This is from another thread in the mobo section where an OC'er says 515 x 7 is a better performing and benching OC than 400 x 9 to get 3.6ghz. His argument that the higher FSB in itself results in higher bandwidth and therefore better benches while my argument is that 400 x 9 is a better more efficient OC that doesnt stress the mobo with higher NB temps and voltages. Furthermore, 400 x 9 allows you to tighten the RAM timings further and should result in (slightly) better performance benches. Is there any technical merit to a 515 x 7 OC to perform and score better than 400 x 9?
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
400 x 9 = 3.6ghz, or 515 x 7?
- Thread starter amenx
- Start date
BillParrish
Supreme [H]ardness
- Joined
- Aug 25, 2006
- Messages
- 7,519
if both are at a 2;1 (or the same memory divider, apples to apples) the higher FSB will result in more memory bandwidth and slightly better performace. CPU performance is a wash of course because ultimately it is running the same speed in both cases. C2D systems have been shown not to be very sensitive to memory timings. Normally you loosen them to acheive higher FSB as your friend suggests to increase cpu performance which overcomes the small penality imposed by looser timings.
If the higher FSB uses some weird memory divider and all but 2;1 (called 1:1 on most non-gigabyte boards) are weird, wait states and buffering eats up some of the bandwidth benifits.
But if the two systems you outline are both running the same divider the faster FSB will probally bench faster, not much but measurable.
Why dont you run some benchmarks with the system set up both ways and see what happens and let the rest of us get some sleep ?
Oh the other hand 517 FSB is a pretty high FSB so the stability and heat and reliabiliy also come into question, but you didnt ask about that
If the higher FSB uses some weird memory divider and all but 2;1 (called 1:1 on most non-gigabyte boards) are weird, wait states and buffering eats up some of the bandwidth benifits.
But if the two systems you outline are both running the same divider the faster FSB will probally bench faster, not much but measurable.
Why dont you run some benchmarks with the system set up both ways and see what happens and let the rest of us get some sleep ?
Oh the other hand 517 FSB is a pretty high FSB so the stability and heat and reliabiliy also come into question, but you didnt ask about that
Thanks Bill, those were my thoughts too. I know that memory bandwidth benches would increase, but not necessarily CPU benches (if at all). As we know AMD far surpasses C2D in memory bandwidth tests, but that dont mean squat in overall performance or FPS in gaming for example, so same thing here. And yes I know mem timings are not a biggie with C2D, but I think even the paltry gains with tighter timings should offset the assumed gains with a higher FSB thus making a 400 x 9 OC roughly equal to a 515 x 7 in performance. Although claiming a performance advantage of 515 x 7 over 400 x 9 is the rationale for this OC, I doubt it is in any way significant in CPU benching.
Yes, that was part of my argument when I said "400 x 9 is a better more efficient OC that doesnt stress the mobo with higher NB temps and voltages."Oh the other hand 517 FSB is a pretty high FSB so the stability and heat and reliabiliy also come into question, but you didnt ask about that
Cheers
http://www.hardforum.com/showthread.php?t=1190286
My real life testing for both setups. Benchmark results included.
Both are stable for hours upon hours of Orthos small form factor.
One more thing I'd like to say is that, while it may not be anything horribly noticable, this is [H]OCP, and as far as I remembered OCing to the highest stable OC was the norm around here, minor gains or not.
My real life testing for both setups. Benchmark results included.
Both are stable for hours upon hours of Orthos small form factor.
One more thing I'd like to say is that, while it may not be anything horribly noticable, this is [H]OCP, and as far as I remembered OCing to the highest stable OC was the norm around here, minor gains or not.
Oh Im not saying that your OC isnt stable or well performing, just that in principle one would tend to go for the least stressful route on the hardware/components to achieve the same results. As we know the NB on the DS3 is typically hotter than those on other mobos and can only get hotter with a higher FSB (I dont know if its improved with the v3.3, but I know mine is hot even at 375fsb).
If the highest stable OC is your goal, I'll bet you can 'slightly' edge past your current bench going with 400 x9 on a 1:1 ratio (yes, lower than your RAM spec) and with tighter timings (like 4-4-4-8 for example). Again it wont be by much, but I'll bet it would be higher than your current OC bench at 515 x7.
