Enthusiast fixes what AMD won't (Phenom II + XP)

[swaaye]

I don't know how many people are aware of this, but Phenom II still is a dog in XP. Vista works better with the new locking of the separate core clocks, but XP still has individual core clocks for CnQ. This causes severe performance issues for all single-processor apps. I noticed myself with Dark Messiah, where load times doubled or so compared to Vista and in-game action strangely stuttered.

There was a bunch of hoopla about how awesomely Phenom II fixed this problem. Nobody really noticed that the fix was only in Vista.

Anyway, I've been following this problem for a few months now. There are some utilities popping up that let you tweak the processor registers yourself. Literally reprogram what CnQ does.

This one (PhenomMSRTweaker) seems to help with XP
http://www.xtremesystems.org/forums/showthread.php?t=220185

There's another one called K10Stat. This rocks for Black Edition owners who want to overclock by multi and still have CnQ working.
http://www.geocities.com/k10stat/

It is rather annoying to say the least that AMD either can't or won't fix the issue. Their processor driver for Phenom 1/2 in XP is from 2007. This kind of nonsense is a real big reason to not buy a Phenom, IMO, regardless of how awesomely they are priced.

 

[homerdog]

If CNQ is so fucked, why don't we hear more about it? I read that CnQ on the Phenoms was a disaster in the AnandTech review a while back, but since they didn't mention it in the Phenom II review I figured it was fixed. Unless... ah... they only tested in Vista

Meanwhile other sites never even mentioned the problem

 

[BRiT]

Why should AMD support XP when the original company doesn't?

 

[I.S.T.]

Why should AMD support XP when the original company doesn't?

Because that just happened yesterday. PIIs have been out for more than just one day. Plus, XP is still mega popular, especially with all the Vista FUD out. Not only that, but people supported Windows 98SE/ME in 2003, two years after XP came out. Why shouldn't companies support XP two years after Vista came out?

Your logic is at best flawed.

 

[Bludd]

And AMD could still pay for the fix and have MS release it for all their users.

 

[swaaye]

Yeah AMD still releases driver packs for all their chipsets and GPUs for XP and XP 64. So, why not fix that ~2 year old CPU driver too?

 

[BRiT]

It's not my logic; I was playing devil's advocate.

Some say it's not about how long an OS is supported after the replacement is out, it's how long the OS is supported since initial release. XP has been out since 2001. Supporting Win98SE/ME in 2003 is 5 years later at longest. Supporting XP now is 8 years later.

I think AMD should support their consumers by supporting their processors in all OS's. Then again, AMD shafted consumers in a quicker timeframe with their lack of continued support on Ssocket939 so this here doesn't surprise me the least.

 

[Silent_Buddha]

I think AMD should support their consumers by supporting their processors in all OS's. Then again, AMD shafted consumers in a quicker timeframe with their lack of continued support on Ssocket939 so this here doesn't surprise me the least.

Yeah just like Intel shafted early adopters of socket 775 among other sockets. And let's not bring in Socket 4 that only worked for Pentium 60/66. Or Socket 6 that was barely even used. How about Socket 1, 2, 3, and 6 all for various 486 CPUs. If you want something more recent, how about Socket 423.

Socket 939 had an very long lifespan in comparison to many AMD/Intel sockets. And supported quite the variety of processors. And definitely lasted longer than say Slot A.

Unless there's some architectural difference in how the OS supports CnQ between XP and Vista, I agree they should get it to work there also.

Regards,
SB

 

[Brad Grenz]

People should probably just get over themselves and run a modern OS.

 

[Scali]

Didn't AMD release a 'dualcore optimizer' for XP that fixed this?
Does that not work on Phenoms anymore? (granted, they should at least rename it 'quadcore optimizer' ).

 

[green.pixel]

People should probably just get over themselves and run a modern OS.

That's hardly an excuse.

 

[DJ12]

Didn't AMD release a 'dualcore optimizer' for XP that fixed this?
Does that not work on Phenoms anymore? (granted, they should at least rename it 'quadcore optimizer' ).

I read that it doesn't work with phenoms at all.

A quick google search seems to confirm it too.

 

[swaaye]

People should probably just get over themselves and run a modern OS.

Or run a CPU that was designed and tested properly. We're not talking a minor bit of inconvenience here. It dramatically slows apps down. Threads can end up running on a core that's clocked at 800MHz as the XP scheduler moves them around.

