c CPUs are the most consistant at hitting high FSB. On the other end of the spectrum, if you want the MAXIMUM performance at box stock speeds, I'd say go with the XMS3200LL TwinX kits+3.2ghz CPU. Then, you are running fantastically high and don't have to worry with any OCing. I'd add the XMS3500 stix in also but it seems that they may be getting hard to find. With either of these stix, you can run a stock FSB and RAM at DDR400 with VERY tight timings. Now, as to timings, I can attest that with my results+posts I've seen by Hawk et al, the PC3700 and PC4000 stix do NOT need tight timings to have AWESOME results. The stock PC3200 is far better on a dual channel MOBO than any previous single channel MOBO but the PC3700 and PC4000 with more loose timings are EVEN BETTER. I attached a JPEG of my last run at FSB270/DDR540 at 2.5,4,4,7,8 on a P4C800 Deluxe NON-E. The voltages were bumped a bit but it just shows the true potential of the PC4000. FORGET about tight timings unless you are running an AMD CPU. So here are my rules of thumb for buying an 875/865 based board and CPU+RAM. If you can't hit a 233FSB, you will be underclocking the PC3700 and PC4000. If you can't hit a 250FSB you will be underlocking the PC4000. 1. If you don't want to OC and want to run stock CPU speed with TIGHT RAM timings buy the TwinX 3200LL or XMS3500 and the FASTEST CPU you can afford. 2. If you want a MILD OC with TIGHT timings, buy the XMS3500 and the FASTEST CPU you can afford. 3. If you want a MEDIUM OC your BEST buy would be the PC3700 with the [U]FASTEST CPU YOU BELIEVE WILL HIT A 233 FSB.[/U] I am on the fence at this point with a 2.8c, will they consitantly hit a FSB233? Buy one and let me know 4. If you want a HOT OC and want to go all out, buy the PC4000 and either a 2.4c or 2.6c. My choice is the w2.4c. It's not as "slow" as it sounds and you can spend the money you save on the lovely Corsair PC4000 :D Anyways, a 2.4c at FSB 250 is 3.0ghz, how much faster do you wanna go? About the most anyone will do without watercooling, which Corsair has, [URL=http://www.corsairmicro.com/main/HydroDATASHEET.pdf]HYDROCOOL[/URL] , is about 3.6ghz due to architectural limitations. This is still RARE and again, hit and miss. Do you want to play hit and miss with a $700 3.2ghz CPU? A 3.2 at 3.6 is only a FSB222 which is hard to get on a 3.2c. I have seen several people post that they cannot hit 220 on a 3.2c. The 3.6 is about the average cap on the P4 Northwood architecture so OCing it up to or past 3.6 is tough. The headroom of a 2.4c is its strength. All P4c CPUs are from the same die, and are just speed binned. The "best" chips become 3.2s and the "worst" become 2.4s. I've had my 2.4c at 3.0 on stock voltage so that tells you how "bad" a 2.4c is. IMO, save the money on a CPU and buy more RAM! In summary, here is my dream system. 2.4c CPU that will do 288FSB:D Asus P4C800 Deluxe E MOBO (THE best OCing 875 board out, PERIOD!) Radeon 9800 Pro 256mb AGP card. (2) WD Raptor HDs on a RAID 0 stripe on the ICH5R controller Antec 1080 Case with Antec True Power 550 PS, no lights or gadgets needed. Corsair HYDROCOOL Water Cooling System You can fill in the rest, this is the core of it.
Link to comment
Share on other sites

Very interesting... But you should have a look at that, even if it is in french... [url]http://www.hardware.fr/articles/473/page3.html[/url] The last chart says it all... it's better to have tight timings at a lower frequency (using the 5:4 ratio) than high frequency and lower timings, in that test at least...
Link to comment
Share on other sites