Cheers
If the highest stable OC is your goal, I'll bet you can 'slightly' edge past your current bench going with 400 x9 on a 1:1 ratio (yes, lower than your RAM spec) and with tighter timings (like 4-4-4-8 for example). Again it wont be by much, but I'll bet it would be higher than your current OC bench at 515 x7.
Cheers
BillParrish
Supreme [H]ardness
- Joined
- Aug 25, 2006
- Messages
- 7,519
yep just posting high numbers without taking system wide considerations into account does not impress me a lot, interesting. but not impressed. I used to play super-pi but it got boring after a couple of days. Best performace on the applications you use makes more sense to me. However in another life I did build a 72 cuda that ran at idle faster than the speed limits in most citys so its big fun to see what "she will do" then tune the beast to your needs. Its all good and all fun
yep just posting high numbers without taking system wide considerations into account does not impress me a lot, interesting. but not impressed. I used to play super-pi but it got boring after a couple of days. Best performace on the applications you use makes more sense to me. However in another life I did build a 72 cuda that ran at idle faster than the speed limits in most citys so its big fun to see what "she will do" then tune the beast to your needs. Its all good and all fun
Ha Ha! Well said. I don't play the game much for superpi, or anything else for that matter really
I do use it a lot to see how different settings run though, like in this instance, it really all started out with me wondering which ran faster, and theres no real world noticeable difference so I had to turn to the synthetics to quantify that tad of preformance increase. Anyways lol, also these sticks I have aren't the greatest, they can do high bandwidth, but hate tighter timings. I've got them running 4-5-4-12 @ DDR 1030, and thats about as thigh as I can get them. even when I tried them @ DDR 800 I couldn't get down to a cas3, so why bother, lol. Figured I'd settle for the faster spec, and in my case, with the setup and luck of the draw I got on mobo's, ram, and proc, best running system is what I'm currently running at.
As for northbridge/southbridge temps, well that was a fairly easy fix for me, ran down and got a little 50x50x20mm fan and hot glued it to the northbridge chipset lol. Dropped the temps way down to acceptable levels and helped to stabilize the OC. As for the southbridge, well I put a pci fan card in about 3 slots over my vid card (originally to help OC the vid card, but that was a failed experiment for another day lol!) I'll post a pic, but I also put it up high enough so that it's over the southbridge as well and wel it keeps it pretty cool.
Heh, all in good fun guys all in good fun
First off, in real life apps, any configuration wont make any noticable difference, 24/7 I run low fsb 1:1 even if my rams are rated at ddr1066....
Now for benchmark performance you can compare them like this.... all results from super pi...
400fsb (1:1) vs 514fsb ( 1333strap) = 514fsb is faster by a small margin and can easily be match by tighting the timings at 400...
400fsb ( 4:5 ) vs 514fsb(1333strap ) = equal
400fsb (2:3 ) vs 514fsb ( 1333strap ) = 400fsb wins by a large margin
400fsb (2:3) vs 514fsb( 1066strap) = almost equal, if ever 400fsb slightly faster...
In synthetic benchmarks, The 2 factors that affects performance the most is straps and memory speed... This goes for 965 chipset not sure about the others......
But like I said above, for 24/7, i like the setting that stresses my system the least...
Now for benchmark performance you can compare them like this.... all results from super pi...
400fsb (1:1) vs 514fsb ( 1333strap) = 514fsb is faster by a small margin and can easily be match by tighting the timings at 400...
400fsb ( 4:5 ) vs 514fsb(1333strap ) = equal
400fsb (2:3 ) vs 514fsb ( 1333strap ) = 400fsb wins by a large margin
400fsb (2:3) vs 514fsb( 1066strap) = almost equal, if ever 400fsb slightly faster...
In synthetic benchmarks, The 2 factors that affects performance the most is straps and memory speed... This goes for 965 chipset not sure about the others......
But like I said above, for 24/7, i like the setting that stresses my system the least...
Ok I've already done this, and I've used more benckmarks OTHER than superpi. I linked the thread where I posted the pics of all the benches, so how on earth are you going to tell me that the system is faster @ 400fsb, when clearly the tests show that it's faster @ 515?
Secondly, maybe you missed the fact that the memory sticks I have don't like tight timings for some reason, and they're already somewhat tight to be running 4-5-4-12 @ 1030?
3rd, when you are talking about the fsb and memory ratio could you be more specific with the dividers? For example you say 4:5 ratio is that on both the 515 and the 400 fsb? Or only for the 400? What about the 2:3? Doubt that you're talking about running over 1500mhz on the memory with a 2:3 divider @ 515 fsb. And also tight timings = ? to you? 3-3-3-9 @ 400?
