errata 123 and its effects on memory bandwidth

**Major** · 01-14-2006, 11:00 PM

Just some quick tests on errata 123 and how it effects memory bandwidth.

Sandra with errata 123 DISABLED in bios

Sandra with errata 123 ENABLED Iin bios

Pi 4m with errata 123 DISABLED in bios

Pi 4m with errata 123 ENABLED in bios

**STEvil** · 01-14-2006, 11:04 PM

what is errata 123?

**eva2000** · 01-14-2006, 11:05 PM

interesting forgot what errata 123 was for ?

**demonbane23** · 01-14-2006, 11:07 PM

just posted this in another thread but this should give you some explanation eva2000:

I was just doing some reading on errata 123 after reading through this and it is suppose to deal with a bug found in some opteron and X2 chips dealing witht he L2 cache,,, if your CPU has it then it needs to be enabled to avoid hang ups or problems when trying to bypass the cache,, but if it doesnt have the bug you get better performance with the option disabled,,

here is the original context:

This BIOS feature is a workaround for the bug. It allows you to disable the cache bypass feature and avoid the bug from manifesting.

When enabled, the processor will not bypass the L2 cache to prefetch data from the system memory.

When disabled, the processor will continue to bypass the L2 cache for certain cache line fill requests. This improves its performance.

When set to Auto, the BIOS will query the processor to see if it is affected by the bug. If the processor is affected, the BIOS will enable this BIOS feature. Otherwise, it will leave it disabled.

If your processor is affected by this bug, you should enable this BIOS feature to prevent the processor from hanging or corrupting data. But if your processor is not affected by this bug, disable this BIOS feature for maximum performance.

**Major** · 01-14-2006, 11:09 PM

^^ yea what he said ! lol

It only applies to dual core, altho I have no way of testing if left on AUTO in bios if it will affect single core memory bandwidth or not.

I've not had any problems with it disabled, but some may so you just have to test !

**Harshal** · 01-14-2006, 11:11 PM

Well nice findings, Major. I am still wondering why it is enabled when you flash the BIOS? I have to restart.....hmmm to disable it

EDIT: Took TIME and you guys posted the findings.

**demonbane23** · 01-14-2006, 11:12 PM

Originally Posted by Major

^^ yea what he said ! lol

It only applies to dual core, altho I have no way of testing if left on AUTO in bios if it will affect single core memory bandwidth or not.

I've not had any problems with it disabled, but some may so you just have to test !

yeah i was trying to determine if you can test it on auto but who knows,,

anyway nice memory scores I must say,,, Im happy usually to break 7000,,,,

**STEvil** · 01-14-2006, 11:18 PM

is there a way we can test for this bug?

**dogsx2** · 01-14-2006, 11:19 PM

I just ran pi 1m 3x each with it on auto and disabled with my 144 and auto was faster then disabled. It was about 0.12 sec faster on auto. I never messed with it before as I did not know what it was for.

**Major** · 01-14-2006, 11:23 PM

Originally Posted by dogsx2

I just ran pi 1m 3x each with it on auto and disabled with my 144 and auto was faster then disabled. It was about 0.12 sec faster on auto. I never messed with it before as I did not know what it was for.

Interesting that it affects single core chips performace too

**dogsx2** · 01-14-2006, 11:28 PM

It's pretty easy to get 0.1 sec difference is 1m so I don't know if it is fact or just normal. 32m would be better and really show if it is faster or slower.

**ozzimark** · 01-14-2006, 11:35 PM

Originally Posted by Major

Interesting that it affects single core chips performace too

kinda makes sense, if it's simply bypassing the cache to go to the ram, it'll do the same if enabled or disable for either dual or single core cpu's

**Major** · 01-14-2006, 11:37 PM

Originally Posted by ozzimark

kinda makes sense, if it's simply bypassing the cache to go to the ram, it'll do the same if enabled or disable for either dual or single core cpu's

I was just under the impression that it was a X2/DC fix, didn't really think about it being enabled on single core.

**K.I.T.T.** · 01-14-2006, 11:38 PM

Under which options in the BIOS is this? I can't search now, as i'm still priming

**mongoled** · 01-15-2006, 12:59 AM

Originally Posted by K.I.T.T.

Under which options in the BIOS is this? I can't search now, as i'm still priming

aahhh tht feeling of waking up in the morning to find your PC is still priming away

Nice find Major, will give this a go with my Opteron 146 and report back.

**K.I.T.T.** · 01-15-2006, 01:06 AM

Originally Posted by mongoled

aahhh tht feeling of waking up in the morning to find your PC is still priming away

Nice find Major, will give this a go with my Opteron 146 and report back.

No....I woke up at 5.00 am, just to get start priming. Core 0 failed twice last night

First time after 1 hour, 26 mins and then i quit and restarted both cores and it failed again at 2 hours, 3 mins

Anyway, seems to be going well now

Imroved my PWM cooling a bit and its been dual priming for 3 hours 45 mins and still going strong!!

**CRUSH-ER** · 01-15-2006, 01:10 AM

I set this option to auto in the bios and it sticks....so I need it enabled. This is on a single core Opty 148. The difference is minimal, but hey, whatever.

**Fhqwhgads6680** · 01-15-2006, 01:49 AM

hmm this is interesting I'll give it a go on my 148 when I get home and post results, at a lan right now

**mongoled** · 01-15-2006, 03:05 AM

Well here are my results, settings in sig, msconfig used to disable all services except MS ones.

SuperPI
Errata123, Status: Auto

Errata123, Status: Disabled

Sandra
Errata123, Status: Auto

Errata123, Status: Disabled

My results are inconclusive, but as others have mentioned, we need to find out a way to see wht AUTO enables for our systems. Then we can see if there is a difference.

**CRUSH-ER** · 01-15-2006, 03:06 AM

According to that article, setting it to auto and then rebooting willl let you know if you need it. It accesses the cpu and if it needs it, it'll stay auto.

**linhvndiy** · 01-15-2006, 04:04 AM

Hi! What is your cpu? Real 280?
How to OC to 32xx? And what mainbaord you use?
Thanks.

**Major** · 01-15-2006, 09:05 AM

Originally Posted by linhvndiy

Hi! What is your cpu? Real 280?
How to OC to 32xx? And what mainbaord you use?
Thanks.

Opteron 180 Engineering Sample, just programed as a 280 since it came out b4 the release of the 180

See sig for rig details

**ben805** · 01-15-2006, 09:17 AM

even running on single core, I'd found the Errata 123 Enhancement stabilized my Opty 148 at higher overclock, do a search on Microsoft database it has a white paper explaining the relation between this feature and the Windows kernel...

by the way, mine take a bigger bandwidth hit to have it enable on XP SP2 wherea not as much with SP1a

**alpha0ne** · 01-17-2006, 10:53 PM

Very interesting and thanks for the heads up, I'll try a few tests and see if there are any meaningful differences

**DltdApS** · 01-18-2006, 09:22 AM

where is this option in the bios ?

Thread: errata 123 and its effects on memory bandwidth

Thread Tools

Search Thread

Rate This Thread

Display

errata 123 and its effects on memory bandwidth

Bookmarks

Bookmarks

Posting Permissions