So what really is faster...3.8GHZ CPU but NB at 2000 or 3000MHZ?

[lopri]

Sure, but the bandwidth is just the same:

http://www.madshrimps.be/?action=get...58&articID=926

Check the Everest chart and compare GD70 to the M2RS in the 3.6G test setting. 600 CL4 beats 800 CL7

Yes, but because at that point the bottleneck is NB. It's my hypothesis that when NB is a bottleneck (i.e. NB frequency < 2 x memory frequency), higher memory frequencies are somewhat wasted. When you run Everest, keep an eye on memory read bandwidth and L3 read bandwidth. You will see both of them moving together very tightly. It seems to me, NB can only process as much as memory can feed it, and memory can send data only as fast as NB can process.

My hypothesis is in the post #27 and #28 in this thread. In your testing, if NB was clocked @2.8GHz, DDR3-1600/CL7 would be pulling ahead, leaving DDR2-1200 behind.

You can see it from Tony's first post in this thread. In the encoding test, look where the performance stops improving with DDR3-1333, yet keeps improving with DDR3-1600. It's somewhere around 2700~2800MHz NB. So DDR3-1333 is good enough to feed 2600MHz NB, but going higher you'll need a comparable (good timings) DDR3-1600 to keep the linear scaling.

So depending on applications - some apps will need to access system memory more often and others don't - NB frequency and memory frequency both matter, and ideally you'd want to have enough memory bandwidth for your NB even if some of it is wasted. If your NB doesn't clock any higher than, say, 2400MHz, then DDR2-1200 or DDR3-1333 will be all that's needed.

[massman]

That seems to me as a pretty solid hypothesis (not the numbers, but the concept)!

When I find the time, I'll try to rerun the everest bandwidth benchmarks and check with a higher-clocked NB frequency (if the AM3 boards allow me to).

[Tony]

Yes, but because at that point the bottleneck is NB. It's my hypothesis that when NB is a bottleneck (i.e. NB frequency < 2 x memory frequency), higher memory frequencies are somewhat wasted. When you run Everest, keep an eye on memory read bandwidth and L3 read bandwidth. You will see both of them moving together very tightly. It seems to me, NB can only process as much as memory can feed it, and memory can send data only as fast as NB can process.

My hypothesis is in the post #27 and #28 in this thread. In your testing, if NB was clocked @2.8GHz, DDR3-1600/CL7 would be pulling ahead, leaving DDR2-1200 behind.

You can see it from Tony's first post in this thread. In the encoding test, look where the performance stops improving with DDR3-1333, yet keeps improving with DDR3-1600. It's somewhere around 2700~2800MHz NB. So DDR3-1333 is good enough to feed 2600MHz NB, but going higher you'll need a comparable (good timings) DDR3-1600 to keep the linear scaling.

So depending on applications - some apps will need to access system memory more often and others don't - NB frequency and memory frequency both matter, and ideally you'd want to have enough memory bandwidth for your NB even if some of it is wasted. If your NB doesn't clock any higher than, say, 2400MHz, then DDR2-1200 or DDR3-1333 will be all that's needed.

You are bang on the money here as it stands there is absolutely NO NEED to run 1600 ram 7-7-7-24 1t with any AM3 CPU at stock, even if AMD up the NB clock to 2.6GHZ you still would do better with 1333 6-6-6-24 1T.
Also remember i did these tests at 3.8GHZ, if i get time I will rerun them at stock 3.2GHZ to see where that pans out although i have a feeling it would be very close to the same.

Today I will run Elpida Hypers (cas 7 blade) at 6-6-6-24 1t 1600 just to see how this one performs and add a new graph.

Overall my conclusion from this testing to date is AMD need to drop 2000NB clock today and implement 2600mhz on BE. There is no need for them to leave it at 2000 and they would see nice gains in performance when compared to Intel CPU's. Regards where OCZ can help with this I have already asked for a 6-6-6- kit at 1600, I will now also ask for a 5-5-5-kit at 1333 for those not OCing as high but wanting the best at that lower clock.