[QUOTE][i]Originally posted by BangoO [/i] [B]Very interesting... But you should have a look at that, even if it is in french... [url]http://www.hardware.fr/articles/473/page3.html[/url] The last chart says it all... it's better to have tight timings at a lower frequency (using the 5:4 ratio) than high frequency and lower timings, in that test at least :wink: [/B][/QUOTE] If I interpret that correctly, the highest they went is DDR490. They also used Quake 3 which introduces AGP performance to the mix. They did not do any memory specific tests like Sandra or Aida32. With a good OC, I am reaching FSB267/DDR534 easily and stable. I have done some benches with Sandra at 270/540. Also, if the GOAL is tight timings, they should have used XMS3500 instead of PC3700. The PC 3700 is not designed to run tight timings. Their results seem to lean more towards what I was saying about DDR400 and DDR433 RAM. To each his own but the ONLY memory benchmarks I have seen that beat my own are from another manufacturers BETA stix of PC4000 and a couple of volt modded boards which don't really apply to what I said. I'll go head to head with any other retail RAM on the planet, loose timings and all! Benchmarks may not really reflect TRUE real-world performance gains (some say yes, others say no) but if they are evenly applied across the comparative products, they can help guage performance and that's what many people base their purchases on. My purpose here is to help mate the right RAM with the right CPU. I mean, a person that buys PC4000 and a 3.2ghz CPU has too much money(;) just kiddin) and they will NOT get to the top end of that RAM. On the other hand, a person with a 2.4c will NEVER run PC3200LL RAM at 270mhz FSB at a 1:1 ratio. There are a great deal of questions here along this line and I am attracting them because I have the PC4000. It was easier to type this thread as a general reply. Mike.
Link to comment
Share on other sites

Their goal was not timings. In fact it was the test of the Corsair 3700. So they decided to compare it to the 3200LL with a 5:4 ratio. I understood the point of your thread which is, as I said, very interesting ! But... the question is... is it really important to stay with a 1:1 ratio ? According to the benchmark scores I saw with PC4000 memory I would say yes... with the PC3700, I would say... hum... maybe not :sigh!: But for sure if you're planning on overclocking like a mad man, then a 2.4C and PC4000 memory is the best deal :greengrin
Link to comment
Share on other sites

[QUOTE][i]Originally posted by BangoO [/i] [B]Their goal was not timings. In fact it was the test of the Corsair 3700. So they decided to compare it to the 3200LL with a 5:4 ratio. [/B][/QUOTE] Do you read French or use and interpreter? I could not tell for sure but it seemed as if they ran the 3200LL at 5:4 but ran the PC3700 at 1:1, is this correct? If so it was a fair test. If not, then the PC3700 may do better at 1:1 because of the Intel PAT. Mebbe not, I am open to owners input, I have no PC3700. The PC3700 may be the equivalent of a 2.8c CPU, sort of sitting on the fence, dead in the middle of the debate. The real choices may be either the 3200LL on one end of the spectrum and PC4000 on the other and each with opposite CPU requirements. I should clarify that my opinion relies on my belief that PAT is disabled (mebbe totally, mebbe partially) at a 5:4 ratio. Or, mebbe the 1 or 1/2 clock cycle delay with asynchronus operation defeats the benefits of PAT. I am basing this on my use of the 1:1 ratio and my gear and observations of benchmarks. Pass on any good info you find that I can read :eyebuldge Mike
Link to comment
Share on other sites

Nice post specmike ! Its a pity most 865/875 boards have a max mem voltage of 2.8/2.85 :grrr: I can run with a 280 FSB.Do you think its worth it , for me , to change memory ? My max v.mem. is 2.8 and i have 2x256MB XMS 3500C2 v1.1. I generally dont see people satisfied with the 3700+4000 memory modules (of any brand). Also i was thinking of doing the voltage mod (v.mem to 3 or 3.1v).Which of the 2 types will this help the most ? The 3500 i already have, or the 4000 ? thanks
Link to comment
Share on other sites

