Bizarre AMD Athlon XP 3200+ problem... really 2500+ or less?

[Steve Kives]

I had a very strange experience lately with a 3200+...

I ordered the retail package from Newegg.com, no problems, the package
(unopened, unmolested) said "3200+ 400MHz FSB!" and the part no. on the
specimen itself even had "3200" in it, so it all checked out. However,
the computer actually got slower (I'm using a 3000+/333MHz FSB).
Multiple utilities (including the BIOS) showed that the cpu was
reporting at "2500+", and the Sandra Lite arithmetic benchmark pegged it
as comparable to a 2200+!

The present cpu checks out consistently at 3000+ w/ 333MHz FSB, with all
the utilities, and the benchmarks bear this out. No crashes (besides
apps and MS crap), and memtest86 runs for 13 hours without a hitch. (I
tried to run the Sandra Lite "burn-in" test but it crashes the app with
some message about the chipset not being supported... maybe I have to
spend money on the full Sandra or such.)

Has anyone else seen this weirdness? I have to assume there was a snafu
at the AMD factory, but I wonder if others have had similar problems.

I RMA'ed it back to newegg without a problem, except for having to pay
return shipping, which sucks, but newegg does everything else right and
it's a small cost so I won't worry about it.


For the curious, here are the details:

ASUS A7V880 (BIOS AMI 1007.005 or 08.00.09, depending where you look)
AMD Athlon XP 3000+ (Barton, 2166MHz, 333MHz FSB effective)
2 x 512 MB TwinMOS PC3200 (DDR 400)
ASUS AX800PRO (ATI Radeon X800 PRO)
SoundBlaster Audigy 2 ZS
Linksys NIC
Lian Li aluminum case
Enermax EG465P-VE


Utilities used:

AMD CPUInfo
CPU-Z
FreshDiagnose
Sandra Lite
Everest


Thanks,
-SK

 

[sorchu_bf]

Hi there.

Does your motherboard supports FBS 400mhz?

I am not surprised if your new 3200+ cpu runs slower at 333mhz...
because you have to run the CPU to its designed FBS + the right
multiplier in order to run it at top speed...

Kev

I had a very strange experience lately with a 3200+...

I ordered the retail package from Newegg.com, no problems, the package
(unopened, unmolested) said "3200+ 400MHz FSB!" and the part no. on the
specimen itself even had "3200" in it, so it all checked out. However,
the computer actually got slower (I'm using a 3000+/333MHz FSB).
Multiple utilities (including the BIOS) showed that the cpu was
reporting at "2500+", and the Sandra Lite arithmetic benchmark pegged it
as comparable to a 2200+!

The present cpu checks out consistently at 3000+ w/ 333MHz FSB, with all
the utilities, and the benchmarks bear this out. No crashes (besides

 

[chocolatemalt]

Hi there.

Does your motherboard supports FBS 400mhz?

I am not surprised if your new 3200+ cpu runs slower at 333mhz...
because you have to run the CPU to its designed FBS + the right
multiplier in order to run it at top speed...

Kev

It definitely supports 400MHz FSB and has the latest BIOS. The RAM also
is ready for it.

What's more, I believe the "AMD Athlon XP 2500+" that was getting
reported by the cpu (now en route back to newegg) is a text string
burned into the cpu itself, so something was certainly fishy.

thanks,
-SK

 

[Zotin Khuma]

I had a very strange experience lately with a 3200+...

I ordered the retail package from Newegg.com, no problems, the package
(unopened, unmolested) said "3200+ 400MHz FSB!" and the part no. on the
specimen itself even had "3200" in it, so it all checked out. However,
the computer actually got slower (I'm using a 3000+/333MHz FSB).
Multiple utilities (including the BIOS) showed that the cpu was
reporting at "2500+", and the Sandra Lite arithmetic benchmark pegged it
as comparable to a 2200+!

The present cpu checks out consistently at 3000+ w/ 333MHz FSB, with all
the utilities, and the benchmarks bear this out. No crashes (besides
apps and MS crap), and memtest86 runs for 13 hours without a hitch. (I
tried to run the Sandra Lite "burn-in" test but it crashes the app with
some message about the chipset not being supported... maybe I have to
spend money on the full Sandra or such.)