Again I'm going to say that I've already run the benchmarks in multiple different applications as to what runs this system I've got faster, how are you going to sit here and argue that what you say is true when I've run everything myself on this rig?
Secondly, maybe you missed the fact that the memory sticks I have don't like tight timings for some reason, and they're already somewhat tight to be running 4-5-4-12 @ 1030?
3rd, when you are talking about the fsb and memory ratio could you be more specific with the dividers? For example you say 4:5 ratio is that on both the 515 and the 400 fsb? Or only for the 400? What about the 2:3? Doubt that you're talking about running over 1500mhz on the memory with a 2:3 divider @ 515 fsb. And also tight timings = ? to you? 3-3-3-9 @ 400?
Again I'm going to say that I've already run the benchmarks in multiple different applications as to what runs this system I've got faster, how are you going to sit here and argue that what you say is true when I've run everything myself on this rig?
If you're interested in synthetics, get the program mbench http://www.x86-secret.com/Download/mbench.zip
try 9x400 with 4:5 (2.50x).
then do 7x515 with 1:1 (2.00x)
mbench should show 400fsb with a significant advantage. That said, you said you've done spi runs. mbench is more synthetic, and not always indicative, but when it comes down to it, real life usage is what counts, right?
try 9x400 with 4:5 (2.50x).
then do 7x515 with 1:1 (2.00x)
mbench should show 400fsb with a significant advantage. That said, you said you've done spi runs. mbench is more synthetic, and not always indicative, but when it comes down to it, real life usage is what counts, right?
Socal, I wasnt arguing with you and I didnt even see your post before i posted just to make it clear.... The results I posted are my findings and its there to give info and that it, I didnt post it to prove a point.... Also those data were from experience running that benchmark, I didnt do those test at one time to find out whats the fastest setting for 24/7 cos I dont really care too much how fast my comp go on a daily basis.....
As far as not mentioning the divider for 514fsb, I thought its pretty obvious that you can only run 1:1 at that speed and the only thing that can be different is the strap.... Right now the only concern I have towards speed is superpi times and trust me, those results are pretty accurate...... in superpi...
As far as not mentioning the divider for 514fsb, I thought its pretty obvious that you can only run 1:1 at that speed and the only thing that can be different is the strap.... Right now the only concern I have towards speed is superpi times and trust me, those results are pretty accurate...... in superpi...
Oh we know mem bandwidth increases with FSB increases, but if mem bandwidth benches are your thing then you'll love AMD X2 or AM2, these kill C2D's in these type of benches. But as we know in overall performance and gaming, they just cant touch the C2Ds
.Interesting results. For some reason, even 400fsb is using the weaker internal latencies. And I meant to run the 400fsb test at 4:5 for 500ddr2 as well to get a more accurate gauge against 500 1:1. Strange! At 400fsb :
Access latency 48.7 ns (156 clocks)
Read datarate (INT) 7441 Mb/s
Write datarate (INT) 3091 Mb/s
Read datarate (MMX) 8108 Mb/s
Write datarate (MMX) 3108 Mb/s
Read datarate (SSE) 8453 Mb/s
Write datarate (SSE) 7264 Mb/s
That's at 8x400 4:5 4-4-4-10. See how the numbers across the chart top your 7x500, with the exception of sse write. I guess the results aren't that major, but if you can get similar with your ds3 somehow at 400fsb, then it might be the best option. Unfortunately, if you can't get into the tighter latencies (and I'm not sure why it isn't doing so), then 7x515 would definitely be the way to go. That said, I can't find my exact results for mbench and 8x400 on my ds3, but I recall them being tight. The results were taken from a p5b, but I could swear ds3 was same scenario at 400fsb.
Hope this helps and gives reason to why people say 9x400 is best.
edit: I recall there being a ram mode that you can set to fast or turbo or something like that. Maybe you should try to fiddle with those settings and see if you can get them to work. They're pretty hard on the northbridge so you may need 1.55vmch (+0.3v), so make sure you have good cooling. That said, 4:5 at 400 may require more vmch than 1:1 as well.