[QUOTE][i]Originally posted by Therion [/i] [B]Nice post specmike ! Its a pity most 865/875 boards have a max mem voltage of 2.8/2.85 :grrr: I can run with a 280 FSB.Do you think its worth it , for me , to change memory ? My max v.mem. is 2.8 and i have 2x256MB XMS 3500C2 v1.1. I generally dont see people satisfied with the 3700+4000 memory modules (of any brand). Also i was thinking of doing the voltage mod (v.mem to 3 or 3.1v).Which of the 2 types will this help the most ? The 3500 i already have, or the 4000 ? thanks [/B][/QUOTE] Thanx. A 280FSB @ 1:1 with RAM that will do it would be AWESOME. However, since you have the "legendary" Winbond BH5 chips in your PC3500, that may be the best RAM on the planet to try a volt mod with. However, with the Hynix 43B chips in the 4000, I have no idea how they stack up next to the Winbond chips. As to the voltage handling characteristics, you should probably ask RAM GUY with an email or a PM. You SERIOUSLY DOUBT he's going to answer that with a public response on this forum. OR, surf around and ask another volt modder. You may want to see if anyone here will own up to running their RAM on a volt modded board and get a more educated opinion. As a responsible adult participating in the Corsair HoH forum, I can't condone volt modding or board modding in any way, you know the warranty RAMifications if you do;) BUT, I expect that you will get unearthly results if you mod and use either PC3500 or PC4000, so let us know how it works out. Just out of curiosity, where are you seeing discontentment with the PC4000 users. I'm as happy as a puppy with 2.....well....nevermind that, I'm just happy:D Are they using Corsair PC4000 or other brands and what are they saying is wrong? Too much money is all I can think of and the old saying holds true," You get what you pay for". Keep in touch, Mike.
Link to comment
Share on other sites

  • Corsair Employees
That is one of the reasons we chose Hynix they do not have a voltage regulator so they will take the over volt direct to the Ram. But I must make it clear we do not suggest that any one over volt there system or the memory Beyond 2.8 Volts. But we have to look forward here. And I think and the trend will be more relaxed timings with higher Freq. at least that’s what’s coming from all of the IC makers. Now I do not know if that will change as the IC makers have new die's, who knows. We will have to wait and see. Ram Guy
Link to comment
Share on other sites

Therion, I have also...2x256MB XMS 3500C2 v1.1. So I am a lucky guy than ?? I use 5:4 at the moment, but I am stuck on 270FSB with 2-2-3-6-8. As soon I use 271, I get errors in Prime even with looser timings or 266 in Bios. I can't be my Memory whats holding me back. I have the P4P800 Deluxe and P4 2.4C. Is the 270 the famous wall than ?? (P4 2.4C) hansje;) Using: 1.6 Vcore and 2.75 Dimm.
Link to comment
Share on other sites

[QUOTE][i]Originally posted by specmike [/i] [B]Do you read French or use and interpreter? I could not tell for sure but it seemed as if they ran the 3200LL at 5:4 but ran the PC3700 at 1:1, is this correct? If so it was a fair test. [/B][/QUOTE] I [B]am[/B] french :greengrin They tested the 3200LL with a 5:4 ratio and the 3700 with a 1:1 ratio.
Link to comment
Share on other sites

By the way, I don't know about other mainboards but on the MSI Neo2 865 FIS2R, you can set the FSB:memory ratio 4:5 and 3:4. That means your memory will go faster than you FSB... anyone tried that ? Cause then it means you could use PC4000 at its specs even with a P4 3.2C, the FSB being 400Mhz and the memory 500Mhz.
Link to comment
Share on other sites

Specmike and Ramguy, I'm currently running the P4P800 Deluxe with Corsair 2x512mb 3500cms. Now, I don't plan on OC'ing the system... as I *really* want to see what "stock" can do Any suggestion on timings to get uber performance from the 3500's? Side note - mine have the blue heat spreaders - they look friggin KEWL with the blue heat spreaders :sunglasse
Link to comment
Share on other sites

Wolf, You do not want to overclock but you do want to set timings below the advertised settings? That sounds like overclocking to me. [url]http://www.hardware.fr/articles/473/page3.html[/url] If I understand these Quake 3 scores correctly, running PC 3500 CAS 2 memory will be faster than PC3700 cas 3 with the CPU @ the same speed, right? I would think that the higher FSB would compensate or why else bother building PC3700 and 4000? Futhermore, do I understand correctly when you run 5:4 instead of 1:1 PAT gets disabled? This because I`m looking for memory to accompany my Asus P4P800 and Watercooled PIV 2.8. So I should get TWINX XMS 3500 CAS 2 Instead of TWIX 3700 CAS 3?
Link to comment
Share on other sites