Has anyone else seen this weirdness? I have to assume there was a snafu
at the AMD factory, but I wonder if others have had similar problems.

I RMA'ed it back to newegg without a problem, except for having to pay
return shipping, which sucks, but newegg does everything else right and
it's a small cost so I won't worry about it.


For the curious, here are the details:

ASUS A7V880 (BIOS AMI 1007.005 or 08.00.09, depending where you look)
AMD Athlon XP 3000+ (Barton, 2166MHz, 333MHz FSB effective)
2 x 512 MB TwinMOS PC3200 (DDR 400)
ASUS AX800PRO (ATI Radeon X800 PRO)
SoundBlaster Audigy 2 ZS
Linksys NIC
Lian Li aluminum case
Enermax EG465P-VE


Utilities used:

AMD CPUInfo
CPU-Z
FreshDiagnose
Sandra Lite
Everest


Thanks,
-SK

A month or so ago, I saw something that may be connected
with this at an online auction site in India. I'm not sure about
the details, but the seller offered an XP 3200+ CPU about
which he said something like its not being a legitimate AMD
product that had been seized in Singapore. He said that it
was not stable at its rated speed and could be run only at
a lower speed. That's all I can remember.

 

[kony]

I had a very strange experience lately with a 3200+...

I ordered the retail package from Newegg.com, no problems, the package
(unopened, unmolested) said "3200+ 400MHz FSB!" and the part no. on the
specimen itself even had "3200" in it, so it all checked out. However,
the computer actually got slower (I'm using a 3000+/333MHz FSB).

What exactly do you mean, "I'm using a 3000+/333MHz" ?

Multiple utilities (including the BIOS) showed that the cpu was
reporting at "2500+", and the Sandra Lite arithmetic benchmark pegged it
as comparable to a 2200+!

Back up and do the basic normal things!
Check the motherboard manual. Look for any CPU speed or
FSB speed jumpers. Check the bios menus. You may need to
set the FSB yourself, it is common for a motherboard to
first default to a lower FSB speed.

The present cpu checks out consistently at 3000+ w/ 333MHz FSB, with all
the utilities, and the benchmarks bear this out.

"Checks out"? What exactly does this mean. You are being
vague when details are most important. Nothing you wrote
matters but the details you didn't write, do. I'm not
trying to be hard on you, rather getting at the point in
shortest manner possible.

Forget about benchmarks until you can be sure you have CPU
and FSB running at correct frequencies and multiplier.

No crashes (besides
apps and MS crap), and memtest86 runs for 13 hours without a hitch. (I
tried to run the Sandra Lite "burn-in" test but it crashes the app with
some message about the chipset not being supported... maybe I have to
spend money on the full Sandra or such.)

Has anyone else seen this weirdness? I have to assume there was a snafu
at the AMD factory, but I wonder if others have had similar problems.

I have no reason to believe there is anything wierd.
Benchmarks are not the way to determine if the CPU is
genuine, they are a performance level. Put a CPU in a board
with loose bios settings, poorly defaulted or changed
underclocked asynchronous bus and an OS problem or two, and
performance will vary. Benchmark apps generally compare to
optimal results, putting parts in the best light possible,
which in this case might mean a tweaked nForce2 chipset and
relatively expensive memory. Maybe there was some CPU
mixup, but it's not certain, subject to question based only
on the info provided.

I RMA'ed it back to newegg without a problem, except for having to pay
return shipping, which sucks, but newegg does everything else right and
it's a small cost so I won't worry about it.

It is not impossible for a CPU to be counterfeit, but I've
not heard of any coming from newegg. I suspect you'll get
another CPU and be faced with same situation, that you need
to change some settings else the benchmarks were simply
painting a deceptive picture.

For the curious, here are the details:

ASUS A7V880 (BIOS AMI 1007.005 or 08.00.09, depending where you look)
AMD Athlon XP 3000+ (Barton, 2166MHz, 333MHz FSB effective)

So is it a 3000+ or a 3200+?

 

[chocolatemalt]

What exactly do you mean, "I'm using a 3000+/333MHz" ?