Keep your eye on the Blade and platinum series for these kits

[SKoob]

Mine doesn't like loading windows with the NB set to 2600, have to use the MSI multistep OC tool that applies the setting once the os has loaded Shows the windows is corrupted screen otherwise

[Zeus]

That means your system is not running stable so you have to raise NB-vid a tad if possible.

Mind you that raising CPU-NB volts makes the cpu run hotter though.

Btw, does msi allow you to raise CPU-NB freqency in windows? I thought it wasn't possible without a reboot.

[mAJORD]

yeah, I have to sy, on my board at least, 2600Mhz NB is a no go. Windows 7 64 s itself on reasonable voltages..

I don't see AMD raising the NB that hig for some time. They have to supply a CPU that will work in any motherboard, at stock voltages, on any OS I predict a 3.4Ghz 2.2Ghz NB part won't be too far off if their SOI process tweaks pay off as well as "usual"

Just my opinion.

[roofsniper]

Mine doesn't like loading windows with the NB set to 2600, have to use the MSI multistep OC tool that applies the setting once the os has loaded Shows the windows is corrupted screen otherwise

same i get the windows corrupted screen even if i set the clocks back. and i have to add volts if i want my nb to run over 2200 and just adding volts alone but no overclock will corrupt windows. kinda sucks and its a little confusing for me.

[gOtVoltage]

same i get the windows corrupted screen even if i set the clocks back. and i have to add volts if i want my nb to run over 2200 and just adding volts alone but no overclock will corrupt windows. kinda sucks and its a little confusing for me.

Hmm when you add volts for Higher NB it may be conflicting with the HT .


Try 2600 with the higher NBvolts and HT at same speed ,boot and Lower HT if it fails to load. Youd be surprized if it swhat i think it is.

Sometimes that the HT setting in a Bios makes adding Volts to NB unstable not the NBfreqency.

Even with the M3A79-T when i set a certain HT freqency and a certain NB Freqency it will fail loading at windows ( But lowering HT one or two stes its 100% stable at same Clocks). In some way the added volts is affecting my Bios while running on a certain HT divider with High Nb and it only does this when i raise the Nbvolts some or not enuff.

The cure for this while keeping the NB High with volts. I simply lowered the HT one or two settings.

Note: Mine does this because the Bios is bugged on some settings when i use certain Ht dividers with High NB.

Its worth a shot

[roofsniper]

ehhhhh ill try it out but i need to get a new dvd drive first incase it corrupts again. my last dvd drive died 2 weeks ago.

[Tony]

updated first post adding cas5 1333 in the mix...cas5 1333 hits back

[roofsniper]

updated first post adding cas5 1333 in the mix...cas5 1333 hits back

hmmm interesting you gonna try a 1600 c6? may want to think about putting a sticky or something on this. it is useful info.

[eRazorzEDGE]

I would be glad to do it on phase but.....thats not going to help you guys on air /water....Trying to avoid that as although its fun....its not much of a help to those on air/water.

I managed to test it on my 940 at 3000+ at one point, I know with ddr II it took an even bigger performance hit.

hey, i'm on phase

i'm gonna do some of my own testing here for s#!ts and giggles as i run 3200MHz NB 24/7 and with the M4A79 DDR2 board.

i really respect you guys doing all these indepth tests and drawing these pretty graphs. i get so impatient with it that i just settle for what seems ok most of the time. u, on the other hand, really push it... love every minute of it.

[Tony]

yes sir cas6 next. I may go back and push the CPU to 4000 also for all tests just to get that last piece of data where I think we may see a things change...its all time though so we will see

[Tony]

hey, i'm on phase

i'm gonna do some of my own testing here for s#!ts and giggles as i run 3200MHz NB 24/7 and with the M4A79 DDR2 board.

i really respect you guys doing all these indepth tests and drawing these pretty graphs. i get so impatient with it that i just settle for what seems ok most of the time. u, on the other hand, really push it... love every minute of it.

The point is this, there are way to many who quote what they think is fast...without actually knowing what is fast.

