Low memory bandwidth - 8RDA +/3+/6+ PRO!

[MietmaN]

Hi everybody,

I've had several nFroce2 mobos during the last few months (Epox 8RDA3+ rev.2.1, Abit NF7 v.2.0, Epox 8RDA3+ PRO rev.1.2), using the same clean system - Win2k SP4. After having done some tests with the current one just a few days ago I marked a real low memory bandwidth /efficiency in comparison with my old (now unfortunately dead) motherboards. I also had a possibility to test two 8RDA+ PRO rev.1.2 for a short period of time and the results were the same - disappointing.

That made me read some Epox 8RDA3+ (PRO) and Epox 8RDA6+ PRO reviews, in which the same issue had been discovered. Since 6+ PRO and 3+ PRO are actually the same mobos, with the exception of the additional SATA controller on the latter, sharing the same bios, I assume they also unfortunately share the same issue. Even more - all of the mentioned "PRO" mobos use exactly the same PCB (well, at least in this 1.2 revision), so I think this is common for all of them.

From my personal experience this afflicts the SiSoft Sandra 2004 memory bandwidth benchmark, both read and write tests in Aida32 /Everest Home, 3DMark 2001SE and SuperPI - these are only examples to show everyone, that the discovered low bandwidth is true and has certainly got some bad influence on the overall system performance, resulting with worse FPS in games and worse benchmarks' results.

Some facts :

http://www.ocworkbench.com/2004/epox/8rda6+pro/b2.htm

http://www.ocworkbench.com/2004/epox/8rda6+pro/b4.htm

http://www.ocworkbench.com/2004/epox/8rda6+pro/b7.htm

"In the tests, we checked out the performance in comparison with older NF2 and other chipsets. Despite being a new chipset that bundles lots of good features, it it slightly slower than its predecessor. The first thing that comes into my mind would be the integrated firewall and dedicated Gigabit Ethernet and SATA that has taken a portion of that bandwidth. We experimented it by turning them off and not installing drivers. We tested with a couple of new BIOS but it still remains the same."

The same goes for me - I've already tried 3 different bioses, changed alfa timings with TicTac's nForce2 Tweaker, turned everything useless off in the bios, checked everything almost hundred times - without any improvement. Sandra's 2004 results with 2x512MB 2-3-3-11 running in dual DDR mode are 100-200 MB/s worse than with 1x512MB 2.5-3-3-11 on my old 8RDA3+ rev.2.1.

400-500 less in 3DMark 2001SE (observed after many runs) - I used the same nForce unified drivers and the same ATI Catalyst 4.8 drivers to make the comparison as objective as I could.

Have a look at my SuperPI 16M calculation's time :

Abit NF7 v2.0

On my current Epox :
11.5x225 - 17m 25s
12x218 - 17m 22s

Some more reviews :

http://www.legionhardware.com/html/doc.php?id=325

http://www.amdzone.com/modules.php?o...rtid=58&page=6

http://www.pcstats.com/articleview.c...id=1623&page=5

http://www.pcstats.com/articleview.c...id=1623&page=6

Exactly the same issue could be discovered in some of the oldest and the newest revisions of 8RDA3+ :

http://www.motherboards.org/reviews/...ds/1254_4.html

http://www.pcper.com/article.php?aid=58&type=expert

http://www.pcper.com/article.php?aid...e=expert&pid=9

My dead, unforgotten 8RDA3+ rev.2.1 equaled in memory performance with the best nFroce2 mobos like DFI Lan Party /Infinity and Abit NF7 v.2.0 (I used D24 original bios, D26 Discovery CPC on by Merlin, D26 Black Mantaray CPC On by TicTac).

.... and don't get me wrong, 'cause I'm moaning all the time - apart from that really irritating low memory bandwidth I'm satisfied with my mobo. I only want to know, if this can be fixed with some bios tweaks and new bios release or this is connected with the PCB and nothing can help to boost the memory performance.

Any suggestions?