Sorry if this was unclear. I'm using, right now, a 3000+/333MHz. It
has run (and continues to run) solidly. I bought this several months
ago thinking I would never upgrade my RAM and therefore didn't need the
400MHz FSB model, which cost another $40 or so at the time. Anyhow,
that was a big mistake, I did upgrade the RAM from 512 MB PC2700 to 1 GB
PC3200, and was pissed at myself for getting stuck with a cpu that
couldn't fully take advantage of the new RAM. Hence, the attempt at the
upgrade to 3200+/400MHz.

Back up and do the basic normal things!
Check the motherboard manual. Look for any CPU speed or
FSB speed jumpers. Check the bios menus. You may need to
set the FSB yourself, it is common for a motherboard to
first default to a lower FSB speed.

I did do the basic normal things, and mentioned it (albeit briefly) by
including "BIOS" in my "utilities" list, but I didn't want to bog down
the newsgroup post with too much boring detail. But, I'll do that now.


I really haven't deviated from the default set-up. Here are the
particulars from the BIOS menus, hopefully all the ones that matter:

CPU Multiple: Auto
AI Overclock Tuner: Standard
Spread Spectrum: Enable
Asynchronous Frequency: Disabled
VCORE Voltage: Auto
DDR Reference Voltage: Auto
AGP Reference Voltage: Auto

VIA KT880 northbridge:
DRAM clock: Auto
DRAM timing: Auto by SPD
DRAM BUS selection: Dual Channel

Only the last setting, forcing "Dual Channel", is something I've changed
in the course of some earlier investigation. I could change it back to
default "Auto" but I don't think it should care.

I've never overclocked this board or played with voltages, not yet
anyway.

As for the motherboard (A7V880), there are very few jumpers to play
with: clear CMOS; wake-on-keyboard; wake-on-usb-devices; and
cpu-overvoltage. All of these are disabled, which is also the default
config.

"Checks out"? What exactly does this mean. You are being
vague when details are most important. Nothing you wrote
matters but the details you didn't write, do. I'm not
trying to be hard on you, rather getting at the point in
shortest manner possible.

Forget about benchmarks until you can be sure you have CPU
and FSB running at correct frequencies and multiplier.

Good suggestions, but I think I lost you when I didn't make clear that
the *present* cpu is a 3000+/333, and the *new* cpu is an
apparently-mislabeled 3200+/400. Without having that info, everything
else looks like mush.

I think I have all the motherboard and BIOS settings right, and the BIOS
is very recent. If you see otherwise, please let me know.

By "checks out" I mean that the present 3000+/333 reports consistent
info about itself in the BIOS and all the other utilities I listed, and
the Sandra benchmark (very basic integer test) compares it almost
exactly with Sandra's reference 3000+/333. This proves nothing, but
it's a lot of corroborating data. Again, if you have other preferred
utilities, or some of the ones I used are for shit, definitely let me
know.

It is not impossible for a CPU to be counterfeit, but I've
not heard of any coming from newegg. I suspect you'll get
another CPU and be faced with same situation, that you need
to change some settings else the benchmarks were simply
painting a deceptive picture.

If it was counterfeit, the crooks did an awesome job on the packaging,
sturdy (thick heavy copper base) heatsink, cert of authenticity,
documentation, etc. The whole package really seemed perfectly legit.

My pet theory is that AMD had a factory screwup and mislabeled a 2500+.
But this doesn't explain why the 2500+ benchmarked as equivalent to a
2200+ on a very simple cpu test. I should've run more simple benchmarks
in other utilities, just for curiosity's sake, but didn't think of it
and it's too late now.

So is it a 3000+ or a 3200+?

Both! I mean... which one?


A final question -- CPU-Z is reporting the following (on my *present*
cpu):

Specification: AMD Athlon(tm) 3000+
Speed: 2166 MHz
Multiplier: x13.0

My understanding is that the "specification" is a string hard-coded into
the cpu itself. Can anyone verify this? If so, the "AMD Athlon(tm)
2500+" I was reading off the pretender "3200+" cpu was a pretty
condemning piece of evidence.


Cheers, and thanks for the help...