Check my sig

[eRazorzEDGE]

lol, well seeing what's been posted around here in the xtreme sections, i know i'm no where near "fast"

[Tony]

Was not a personal dig m8...please don't take it as one, but many just set systems up and don't actually test what is fast...

Also consider it may be easier for OCZ to make 1333 5-5-5-24 than 1600 6-6-6-6-24 and we now know cas5 1333 is faster than cas7 1600 so its a no brainer

Also looking at the graph 2400 to 2800NB is where most will land, and 1333 5-5-5-24 is killer here with air cooled CPU's from i would guess 3.6 to 4GHZ...so again its no brainer.

[eRazorzEDGE]

i don't take things to heart, tony, no matter what you say to me... and i didn't think anything wrong of what was being said until you brought it up. i was just going a different direction off your comment, taking a stab at myself

was anyone else able to hit 3000MHz on the NB under air or water? i never could on my M3A79-T, but once i got the M4A79 it as almost easy.

[demonkevy666]

yea i think when you overclock HT it also overclocks the HT connecting the cores as well.

I think ACC will/should show that.

however I also remembered we can split the hyper transport links going to the cpu in two half speed duplex.

though that seem pointless, unless we can find the inter core speed Ht links somewhere.

(the phenomenons of phenom's )

[roofsniper]

I think ACC will/should show that.

however I also remembered we can split the hyper transport links going to the cpu in two half speed duplex.

though that seem pointless, unless we can find the inter core speed Ht links somewhere.

(the phenomenons of phenom's )

yea whats the option called that lets you split it? im too lazy to go look. and how would it be useful for inter core links?

[Enoc]

I will now also ask for a 5-5-5-kit at 1333

updated first post adding cas5 1333 in the mix...cas5 1333 hits back

thanks tony

[massman]

Have you already figured out why the performance decreases when you increase the HT Link frequency? I have a completely different dataset, so I'm quite interested in your findings :-)

[Tom128]

The point is this, there are way to many who quote what they think is fast...without actually knowing what is fast.

I've actually been thinking this as well. I have not really verified on my setup what is the best speeds. My current DDR2 sticks can either hit 570Mhz 5-5-5 or 430 4-4-4. I am curious which is faster, and if I can tweak a few things to try and reach 600Mhz 5-5-5 or 500Mhz 4-4-4 (unlikely).

[Tony]

Have you already figured out why the performance decreases when you increase the HT Link frequency? I have a completely different dataset, so I'm quite interested in your findings :-)

post it up

[roofsniper]

Have you already figured out why the performance decreases when you increase the HT Link frequency? I have a completely different dataset, so I'm quite interested in your findings :-)

i have been told that it just gets the timing off and can cause cache misses and stuff. so with the HT link find the best clock not the fastest.

[tiborrr]

This is my stock cooled X3 720BE @ X4. Interesting fact that every mem kit i had (Crucial 8500F D9HXT; CSX Diablo DDR3-2000 D9GTR; OCZ Platinum DDR3-1800 8-8-8-x 1.95V) caps out at 855 CAS8 8-8-24 1T, 830 CAS7 7-7-21 1T.

And yes, 2700MHz is the hard limit for NB speed for given settings and poor cooling.

System:
Phenom X3 720BE @ X4 (ACC enabled) - stock cooled
2x512MB Micron D9HXT
ASUS M4A78T-E
PoV 8400GS 256MB PCIe
LCPower 420W

Settings:
Core voltage = 1.35V
CPU/NB voltage = 1.375V
DIMM voltage = 2.1V
HTT voltage = 1.24V
NB voltage = 1.44V
NB 1.8V voltage = 2V
SB voltage = 1.2V

CAS 6 hardly worth the effort? I have re-runned CAS7 with fresh reboot, opened SPi automatically on start-up to make sure no wazza-like effect happened in between.

WinRAR 3.4x benchmark gives only about ~ 20kB/s increase, almost the same as squeezing up the Trc.


CAS 6-7-6-18-34 1T:


CAS 7-7-6-18-34 1T: