AOA Forums AOA Forums AOA Forums Folding For Team 45 AOA Files Home Front Page Become an AOA Subscriber! UserCP Calendar Memberlist FAQ Search Forum Home


Go Back   AOA Forums > Hardware > EPoX MotherBoards

EPoX MotherBoards EPoX Intel and AMD Motherboards.


Reply
 
LinkBack Thread Tools Rate Thread
  #1 (permalink)  
Old 5th February, 2005, 12:27 AM
Member
 
Join Date: January 2005
Posts: 12

Lower than expected memory bandwidth 9nda3j

I got PC4000 HyperX ram, have the board running 1:1, 255 mhz and I barely break 6k in SiSoft Sandra memory benchmarks. Most people who I've seen with similar #'s are closer to 7k, my htt is at 4x. I don't understand
__________________
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #2 (permalink)  
Old 5th February, 2005, 12:50 AM
Member/Contributor/Resident Crystal Ball
 
Join Date: March 2004
Posts: 7,451

what multi are you using. the higher multi's perform better for some reason.
__________________
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #3 (permalink)  
Old 5th February, 2005, 01:20 AM
Finetobe's Avatar
Member
 
Join Date: June 2004
Location: Denmark
Posts: 525

are your memory runing at 2T.
i get about 7400 mb/s at 251x11 2.5.4.4.8 1T.
but cpu speed has some thing to do with it. higher cpu clock .higher memory bandwidth.
__________________
2500K-4600mhz/8gb ram /2x6970.
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #4 (permalink)  
Old 5th February, 2005, 01:21 AM
Member
 
Join Date: January 2005
Posts: 12

Running 3-4-4-8 @ 500mhz DDR
winnie 3200+ @ 2.5 ghz
command rate 1t and 2t make hardly any diff in score
__________________
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #5 (permalink)  
Old 5th February, 2005, 01:52 AM
Finetobe's Avatar
Member
 
Join Date: June 2004
Location: Denmark
Posts: 525

i have just test my pc in sandra.
250x10 an memory at 2.5.5.5.10.
got 6887 mb/s.
have you testet in memtest. if your memory make any error. you lose a lot of bandwidth.
http://www.memtest86.com
__________________
2500K-4600mhz/8gb ram /2x6970.
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #6 (permalink)  
Old 5th February, 2005, 02:10 AM
Member
 
Join Date: February 2003
Posts: 47

try everest, used to be aida32, i don't trust sandra anymore, seems nforce chipset support is lacking.

1/2T command rate seems to make much more difference on ram write bandwidth than ram read bandwidth, for me at least.

also, try to drop to cas 2.5 instead of cas 3.0, that should make a difference too, provided your ram can handle it.

finetobo, did you find your ram timings of 5-5-10 are better than 3-3-10? just wondering since i rarely see timings like you've shown.
__________________
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #7 (permalink)  
Old 5th February, 2005, 02:24 AM
Finetobe's Avatar
Member
 
Join Date: June 2004
Location: Denmark
Posts: 525

my memory cant run lower then 2.5.4.4.8 but they can do it at 290mhz 1T or 300 mhz at 2T. but the memory controler dont like that speed. anything over 260 mhz and it crash.
__________________
2500K-4600mhz/8gb ram /2x6970.
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #8 (permalink)  
Old 5th February, 2005, 02:32 AM
Member
 
Join Date: February 2003
Posts: 47

i know the feeling, i have ram that can do about the same, and my cpu memory controller craps out at 252mhz. what a waste to have ram that can run 2.5 3-3-7 1T at 275mhz (rated speed) with this particular cpu.
__________________
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #9 (permalink)  
Old 5th February, 2005, 02:53 AM
Finetobe's Avatar
Member
 
Join Date: June 2004
Location: Denmark
Posts: 525

we have to hope that amd make a better memory controler in the future.
my old setup. 8kda3+ and a 3400 newc. could run 275 mhz before the memory controler craps out, with 1gb of ram or 300 mhz with 512 mb ram.
__________________
2500K-4600mhz/8gb ram /2x6970.
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #10 (permalink)  
Old 5th February, 2005, 09:22 AM
EPoX Tech's Avatar
EPoX Tech
 