 

[John]

CPU Multiple: Auto
AI Overclock Tuner: Standard
Spread Spectrum: Enable
Asynchronous Frequency: Disabled
VCORE Voltage: Auto
DDR Reference Voltage: Auto
AGP Reference Voltage: Auto


My understanding is that the "specification" is a string hard-coded into
the cpu itself. Can anyone verify this? If so, the "AMD Athlon(tm)
2500+" I was reading off the pretender "3200+" cpu was a pretty
condemning piece of evidence.


Cheers, and thanks for the help...

Someone was posting something about that but Im not sure if it was
with the bartons. I know with my 2500 XP it would come out as a 3200
when I cranked my bus up to 200.

You are saying it shows 2500 so its getting the string so it MUST be a
2500. I dont know about that. And you are saying you have it on
default and it comes out as a 2500 so it must be a 2500.

With my older nforce2 system in fact all my boards recently it didnt
quite detect the CPU accurately most of the time in fact thats what
caused it to hang many times and required a clearing of the CMOS.

With your system under
ADVANCED AI overclock tuner you should have checked the FSB.
If it wasnt 200 and only 166 then it accounts for the 2500 - IF that
is the string thing you mention isnt true. Somehow I cant picture AMD
making a mistake and putting a 2500 in there but who knows.

 

[kony]

Sorry if this was unclear. I'm using, right now, a 3000+/333MHz. It
has run (and continues to run) solidly. I bought this several months
ago thinking I would never upgrade my RAM and therefore didn't need the
400MHz FSB model, which cost another $40 or so at the time. Anyhow,
that was a big mistake, I did upgrade the RAM from 512 MB PC2700 to 1 GB
PC3200, and was pissed at myself for getting stuck with a cpu that
couldn't fully take advantage of the new RAM. Hence, the attempt at the
upgrade to 3200+/400MHz.

You haven't lost out on much, a few % if that. Hardly worth
the cost to upgrade especially since the highest speed parts
(in this case the 3200) are always disproportionately
higher priced.

I did do the basic normal things, and mentioned it (albeit briefly) by
including "BIOS" in my "utilities" list, but I didn't want to bog down
the newsgroup post with too much boring detail. But, I'll do that now.

Ok, I didn't realize this. It is significant though, to
first and foremost continue to focus on the bios settings to
ensure the correct multiplier and FSB are set. This is to
ensure the reduction of variables, so once you are sure
it's at correct speed,the rest of the variables can be
addressed. If you did happen to have the wrong speed of
CPU but it was running at correct multliplier and FSB for
what you "thought" you'd bought, it would have same
performance. Actually, if your current CPU is not
multiplier locked, it should be easy enough to just change
the FSB speed and multiplier without having to buy another
CPU at all... but again, the difference isn't much.

I really haven't deviated from the default set-up. Here are the
particulars from the BIOS menus, hopefully all the ones that matter:

CPU Multiple: Auto

Then try manual, setting what it should be.

AI Overclock Tuner: Standard
Spread Spectrum: Enable

Disable that, it's never good except for specific
circumstances where outside equipment is picking up
interference.

Asynchronous Frequency: Disabled
VCORE Voltage: Auto
DDR Reference Voltage: Auto
AGP Reference Voltage: Auto

VIA KT880 northbridge:
DRAM clock: Auto

Don't use auto. Set it to same speed as the CPU FSB, or
maybe it's worded as "100%", it varies per board.

DRAM timing: Auto by SPD
DRAM BUS selection: Dual Channel

Only the last setting, forcing "Dual Channel", is something I've changed
in the course of some earlier investigation. I could change it back to
default "Auto" but I don't think it should care.

If it tests stable in dual channel mode, leave it enabled.

I've never overclocked this board or played with voltages, not yet
anyway.

As for the motherboard (A7V880), there are very few jumpers to play
with: clear CMOS; wake-on-keyboard; wake-on-usb-devices; and
cpu-overvoltage. All of these are disabled, which is also the default
config.

It does seem to have all necessary settings in the bios, but
sometimes the bios settings are limited based on the jumpers
too... which doesn't seem to be the case but I dont' have
that board, can't see your bios.

Good suggestions, but I think I lost you when I didn't make clear that
the *present* cpu is a 3000+/333, and the *new* cpu is an
apparently-mislabeled 3200+/400. Without having that info, everything
else looks like mush.

yes that clarifies things.