Access latency 48.7 ns (156 clocks)
Read datarate (INT) 7441 Mb/s
Write datarate (INT) 3091 Mb/s
Read datarate (MMX) 8108 Mb/s
Write datarate (MMX) 3108 Mb/s
Read datarate (SSE) 8453 Mb/s
Write datarate (SSE) 7264 Mb/s
That's at 8x400 4:5 4-4-4-10. See how the numbers across the chart top your 7x500, with the exception of sse write. I guess the results aren't that major, but if you can get similar with your ds3 somehow at 400fsb, then it might be the best option. Unfortunately, if you can't get into the tighter latencies (and I'm not sure why it isn't doing so), then 7x515 would definitely be the way to go. That said, I can't find my exact results for mbench and 8x400 on my ds3, but I recall them being tight. The results were taken from a p5b, but I could swear ds3 was same scenario at 400fsb.
Hope this helps and gives reason to why people say 9x400 is best.
edit: I recall there being a ram mode that you can set to fast or turbo or something like that. Maybe you should try to fiddle with those settings and see if you can get them to work. They're pretty hard on the northbridge so you may need 1.55vmch (+0.3v), so make sure you have good cooling. That said, 4:5 at 400 may require more vmch than 1:1 as well.
Good ideas I'll have to try some of them, thing is this memory just isn't as well preforming as some of the other micron I've seen out there. It doesn't like tight timings @ low fsb and it's kinda sketchy with the turbo settings too. I'll give it a shot.
If it's of any help, I wasn't able to use the fastest setting either. I think it went normal, fast, turbo or something. However it was, I was able to use the middle setting. Wish I had made more ds3 screenshots.
Well I think I've got one halfway bad stick and it's what's been holding me back timings/speed wise. Boot issues just started to show up with one of the modules, and I've already started the RMA process with OCZ.
Dunno what the deal is because it was never run out of spec (except for a few times I ran some benchmarks @ DDR 1200 5-5-5-15) with the voltage @ 2.32 (the memory is supposed to be operated at 2.3-2.35 according to OCZ's website.). So who knows maybe that's all that was wrong and things will get fixed here shortly.
Dunno what the deal is because it was never run out of spec (except for a few times I ran some benchmarks @ DDR 1200 5-5-5-15) with the voltage @ 2.32 (the memory is supposed to be operated at 2.3-2.35 according to OCZ's website.). So who knows maybe that's all that was wrong and things will get fixed here shortly.
That bright blue fan is right over the modules at all times, so yea they're being cooled. Not to mention the back case fan is pulling cold air into the case instead of drawing out.
So where exactly do the Northbridge straps kick in on the Gigabyte 965 boards? I did some research on this and I've got about 15 mbench tests to check where the strap kicks in but from 400 to 515, I can't seem to find any definable difference in chipset latency or clockspeed. In fact, Sandra reports the actual chipset clock speed and it stays on a constant increase with the fsb. with it at 515, the chipset is at about 640 internal clock and 400 @ 400fsb (go figure lol
). I'll post some of the mbench results in a little but theres no point beyond 400fsb that I can tell if it switches the chipset strap to the 1333 strap like on the asus boards, I know those kick in @ 401. Or any decrease in memory preformance. Maybe the strap kicks in at 333? and stays there all the way up to 515 where I'm at?
BTW it was one bad stick of ram, it's been RMA'd
well from the results you posted at 400fsb, it looks like internal latencies already weakened at that point for some reason. Otherwise, if you're running 1:1, you should see latencies in the 50's, and 4:5 (1000), upper 40's. I saw low 70's when 1:1 at 401fsb. When I used my ds3, I used mainly F7 bios. Maybe later ones changed the changeover at a lower fsb to make people stop complaining about an fsb hole and thinking that it doesn't clock well. Not sure though
Well I promised all the tests and well here they are!!!! 12 different ones all mbench, and I still tthink that the strap is applied way under 400, theres even a 332 and 333 test because I was trying to pinpoint the holes.
I couldn't find it lol. One note these were all done with 1 stick of ram (as it was all I had last night when I was testing since the other stick was bad lol.)
Let me know what you think!!!
PS- Look to the titles of the documents for the system settings for those tests (CPU Clock, Multi x Divider, latencies, and overall DDR speed.
I couldn't find it lol. One note these were all done with 1 stick of ram (as it was all I had last night when I was testing since the other stick was bad lol.)
Let me know what you think!!!
PS- Look to the titles of the documents for the system settings for those tests (CPU Clock, Multi x Divider, latencies, and overall DDR speed.