[Wolf2000me]

Hey,

I honestly don't know what you mean with bios' alpha timings. But if it's the same thing as the bios' romsip table then ignore the following. What i did hear was the this.
If you set your fsb to 166 mhz, the bios implements a more agressive romsip table. Try to overclock from that setting in Windows. I know of at least one user gaining performance that way.

At any rate, i did the same test you did in super_pi. However I do get better results than you while my fsb and cpu speed are a bit lower. I am running dual channel so that might explain it.

these are my results. Don't be fooled by the other results, they are at 215x12 which i don't get a 100% stable in dual channel mode.
I was in the meantime typing the stuff above, using Opera.
I'm using 2 Western Digital 160gb 8mb cache in RAID 0 on the nvraid interface. I also have 2 other disks, one on the pata and one on the silicon image.

Btw, do you get the same OC results with the same cpu in all nforce2 boards you used? I assume you use at least once the same cpu when you switched mobos.
Also, if the bios' alpha timings are not the same as the romsip, could you give a short explanation and a link?
My previous rig was an abit kr7a-raid with a mobile 2400 so that won't be much of a fair comparison.
I don't know if you tried the integrated firewall? My findings with it are a bit on the negative side. While browsing with Opera it can slow down any loading website for even 30 seconds.

The rest of my system is in my sig, i'm using winxp pro sp1.

I also may have more info, but then ask because nothing really pops in mind.

Cheers dude

Oh and I too am happy with the board. It's brought me the performance I wanted without spending too much. Of course any performance gained is gladly taken

[EPoX Tech]

It should be that if you wish for highest performance at greater than 200MHz FSB - you should use the 166MHz ROMSIP - solving all of your "findings" above.

[labtech1]

I found a reduction in the memory bandwith with my board as well when running at 200Mhz FSB or higher. Fortunately there is an easy solution as referred to above. If you set the FSB in the bios at 166~198Mhz and the cpu interface on "aggressive" then overclock from within Windows memory bandwith should be very close to what you have been used to.

[Wolf2000me]

Hey,

ET, a little question for you...
Have you looked at the config files of USDM yet?

Thanks

btw, the romsip thingy, didn't I already say that?

[MietmaN]

Originally Posted by Wolf2000me

Hey,

I honestly don't know what you mean with bios' alpha timings. But if it's the same thing as the bios' romsip table then ignore the following.

These are sort of hidden memory timings, which u can change in bios while using VIA chipsets for socket A. As far as I know they've got nothing to do with bios' romsip table - they're just memory timings, not important enough, to be set manually (the performance increase is almost marginable).

Originally Posted by Wolf2000me

What i did hear was the this.
If you set your fsb to 166 mhz, the bios implements a more agressive romsip table. Try to overclock from that setting in Windows. I know of at least one user gaining performance that way.

I've never overclocked my CPUs in Windows - so you mean I should use 8rdavcore or similar sowftware? Why should I bother doing this, when I don't have to that while using other mobos, e.g. 8RDA3+ rev.2.1? I'm gonna try this method anyway .... thanx, mate

Originally Posted by Wolf2000me

At any rate, i did the same test you did in super_pi. However I do get better results than you while my fsb and cpu speed are a bit lower. I am running dual channel so that might explain it.

these are my results. Don't be fooled by the other results, they are at 215x12 which i don't get a 100% stable in dual channel mode.
I was in the meantime typing the stuff above, using Opera.
I'm using 2 Western Digital 160gb 8mb cache in RAID 0 on the nvraid interface. I also have 2 other disks, one on the pata and one on the silicon image.

I've forgotten one important thing - under Wink2k everybody gets worse results than under WinXP SP1/SP2 or Win2k3 Server - these values varies from 40 sec to 1 minute (16M). And, to be honest, I'm also running in dual channel mode with those 2x512MB mem sticks (FSB222 stable, FSB225 stable enough to count SuperPI 16M).

Originally Posted by Wolf2000me

Btw, do you get the same OC results with the same cpu in all nforce2 boards you used? I assume you use at least once the same cpu when you switched mobos.

Yes, it's my Athlon XP-M 2500+ 1.45v, L12 mod done - all the other tests (Sandra 2004, 3DMark 2001SE, Everest) were done at 11x200.