I think I have all the motherboard and BIOS settings right, and the BIOS
is very recent. If you see otherwise, please let me know.

I would still try manually setting what you know to be
correct.

By "checks out" I mean that the present 3000+/333 reports consistent
info about itself in the BIOS and all the other utilities I listed, and
the Sandra benchmark (very basic integer test) compares it almost
exactly with Sandra's reference 3000+/333. This proves nothing, but
it's a lot of corroborating data. Again, if you have other preferred
utilities, or some of the ones I used are for shit, definitely let me
know.

I'd not rely on Sandra though, it only knows hardware that
preceeded it's (particular version) release. When dealing
with newer hardware it can sometimes be misleading, though
perhaps it's better on newest versions, I dont' use it to ID
CPUs.

If it was counterfeit, the crooks did an awesome job on the packaging,
sturdy (thick heavy copper base) heatsink, cert of authenticity,
documentation, etc. The whole package really seemed perfectly legit.

The typical counterfeit would be a genuine AMD CPU,
box/heatrsink/etc, that they'd simply manipulated the
bridges on so it ran at different muliplier and/or FSB, and
perhaps changed the voltage... with a casual observation it
would look fine, though perhaps the label looked off but
only compared to the genuine label. I doubt this is the
case though from Newegg's parts, they have never been known
to have any counterfeit parts AFAIK.

My pet theory is that AMD had a factory screwup and mislabeled a 2500+.
But this doesn't explain why the 2500+ benchmarked as equivalent to a
2200+ on a very simple cpu test. I should've run more simple benchmarks
in other utilities, just for curiosity's sake, but didn't think of it
and it's too late now.

I think it was running at the wrong FSB, not that it was a
2500+.

Both! I mean... which one?

To determine correct MHz, Google search a CPU type (barton)
and XP(nnnn) speed. A Barton @ 2166 would be an XP3000.

A final question -- CPU-Z is reporting the following (on my *present*
cpu):

Specification: AMD Athlon(tm) 3000+
Speed: 2166 MHz
Multiplier: x13.0

My understanding is that the "specification" is a string hard-coded into
the cpu itself. Can anyone verify this? If so, the "AMD Athlon(tm)
2500+" I was reading off the pretender "3200+" cpu was a pretty
condemning piece of evidence.

I think CPU-Z simply consults an internal chart and based on
the FSB speed, misreported the CPU. I think the odds are
overwhelming that the only problem you had was that the
motherboard was running @ 166MHz FSB instead of 200MHz FSB,
that it was in fact a genuine XP3200 Barton you returned.

So, when you get the new CPU, set the FSB to 200MHz
(DDR400).

 

[chocolatemalt]

You haven't lost out on much, a few % if that. Hardly worth
the cost to upgrade especially since the highest speed parts
(in this case the 3200) are always disproportionately
higher priced.

Agreed. But I'm an avid gamer and lots of poring through benchmarks at
tomshardware.com showed me that I could expect a 3%-15% performance
improvement (very dependent on game or benchmark) just with the FSB
jump, and the 3000+ -> 3200+ was another $40 for another couple %, so I
figured what the heck, since I was going to be getting the 3000+/400 at
the very least. And anyway, upgrading is fun... right?

I'll overclock the 3200+ too, once I actually get one.

Ok, I didn't realize this. It is significant though, to
first and foremost continue to focus on the bios settings to
ensure the correct multiplier and FSB are set. This is to
ensure the reduction of variables, so once you are sure
it's at correct speed,the rest of the variables can be
addressed. If you did happen to have the wrong speed of
CPU but it was running at correct multliplier and FSB for
what you "thought" you'd bought, it would have same
performance. Actually, if your current CPU is not
multiplier locked, it should be easy enough to just change
the FSB speed and multiplier without having to buy another
CPU at all... but again, the difference isn't much.

With the 3200+/400 working as expected, then overclocked, all those
marginal improvements should add up to me being the terror of the online
shoot-em-ups.

Then try manual, setting what it should be.

"Auto" is working flawlessly and as expected with the 3000+/333, but
I'll definitely adjust this if the new 3200+ gives me problems. And
when I overclock it, of course.