Join Date: October 2001
Location: Seems not to matter
Posts: 4,720

In retrospect 275 can't be bad for a part thats rated at 200
__________________
If you need EPoX BIOS eeproms in the UK, goto http://www.epox.org.....................
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #11 (permalink)  
Old 5th February, 2005, 11:53 AM
MietmaN's Avatar
Member
 
Join Date: January 2005
Location: Heidelberg (DE), Kraków (PL)
Posts: 413

Talking

I've just found something interesting for you guys - according to AnandTech the TRas (RAS precharge) memory timing gives better memory bandwidth when set to 10 instead of the lower possible setting (e.g. 7 or 8).

http://www.anandtech.com/mb/showdoc.aspx?i=2253&p=7

"Several memory tests have shown that memory performs fastest on the nVidia nForce chipsets at a TRas (RAS Precharge) settings in the 9 to 13 range. We ran our own Memory Bandwidth tests with memtest86 with TRas settings from 5 to 15 at a wide range of different memory speeds. The best bandwidth was consistently at 9 to 11 at every speed, with TRas 10 always in the best range at every speed. The performance improvement at TRas 10 was only 2 to 4% over TRas 5 and 6 depending on the speed, but the performance advantage was consistent across all tests."

It's like with the "magic eleven" on nFroce2 chipset.
__________________
AMD AXP-M 11x220 1.675v| Epox 8RDA3+ PRO 1.4| Thermalright SLK900U /Y.S.Tech 92mm|
2x512MB Mushkin PC3200 L.II V2| Sapphire Radeon 9800 PRO@XT| S.B. Live 5.1| Maxtor 120GB|
Seagate 80GB| Teac CD-W540E| Asus DVD| BeQuiet Blackline 450W| Chieftec BX| Iiyama HM704UTc|
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #12 (permalink)  
Old 5th February, 2005, 11:54 AM
EPoX Tech's Avatar
EPoX Tech
 
Join Date: October 2001
Location: Seems not to matter
Posts: 4,720

Yes - 10 or 11 is a good timing for NF2 and also AMD64.

It makes sense when you think of the JEDEC calculation for sum rise and fally command plus strobe - adding more latency at peak by iincreasing the whole wavelength means that less error correction is needed and you get a better hitrate.

If I elaborate - because the flat at the high peak of the 2 instructions (DDR) is longer it had less chance of failing and hence the instructions are more likely to be succesfull and not need to be re-comitted. Memory efficiency is therefore increased.
__________________
If you need EPoX BIOS eeproms in the UK, goto http://www.epox.org.....................
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #13 (permalink)  
Old 5th February, 2005, 12:26 PM
MietmaN's Avatar
Member
 
Join Date: January 2005
Location: Heidelberg (DE), Kraków (PL)
Posts: 413

Question

Quote:
Originally Posted by EPoX Tech
Yes - 10 or 11 is a good timing for NF2 and also AMD64.

It makes sense when you think of the JEDEC calculation for sum rise and fally command plus strobe - adding more latency at peak by iincreasing the whole wavelength means that less error correction is needed and you get a better hitrate.

If I elaborate - because the flat at the high peak of the 2 instructions (DDR) is longer it had less chance of failing and hence the instructions are more likely to be succesfull and not need to be re-comitted. Memory efficiency is therefore increased.
Thanks a lot!

I've heard about the tRAS 11 on nForce2, now I know it works on nForce3 as well with the tRAS 10 as the best setting possible, but I've never had the faintest idea why it is so ...

Does it concern DDR memory only? I mean is it connected with the fact, that 2 instructions are carried in one memory cycle (hope I'm using the proper terminology) instead of just one instruction like SDR?
__________________
AMD AXP-M 11x220 1.675v| Epox 8RDA3+ PRO 1.4| Thermalright SLK900U /Y.S.Tech 92mm|
2x512MB Mushkin PC3200 L.II V2| Sapphire Radeon 9800 PRO@XT| S.B. Live 5.1| Maxtor 120GB|
Seagate 80GB| Teac CD-W540E| Asus DVD| BeQuiet Blackline 450W| Chieftec BX| Iiyama HM704UTc|
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #14 (permalink)  
Old 5th February, 2005, 12:39 PM
EPoX Tech's Avatar
EPoX Tech
 
Join Date: October 2001
Location: Seems not to matter
Posts: 4,720

At the peak of the rising edge and very beginning of the falling edge an instruction is carried on two parsed planes of voltage to get the effect of double data rate - but if through error correction (failing instructions) the efficiency is lowered the double data effect lessens. Early DDR controllers such as the KT266 and AMD761 whilst offering performance superlative to that of the KT133a were quite poor in getting a good hit rate.