Originally Posted by Wolf2000me

Also, if the bios' alpha timings are not the same as the romsip, could you give a short explanation and a link?

http://www.xtremesystems.org/forums/...p/t-39330.html

Originally Posted by Wolf2000me

My previous rig was an abit kr7a-raid with a mobile 2400 so that won't be much of a fair comparison.
I don't know if you tried the integrated firewall? My findings with it are a bit on the negative side. While browsing with Opera it can slow down any loading website for even 30 seconds.

I've been using Sygate Personal Firewall for more than two years so far and it doesn't really absorb to mouch of the system's resources. Going back to my test - I always turn everyting off, leaving only about 16-17 system processes activated.

Originally Posted by Wolf2000me

The rest of my system is in my sig, i'm using winxp pro sp1.

I also may have more info, but then ask because nothing really pops in mind.

Cheers dude

Oh and I too am happy with the board. It's brought me the performance I wanted without spending too much. Of course any performance gained is gladly taken

If I come across something interesting, I'll let everybody know.
Shame I'm not into bios editing and Merlin ist too busy these days, to do something in this case.

Cheers mate

PS. I don't do that normally, but I think I will answer the other posts in a separate one.

[MietmaN]

Originally Posted by EPoX Tech

It should be that if you wish for highest performance at greater than 200MHz FSB - you should use the 166MHz ROMSIP - solving all of your "findings" above.

Ok, I'm gonna go for it anyway, but could you explain, why shuold I do such little oc-tricks, when memory bandwidth and efficieny were so high e.g. with my old Epox 8RDA3+ rev.2.1 at FSB200 and even higher? They really were comparable with Abit NF7 v.2.0 (at FSB200 no matter which bios I used - stock D24 or D26 Black Mantaray) and I didn't have to do any OC in Windows. Were romsip tables at FSB200 more performance oriented and the whole bios more mature on my 8RDA3+ rev.2.1?

@labtech1 - as already said a couple of times, I'm gonna try this. CPU Interface was always set to agressive - I even did some tests to discover the differences and they were quite big. (btw. the same goes for Abit nad CPU Interface Enabled/Disabled - exactly the same thing, I assume).

Thanx everybody, once again

UPDATE : Yes, it works like it should
11x166 in bios and then 11x200 in Windows thanks to 8rdavcore. 3D2001SE back again at 18400-18600 instead of 18000-18200; Memory bandwidth (Sisoft Sandra 2004) back at 3080MB/s Int and 2840MB/s Float like in the good old times - everything all right.

SuperPI 16M - 19:15 on my dead Abit with D26 Black Mantaray (Win2k SP4), now 19:22 (Win2k SP4) - almost no difference.

ET, could it be fixed in the next bios release? - I mean the FSB200 romsip table with more aggressive values?

[Wolf2000me]

Originally Posted by MietmaN

UPDATE : Yes, it works like it should
11x166 in bios and then 11x200 in Windows thanks to 8rdavcore. 3D2001SE back again at 18400-18600 instead of 18000-18200; Memory bandwidth (Sisoft Sandra 2004) back at 3080MB/s Int and 2840MB/s Float like in the good old times - everything all right.

SuperPI 16M - 19:15 on my dead Abit (Win2k SP4), now (Win2k SP4).

ET, could it be fixed in the next bios release? - I mean the FSB200 romsip table with more aggressive values?

I'd like that too in the next bios release. Btw, ET could you upload a bios binary for 8rda3/6 pro on your .org website? That way i could check your website first for new releases. Now i feel a bit left out because there's so much listed but no binary for our kind of board. Unless i'm mistaken but the only thing i've seen is the RAID-5 bios and it's in a different section.

MietmaN, do you note the exact same stability as without the more agressive romsip table? I mean, if it errors in prime95 does it error at almost the same moment? I don't have time to test it out myself because i'm in the examination period. I'm actually leaving now to take the datacommunications and networking exam. Of course i will do it after the 22nd of januari but since you seem to have experience already on the exact subject

Great to have you posting and sharing your findings here MietmaN
Btw, when you run at the fsb you did the prime tests with, how stable are you? Because at 215 fsb i'm not sure if it's my memory failing or the memory controller not being able to keep up. I crash after 7 minutes in prime blend test with 1.8v vdd. It crashes later when i set 2v vdd but my applications and full screen graphical appz crash earlier on 2v. Did you lap and reapply thermal compound on the northbridge heatsink?