[QUOTE][i]Originally posted by WOLF359 [/i] [B]Specmike and Ramguy, I'm currently running the P4P800 Deluxe with Corsair 2x512mb 3500cms. Now, I don't plan on OC'ing the system... as I *really* want to see what "stock" can do Any suggestion on timings to get uber performance from the 3500's? Side note - mine have the blue heat spreaders - they look friggin KEWL with the blue heat spreaders :sunglasse [/B][/QUOTE] In the order that the AMI bios places the RAM timings try for SUPER LOW timings 2,2,3,5,8 at first with NO turbo, NO MAM, and Vdimm at 2.75v. Those are the lowest timings I have seen anyone run with only about 3-5 RARE exceptions of 2,2,2,5,8. Be sure to backup your data or make a ghost image of your system drive. I recently had an OCing incident and corrupted Windows. Uber RAMster RAM GUY suggests running a Memtest run before booting with new OCing settings. It is slow but not as slow as reinstalling an OS+Apps+Tweaks on a DV editing machine! Let us know how those timings work out. Just FYI, with great RAM like that, you just gotta try a wee little OC:D Happy Computing, Mike
Link to comment
Share on other sites

Cool, will give it a shot over the weekend and let ya know what it does... I read that post on the dual bios hack in the compatibility section. /me to big of a chicken to try it at this point, guess I'm optimistic that ASUS will post a bios update one of these days that fully enables all the goodies on this board (but what do I know, I'm still waiting for Ozzy to win his life time achievement Grammy)
Link to comment
Share on other sites

Great post! This ought to be a "sticky". I lurked here a while before buying and came to the same conclusion (this was before the 4000's came out though). After some thought I decided I was a #2 overclocker, so I picked the 3500C2+3.0c (the 3.2's price was just silly). I am really happy with it running 220fsb 1:1 at 2-2-3-5-8. (I have tested it at those timings at 222, maybe I'll give 223 a go and see). I am still unsure what the deal is with "Turbo" mode, it doesnt seem to make a significant difference in speed for me nor does it seem to affect stability at the speed/timings I'm running. For the record though, I am seeing a (very) slight decrease in some of the CPU benchmarks with it ON, and few more of the benchmarks (mostly FP related for some reason) were (very) slightly higher. I've left it on for now, to give me something to try disabling that doesnt hurt performance should I ever have a future stability problem ;-) Specmike: "about 3-5 RARE exceptions of 2,2,2,5,8" I got 2-2-2-5-8 at 200fsb with the 2x512 XMS-3500C2's last weekend when I was checking things out for that guy wanting someone to "confirm" they'd run at 2-2-2-6. I just didnt bother to report that result to him for fear of more questions ;-)
Link to comment
Share on other sites

[QUOTE][i]Originally posted by Zakamoto [/i] [B]Great post! This ought to be a "sticky". I lurk here a while before buying and came to the same conclusion (this was before the 4000's came out though). After some thought I decided I was a #2 overclocker, so I picked the 3500C2+3.0c (the 3.2's price was just silly). I am really happy with it running 220fsb 1:1 at 2-2-3-5-8. (I have tested it at those timings at 222, maybe I'll give 223 a go and see). I am still unsure what the deal is with "Turbo" mode, it doesnt seem to make a significant difference in speed for me nor does it seem to affect stability at the speed/timings I'm running. For the record though, I am seeing a (very) slight decrease in some of the CPU benchmarks with it ON, and few more of the benchmarks (mostly FP related for some reason) were (very) slightly higher. I've left it on for now, to give me something to try disabling that doesnt hurt performance should I ever have a stability problem ;-) Specmike: "about 3-5 RARE exceptions of 2,2,2,5,8" I got 2-2-2-5-8 at 200fsb with the 2x512 XMS-3500C2's last weekend when I was checking things out for that guy wanting someone to "confirm" they'd run at 2-2-2-6. I just didnt bother to report that result to him for fear of more questions ;-) [/B][/QUOTE] LOL....that's funny and sometimes I do the same. A given answer results in 20 more questions when a simple yes or no would suffice. Of course, I am still learning and prone to turn an answer into about 20 more questions myself. Those are NICE timings though. And, I know fer sure that 2 others with P4P800s have gotten those timings. Thanx for the input. Mike
Link to comment
Share on other sites