Enter the KT266a which is the basis of all VIA DDR subsystems (the only new technologies being Faststream 64 and dual channel with the KT400a and KT880 as driving models) - the techniques for purging real world performance increases were near perfected and the only major improvements we have seen have been with Intel's 865/875 and the NForce 2 chipset and the clever low logic to memory linkage of the AMD64 series.

Part of the amazing increase within the newer AMD / NVidia influenced designs is the way in which a higher TRAS can be valid and not send the memory unstable since the calculation for the total wavelength of each cycle may fall outside of the stability envelope of the memory module. A trick and a half to gain a better hit rate whilst not making the memory trip up.

This is why I don't like certain kinds of so called enthusiast memory that have very stupid SPD programmed on them and unworldly ratings - often they mislead purchasers - in 99% of cases good chipset or CPU manufacturers approved modules will offer more stability for Joe Six Pack than anything called Corsair, Geil, OCZ etc. Matter of fact for people that do know what they are doing though is that these memories when configured correctly will blow anything else clean out of the water.

Memory bandwidth therefore as measured by Sandra, Everest or whatever can only be as good as the memory and ability of the person configuring the system.

Poor memory and poor ability = poor bandwidth, basic memory and good ability = good bandwith, corsair like memory and good ability = amazing bandwidth.
__________________
If you need EPoX BIOS eeproms in the UK, goto http://www.epox.org.....................
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #15 (permalink)  
Old 5th February, 2005, 01:13 PM
Chief Systems Administrator
 
Join Date: September 2001
Location: Europe
Posts: 13,075

Quote:
Originally Posted by MietmaN
but I've never had the faintest idea why it is so ...
Really, really basically, the setting works the opposite way that you think it does. It's basically a timer to tell the chipset the longest time it can keep the memory open for data. When the timer expires, the transfer is interrupted. So, if you set it too low, then the transfers are interrupted prematurely, dropping your bandwidth. At the same time, going over the optimal setting doesn't increase performance at all, and can hurt it slightly.

It doesn't really matter which chipset is in use, as the memory timings are set by the memory rather than the chipset. Hence the calculations for the nForce 2 chipset should also be applicable to other chipsets too!
__________________
Any views, thoughts and opinions are entirely my own. They don't necessarily represent those of my employer (BlackBerry).
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #16 (permalink)  
Old 5th February, 2005, 01:16 PM
EPoX Tech's Avatar
EPoX Tech
 
Join Date: October 2001
Location: Seems not to matter
Posts: 4,720

To add to that ..........

In the case of NF2 and AMD64 the ROMSIP / XGroup's timings determine the ability to have an efficient long clock.
__________________
If you need EPoX BIOS eeproms in the UK, goto http://www.epox.org.....................
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #17 (permalink)  
Old 5th February, 2005, 02:23 PM
MietmaN's Avatar
Member
 
Join Date: January 2005
Location: Heidelberg (DE), Kraków (PL)
Posts: 413

Post

Quote:
Originally Posted by Áedán
Really, really basically, the setting works the opposite way that you think it does. It's basically a timer to tell the chipset the longest time it can keep the memory open for data. When the timer expires, the transfer is interrupted. So, if you set it too low, then the transfers are interrupted prematurely, dropping your bandwidth. At the same time, going over the optimal setting doesn't increase performance at all, and can hurt it slightly.
Well, having read all of the above posts, I don't think I was thinking incorrectly, but providing we speak about all memory timings : setting a timer too low, we get some system's instability or our motherboard just won't boot at all, if timings are really too sharp for our memory to handle them. The instability can be easily discovered while running Prime95 Blend Test in most cases or MemTest86 to find some more subtle instability issues - Gold Memory is also a good way to test.