Seems like i'm asking info instead of sharing more like i said in the first reply

Anywayz, wish me some luck and cheers m8

[EPoX Tech]

I have just put up da6p4a05.bin for you so you will now see the file upgrade as you wish.

[MietmaN]

Originally Posted by Wolf2000me

I'd like that too in the next bios release. Btw, ET could you upload a bios binary for 8rda3/6 pro on your .org website? That way i could check your website first for new releases. Now i feel a bit left out because there's so much listed but no binary for our kind of board. Unless i'm mistaken but the only thing i've seen is the RAID-5 bios and it's in a different section.

The last two releases can be found on the american Epox's website - no binary files for 3+ PRO,

http://www.epox.com/USA/product.asp?id=EP-8RDA3plusPRO

cause it shares the same bios with 6+ PRO :

http://www.epox.com/USA/downloads.as...P-8RDA6plusPRO

Originally Posted by Wolf2000me

MietmaN, do you note the exact same stability as without the more agressive romsip table? I mean, if it errors in prime95 does it error at almost the same moment? I don't have time to test it out myself because i'm in the examination period. I'm actually leaving now to take the datacommunications and networking exam. Of course i will do it after the 22nd of januari but since you seem to have experience already on the exact subject

1) Old romsip table (original FSB200) : I can achieve rock stable FSB220, checked with hours of Prime95 Blend and MemTest as well. I could even reach FSB222, but this is the limit of rock stable frequency. FSB225 - I can run all well known 3D benchmarks as well as SuperPI 16M, but Prime95 blend crashes after few minutes. FSB230 - I can get into Windows, but my system is very unstable. I'm sure it isn't the memory (PC4000) , that's holding me down.
On Abit NF7 I could reach rock stable FSB225 with D26 Black Mantaray CPC On, and rock stable FSB235 with the same bios, but CPC Off.

http://www.insanetek.com/index.php?page=cpcnf2

FSB200 : 2-3-3-11
FSB213 : 2.5-3-3-11
higher : 2.5-4-4-11 (those sticks should do FSB250 at 2.5-4-4-10)

It's either the chipset itself, that can't handle 2x512MB in dual DDR mode, or there's something wrong with my L12 mod.

Raising the chipset voltage doesn't help at all at any time - I've heard it's simply so on many boards. I only applied 1.80v, didn't try 2.0v - not with my passive cooling - too risky

2) New romsip table (original FSB166) :
Just testing, although I haven't got much time and some serious exams coming really soon.

Originally Posted by Wolf2000me

Great to have you posting and sharing your findings here MietmaN
Btw, when you run at the fsb you did the prime tests with, how stable are you? Because at 215 fsb i'm not sure if it's my memory failing or the memory controller not being able to keep up. I crash after 7 minutes in prime blend test with 1.8v vdd. It crashes later when i set 2v vdd but my applications and full screen graphical appz crash earlier on 2v. Did you lap and reapply thermal compound on the northbridge heatsink?

See, what I've just written above - raising Vdd doesn't help at all, well, at least in my case on my mobo.
I've got the blue Zalman passive cooling on my NB with Arctic Ceramique in between - should do as well as the stock active cooling. If u wnat to keep the stock cooling I would advise you to remove that thermal pad and apply some good silicone thermal compound (Arctic Ceramique or Cooler Master Performance).

Seems like i'm asking info instead of sharing more like i said in the first reply
Anywayz, wish me some luck and cheers m8

Yeah, it seems so, but I don't care - sharing experience is something really positive to me.

Good luck!

[chrisbard]

MietmaN did you try Folding@Home to see how well you PC performs???

[MietmaN]

Originally Posted by chrisbard

MietmaN did you try Folding@Home to see how well you PC performs???