OK, I got all type-happy again so here's my take on buying a new P4 system. Again, it's long but I hope it helps. [QUOTE][i]Originally posted by CrackFerret [/i] [B]Mike, care to explain memory timings or should a newb such as myself on worry about messing with the FSB and FSB ratio for now? [/B][/QUOTE] I'll give it a shot sooner or later but there are better here and other places than me to do it. [URL=http://www.rojakpot.com/default.aspx?location=1]Adrian Wong does a great job here[/URL] Essentially, ANY latency inside a computer means some action or process is waiting for, or dependent upon, another action or process to finish before certain instructions can be completed (I think). Ideally, a computer would have no latency. RAM timings ideally are lowered (smaller numbers on RAM timings = less latency) to their lowest points for fastest performance. Some RAM is better than others at running with low latencies. Some systems are more dependent upon low latencies for best performance. For example, the current AMD systems gain way more performance from low latencies that the current Intel 875/865. These Intel boards that I am speaking of seem to be getting BEST performance from pure high FSB speed + high RAM frequencies. But, physics limit us (I don't know enough specifics to splain it clearly Lucy) and the RAM companies cannot yet get the low latencies we are used to in the PC3700 and PC4000 DIMMs. However, as the results will show, memory bandwidth and performance is still increasing with the PC3700 and PC4000 RAM. That's why I sort of drew the line there at the PC3200/PC3500 vs PC3700/PC4000. IMO, you can use that to select the RAM that best suits your wants/needs and match it logically with the right CPU. To illustrate my point, I got some prices together from Googlegear.com and Newegg.com. Intel P4c CPUs 3.2c..........$655 3.0c..........$393 2.8c..........$266 2.6c..........$211 2.4c..........$170 RAM Prices Frequency...............Corsair Memory PC4000 2 x 512mb...$380 PC3700 2 X 512mb...$315 PC3500 2 x 512mb...$363 PC3200LL 2 x 512mb...$312 Let's say now that you want the absolute best performance to be had on an Intel D875PBZ Canterwood board. Intel boards are noted to be very stable, very fast at stock speeds, but will not OC well at all, 4% at most on that particular model. My pick would then be a 3.2ghz CPU with the Corsair3200LL 1024mb kit. Corsair+CPU=$967( @ 2,3,2,6 timings) If you bought PC4000 RAM, it will NOT run the timings tight, about 2.5, 4,3,5 at best based on my tests of the Corsair. Corsair 3200LL at DDR400 will outperform Corsair PC4000 being run at DDR400. I don't know of any combinations that will beat a 3.2c+3200LL/ as far as CPU frequency and nice tight low latency RAM at FSB200/DDR400. So, you have saved $68 by buying the "correct" RAM. You have also saved yourself the headache of trying to underclock your PC4000 RAM and/or run it at timings you expect for $380 but that it CANNOT do. Other opinions please. Now let's say you want an average MOBO that will OC somewhat well but you are on a budget of $600 to upgrade. You have gotten yourself an Abit IS7-E for about $100, leaving $500 for RAM+CPU. It makes no sense to me to spend an inordinate amount of money on either RAM or CPU, you need a good balance. Abit IS-7$100 2.8c.......$266 Corsair…$315 TOTAL:$675 And, you could save $96 and still be able to OC VERY WELL with a 2.4c. Or, you could spend the $96 on more RAM. Abit IS-7 $100 2.4c........$170 Corsair...$315...(PC3700 TwinX1024 kit, 2 x 512mb) TOTAL.....$585 See the IS-7 review [URL=http://www.abxzone.com/forums/showthread.php?s=&threadid=47023]HERE[/URL] by Prometheus. So now, let's say that you absolutely have to spend all the money possible on a CPU+RAM combo and stick it in a Gigabyte GA-8KNXP Ultra MOBO. You get a 3.2c and either brand of PC4000 for Corsair's listed price @ Googlegear of $380. MOBO...$372 3.2c.....$655 RAM.....$380 TOTAL:...$1035 spent on a combination that most likely will NOT work that well together. It will work together but take the MAX FSB speed that Hawk was able to get with his 3.2c, of 238. That's DDR476. at a CPU frequency of 3.7ghz. Hawk has also run a 2.4c in the area of 3.25ghz with PC4000 OCd to about FSB 275/DDR550. The RAM bandwidth between DDR476 and DDR550 is SUBSTANTIAL. I don’t think that there is a 3.0c or a 3.2c on the planet that will run a FSB275 outside of a liquid nitrogen cooling system if even then. So, let's say you still want CPU speed so you buy a 2.6c for $211, SAVING $444 over the 3.2c, and you will have more memory bandwidth with the cheaper CPU. And, you will probably be able to run the 2.6c at 3.2ghz or faster but with enough FSB to really use the RAM. Possibly the same type of results with a 2.8c, Hawk got a FSB of 266 (which = 3.2ghz) out of one. If you are so lucky, you have your CPU at FSB266 and RAM at DDR524, still spanking the bandwidth you'll likely get from the 3.2c at FSB238/DDR476. The examples are endless but I hope these are adequate. I've applied my logic here. There are endless debates about brands but I am not debating Corsair vs OCZ vs Kingston (but look what forum I’m in ;-) or Asus vs Albatron etc. That's up to you, just put a sensible combination of parts into it and spend your money wisely. As always, gurus, experts, soothsayers, here it is, fire away! Happy Computing, Mike.
Link to comment
Share on other sites