All this concerns a specific individual memory stick, that has to be tested in order to make sure we won't get any errors. Setting e.g the lowest possible CAS Latency increases memory bandwidth, as a rule, but we must be sure we won't get any errors. Well, of course 99% of certainty - 100% certain are only death and taxes.

As to RAS precharge - I must commit I didn't know the meaning of this particular timing and therefore I was thinking the other way round. Some time ago I read an article on "magic 11" with the nForce2 chipset ( www.madshrimps.com ), but still I've been using the setting blindly, until now.

@ EPoX Tech :

I think this is becoming an OT a little bit, but still VERY interesting - I think you can both write a short article on memory timings. I'm serious - most of the information is already here, in this topic. Then it could be sticked either here or somewhere else in the AOA forums.
__________________
AMD AXP-M 11x220 1.675v| Epox 8RDA3+ PRO 1.4| Thermalright SLK900U /Y.S.Tech 92mm|
2x512MB Mushkin PC3200 L.II V2| Sapphire Radeon 9800 PRO@XT| S.B. Live 5.1| Maxtor 120GB|
Seagate 80GB| Teac CD-W540E| Asus DVD| BeQuiet Blackline 450W| Chieftec BX| Iiyama HM704UTc|
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #18 (permalink)  
Old 5th February, 2005, 03:03 PM
Member
 
Join Date: January 2005
Posts: 12

I agree, this is becoming very insightful
But,back to my lesser topic
I've done about everything I can do to get normal benchies in Sandra, I've reinstalled windows ( I was a little overdue on a reinstall ), I've tried different settings voltages, timings, trying different versions of motherboard drivers.
Though, if I have the ram running at 200mhz instead of 250, I'm at 4,500. Which is considerably lower than it should be. Just to clear up overclocking being the culprit.
I'm at a complete loss. I'll take any other suggestions, could the ram just be underperforming? I also ran memtest86 for 3 hrs and had no errors at 250mhz
memory controller on the cpu hozed?

Slowly getting frustrated
Thanks!
__________________
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #19 (permalink)  
Old 5th February, 2005, 03:16 PM
MietmaN's Avatar
Member
 
Join Date: January 2005
Location: Heidelberg (DE), Kraków (PL)
Posts: 413

Have you got a matched pair of memory sticks? I just can't think of anything else, that could afflict memory performance so badly. How many sticks, what make and in which mem slots? If we know this, I think the more experienced guys here will be able to help.

BTW. While changing chipset drivers it is good to run DriverCleaner to get rid of everything, that remains in the system after uninstalling the previous ones.
__________________
AMD AXP-M 11x220 1.675v| Epox 8RDA3+ PRO 1.4| Thermalright SLK900U /Y.S.Tech 92mm|
2x512MB Mushkin PC3200 L.II V2| Sapphire Radeon 9800 PRO@XT| S.B. Live 5.1| Maxtor 120GB|
Seagate 80GB| Teac CD-W540E| Asus DVD| BeQuiet Blackline 450W| Chieftec BX| Iiyama HM704UTc|
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #20 (permalink)  
Old 6th February, 2005, 03:21 AM
Member
 
Join Date: January 2005
Posts: 12

Alrighty, I have a matched pair of Kingston HyperX PC4000 ram(2x512). I've tried them in slots 1&2 and 3&4 and I get the same performance either way.
__________________
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
Reply



Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
 
Thread Tools
Rate This Thread
Rate This Thread:

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is Off
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
Memory bandwidth seems low danrok General Hardware Discussion 11 28th April, 2004 05:54 PM
8RDA3+ rev 1.1 Vcore increment => lower memory performance... Why? Nfountou EPoX MotherBoards 4 20th March, 2004 04:48 PM
2500 = lower bandwidth 8rda3+? speeduk EPoX MotherBoards 8 7th August, 2003 07:43 AM
low memory bandwidth m4a2t0t EPoX MotherBoards 11 27th May, 2003 06:57 PM
More memory = lower performance?? Genius EPoX MotherBoards 13 18th November, 2002 06:36 PM


All times are GMT +1. The time now is 10:52 AM.


Copyright ©2001 - 2010, AOA Forums
Don't Click Here Don't Click Here Either

Search Engine Friendly URLs by vBSEO 3.3.0