No, I haven't tried Folding@Home yet - although some friends of mine wanted me to join their team.
I spend most of my time in Germany, where I've got a simple notebook, while my PC stays at home, in Poland.

Btw. I've just discovered that alpha memory timings change with the romsip table - they're more agressive in the FSB166 romsip table than in the FSB200 one.

[Wolf2000me]

Originally Posted by EPoX Tech

I have just put up da6p4a05.bin for you so you will now see the file upgrade as you wish.

Thanks, i noticed it's now the first one listed
Not that i couldn't get those bios files elswhere but it does belong on your site and it's greatly appreciated.

Also, ET you did mention to take a look at the ini file(s) of USDM. Have you made any progress? Just a status update, so far i am able to live with the fact that the program doesn't read my fan speeds correctly. Other programs however do read it correct so i assume it's probably no big deal adjusting it.
No rush of course

Btw, another thing
You said mosfeds are better cooled with decent air flow than just heatsinks, like the ones in the epox power pack. With that you suggested they were in fact ramsinks. So i put them on my videocards memory chips. Which seems to have done more harm than good. I assume the chips are better to be cooled with air? Or perhaps more likely, the sinks don't cover the chips completely, not even half as i remember. That might have had something to with it. Of course that leaves me with a bunch of heatsinks lying around so I am going to put them on my mosfeds again, on the places MietmaN suggested. That is if i find the time to power off my rig and actually do it. I'll report the differences i notice, if any.

[MietmaN]

I'm gonna post the exact results of performance increase in a couple of days, since my life consists of something more than computers

This is gonna be a comparison of 11x200 with FSB200 romsip table and FSB166 romsip table (at least at FSB200 seems stable so far) : as usual, Sandra 2004 memory bandwidth's benchmark, read and write memory tests in Everest Home, 3DMark2001SE to show the overall system performance plus SuperPI 2M and 16M. No other tests - I ain't got enough time and I've already done FSB166 11x166 (romsip FSB166) and FSB200 11x200 (romsip FSB200) benchmarks.

Now checking the stability at FSB200 with FSB166 romsip tables to make sure, that it works like it should

Anyone with the same experience (11x200 required)????

Drivers : nVidia 4.27 International Unified Drivers, ATI Catalyst 4.12.
OS : Win2k SP4 with all possible updates installed.

now to work ....

[Wolf2000me]

MietmaN, i'll install the unified drivers i found on guru3d. Filename is nforce-6xx-[Guru3D.com].exe. Of course, that won't do anything on the romsip table issue but it might give other results, improvements and such. I'd lend you a hand if i had the time to test for stability. Right now i need my pc for programming and to study operating systems. Like i said, i will probably have time after the 22nd. That is if you're not done by then. At least, if guys like us ever pronounce a system "done"
btw, the exam on datacommunications went well Next one will be OO-programming on next monday.

[MietmaN]

11x166, FSB166 ROMSIP table
V-core 1.30v, 2x512MB dual 2-3-3-11, Win2k SP4, nVidia unified drivers 4.27, Catalyst 4.12 (balanced settings), DirectX 9.0c

* Sandra 2004 SP1
Int : 2549 MB/s, 2552 MB/s (95-96% efficiency)
Float : 2393 MB/s, 2401 MB/s (90% efficiency)

* Everest Home
read - 2581 MB/s
write - 993 MB/s

* 3DMark 2001SE
16063, 16147, 16125

* SuperPI
2M - 02m 05s
16M - 23m 12s


11x200, FSB166 ROMSIP table
V-core 1.50v, 2x512MB dual 2-3-3-11, Win2k SP4, nVidia unified drivers 4.27, Catalyst 4.12 (balanced settings), DirectX 9.0c

* Sandra 2004 SP1
Int : 3070 MB/s, 3071 MB/s (96% efficiency)
Float : 2877 MB/s, 2891 MB/s (90% efficiency)

* Everest Home
read - 3108, 3112 MB/s
write - 1196, 1204 MB/s

* 3DMark 2001SE
18535, 18617, 18614, 18545, 18692

* SuperPI
2M - 01m 44s, 01m 44s
16M - 19m 2