[QUOTE][i]Originally posted by Zakamoto [/i] [B]Well, as I wasn't able to get my 3500 over 210fsb at 2-2-2-5-8, I loosened them up a bit to 2-2-3-5 so I could take it to 220 ;-) This stuff is unbelievable. Too bad it's dissapearing fast :-) [/B][/QUOTE] :( Yep. Last of the Winbond Miracle Chips:( RIP:zzzz:
Link to comment
Share on other sites

[QUOTE][i]Originally posted by Therion [/i] [B]Ok at this point i have to ask : which is better 1) 2 sticks of 256MB ram running as high as possible dual channel 2) 1 stick of 512MB ram that will SURELY run faster,but no dual channel ? [/B][/QUOTE] Dual channel hands down, probably even at SPD, more bandwidth. You could test it on your system if you like, even with 256 X 2 Dual vs 256 X 1 Single. That may not seem fair but look at it this way, a REALLY FAST single channel MOBO with 2 X 512 does not provide as much memory bandwidth as a dual channel MOBO w/2 X 256. It's the "width", not the "depth" which is what you get when you have multiple DIMMs in a single channel MOBO. The memory instructions simply cannot go into DIMMs in a "line" as fast as they can to 2 sticks side by side. Look at it as people performing manual labor, like digging a ditch. In single channel mode, only the 1st guy (digger) in line can dig until he exhuasts his capacity. Then, the 2nd digger (DIMM) can begin to help out and only by digging in the same direction (which= attempting to complete the same instruction before moving on to the next set of instructions) . In dual channel, not only are both digging at the same time, they can dig in different directions (which = different instructions) and therefore complete more digging and better digging independent of each other. I hope that is a clear analogy. Try it with your system and let us know. I bet it won't be much better in pure memory benches than a P4b CPU with HT at the same speed in a single channel board if you try it in single channel. Mike.
Link to comment
Share on other sites

I understood the analogy, but what i'm asking, i cannot test myself,as you know. I should have a stick of XMS3500C2-512MB to ba able to. I have seen that dual channel gives more mem bandwidth BUT !.... in order to stay in dual channel (and even more importantly, 1:1) , one has to run the FSB lower .Which means less memory bandwidth AND less CPU speed. So , if there are no important reasons to maximize mem bandwidth , then the question becomes intresting ! For example, in which tasks(programms) a memory bandwidth is notably more important than cpu speed ? This is the question !
Link to comment
Share on other sites

I have seen others more knowledgeable than me (I know very little about 3D/Video/AGP/Complete System based benches and programs) stating that in those instances CPU speed is really important for frame rates, video rendering, graphics rendering etc. However, I don't think that they are independent of the memory and will only get better with more bandwidth. I am planning on getting some other benchmark utilities to test these things with. My main drawback is my AGP card, a 32mb Matrox G550 Dual Head. No good at all for video benches. As far as staying at 1:1 I am changing my stance on one part of that. IF you can believe Sandra MAX2003, PAT DOES STAY ENABLED at 5:4. However, the inherent latency incurred by the memory running asynchronously seems to offset PAT to a great degree. I don't know any percentages but try it yourself. Run Sandra 1:1 at stock speed, buff or unbuff either one. Then, set to 5:4, run the FSB as high as you can get it and test again. My money is on the 1:1 score by a substantial margin. I just did a couple of runs that way and it bears out. Other than that, I still say 1:1 is KING! That's where the value of a CPU that will run a HIGH FSB and some PC4000 (or better even:D ) RAM will pay off. You can get your CPU speed back up pretty high AND have huge bandwidth. You could probably think of some ways to test more theories than you think with your current system. Keep in touch. Mike.
Link to comment
Share on other sites

Archived

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

×
×
  • Create New...