Besides, who wants to own Vista anymore with Win7 looming? And who wants to trust Win7 beta for anything significant? Being a voluntary MS beta tester bitch doesn't appeal to everyone.

Frankly I'm surprised that this crazy CnQ issue doesn't get people up in arms ala the TLB bug. The problem was even worse with the original Phenom cuz Vista was affected as well. Add this issue together with the TLB BIOS workaround gimpage and you're running with an imaginary quad core Athlon Tbird in the performance dept.

 

[swaaye]

Didn't AMD release a 'dualcore optimizer' for XP that fixed this?

No that fixed their timestamp counter problems with Athlon 64 X2. At least they fixed that though.

 

[swaaye]

Yeah just like Intel shafted early adopters of socket 775 among other sockets. And let's not bring in Socket 4 that only worked for Pentium 60/66. Or Socket 6 that was barely even used. How about Socket 1, 2, 3, and 6 all for various 486 CPUs. If you want something more recent, how about Socket 423.

Yeah the folks who hate on AMD for their supposed shafting of Socket 939 have seriously selective memory or short term memory. Ah well.

My biggest qualm with 939 was how unreliable the memory controllers were. It was not uncommon to have serious problems when running 4 DIMMs. Although I suppose that's true of recent mobos+CPUs+chipsets too.

 

[BRiT]

I haven't had any issues after switching to Intel chipsets with running 4 dimms on my P35 motherboard. Once I replaced some faulty dimms I didn't have issues running on a P965 either.

As for selective memory, I bitched equally as much on other artificially limiting situations. At least on the Intel front, they had socket-370 to slot-1 adapters which worked great. When AMD abandoned Socket939, their Socket-AM offered lower performance, while Intel always provided higher performance.

 

[Blazkowicz]

it's almost identical performance, same as northwood vs prescott which was probably worse (worst CPU ever is pentium 4 1.4GHz with rdram, there was that failure with pentium fsb 133 and rdram chipset too.)

but hell if we're going that way I still don't forgive ATI for its lackluster ATI Rage Pro Turbo 2 (which I didn't even own)
I believe Intel managed to get three versions of socket 370.

 

[Scali]

No that fixed their timestamp counter problems with Athlon 64 X2. At least they fixed that though.

Yea, but I thought this was basically the same thing... Different cores running at different speeds, and as such, timestamp counters updated at different rates.

 

[swaaye]

A64 cores don't clock independently of each other. The TSC thing was just a early multi-core bug. XP actually got a new method of managing that counter with a hotfix, called the "pmtimer". So I think you can skip the dual core optimizer entirely as long as you have a recent XP kernel version and run it with /usepmtimer.

Phenom's problem is related to Cool'n'quiet not properly clocking cores in response to XP's thread scheduler changing thread affinity. It seems that AMD's fancy-and-highly-touted independent core clocking never worked right, regardless of OS. In Vista they made Phenom 2 work like every other CPU out there in that the cores all run at the same speed. In XP, the cores are independent still and it doesn't work right.

 

[swaaye]

I haven't had any issues after switching to Intel chipsets with running 4 dimms on my P35 motherboard. Once I replaced some faulty dimms I didn't have issues running on a P965 either.

As for selective memory, I bitched equally as much on other artificially limiting situations. At least on the Intel front, they had socket-370 to slot-1 adapters which worked great. When AMD abandoned Socket939, their Socket-AM offered lower performance, while Intel always provided higher performance.

I've been going through some non-POSTing joy with a friend's Abit IP35 and 4 DIMMs. Whether or not you have problems really comes down to the quality of the DIMMs. Stick with the 1.8v DIMMs for the best luck IMO.

My confusion with Socket 939-related AMD hate is caused by my having moved on from it in 2006. AMD wasn't going to port Phenom over to 939, so K8x2 was all it would ever have. Once Core 2 Duo came out, getting more K8x2 chips was anything but exciting for such an expensive platform. RAM costs a bucketload for it if you want lots of it. 939 was only interesting for a short while IMO (a year or two) and that's not really entirely related to AMD moves. Right now, AM2+ is actually way more interesting and carries more value than 939 ever did IMO, with access to $40 dual cores and $40 4gig kits. It's a bitchin' low cost platform today.