Disable that, it's never good except for specific
circumstances where outside equipment is picking up
interference.

Interesting! I'd never run into this before on prior motherboards, nor
gotten around to researching it until just now. I doubt my non-windowed
aluminum case needs help holding back EMI, so I'll take your advice.

I gather the "Enable" setting is just fine for non-overclocked machines,
at some discernable benefit to monitors, etc, else ASUS wouldn't make it
the default. Or maybe that's too trusting of me.

I'd not rely on Sandra though, it only knows hardware that
preceeded it's (particular version) release. When dealing
with newer hardware it can sometimes be misleading, though
perhaps it's better on newest versions, I dont' use it to ID
CPUs.

Nor did I -- I used AMD's own "CPUInfo" and "CPU-Z", which seemed to be
more complete. Again, all info on both the present cpu and that
hypothetical "3200+" reported identical info in those utilities, as well
as the BIOS boot screen.

I think CPU-Z simply consults an internal chart and based on
the FSB speed, misreported the CPU. I think the odds are
overwhelming that the only problem you had was that the
motherboard was running @ 166MHz FSB instead of 200MHz FSB,
that it was in fact a genuine XP3200 Barton you returned.

So, when you get the new CPU, set the FSB to 200MHz
(DDR400).

The problem I have with your theory is that it wasn't just CPU-Z, but
also AMD CPUInfo and the BIOS itself that all believed it was a "AMD
Athlon(tm) 2500+". So they would all have to be using the same trick
to come up with a dynamic value rather than the "real" model number of
the cpu.

But you could be right... I'm no expert in the PC world, just a
hobbyist, and too many years in the big iron unix field may have made me
too optimistic in my hardware diagnostics. I'm also surprised that
"Auto" is not to be trusted on these ASUS boards, based on your advice.
My current cpu works fine with the "Auto" values and reports correct
info about its model no., so I'll be surprised if it turns out that the
3200+/400 doesn't behave as politely. (Why would it lie? Is it evil?)

In any case, I'll post about how the replacement works out.

Thanks...

 

[chocolatemalt]

Someone was posting something about that but Im not sure if it was
with the bartons. I know with my 2500 XP it would come out as a 3200
when I cranked my bus up to 200.

You are saying it shows 2500 so its getting the string so it MUST be a
2500. I dont know about that. And you are saying you have it on
default and it comes out as a 2500 so it must be a 2500.

At this point I believe you (and kony) are right, the "AMD Athlon(tm)
2500+" (or whatever) string is dynamically determined by the BIOS, or
any other utility that wants to determine the processor type. There in
fact appears to be no way to find this info from within any OS, only the
current operating speed and whatever that corresponds to in a table.

I was playing with the multipliers and FSB speeds earlier tonight,
managing to hose up my box a few times and needing a hard reboot to get
the BIOS back, and with all settings wiped clean. In the process, it
would often come up as "AMD Athlon(tm) 1800+", clearly not in tune with
the "3000+" that it is. So, the number doesn't reflect the actual
version.

The cpu utilities (and probably the BIOS) use the CPUID and RDTSC
instructions to figure out all the processor attributes and real-time
running speed. Here are some details:

http://www.sandpile.org/ia32/cpuid.htm
http://www.paradicesoftware.com/specs/cpuid/

As shown by all the fields buried in the CPUID specs, there is ample
room for tons of info, including strings such as "AuthenticAMD" and
"It's Hammer Time!", but AMD doesn't store the marked speed anywhere. I
thought for a moment that the stepping code would have this info, but
it's only obliquely related.

So, that's an education. It's also disappointing that the 3200+ choked
on the Auto settings in the BIOS, but I'll figure that out soon enough.

With my older nforce2 system in fact all my boards recently it didnt
quite detect the CPU accurately most of the time in fact thats what
caused it to hang many times and required a clearing of the CMOS.

With your system under
ADVANCED AI overclock tuner you should have checked the FSB.
If it wasnt 200 and only 166 then it accounts for the 2500 - IF that
is the string thing you mention isnt true. Somehow I cant picture AMD
making a mistake and putting a 2500 in there but who knows.

The AI Overclock was/is set to "Auto", like most everything else in the
BIOS... the default settings. Currently the FSB is at 166,
"effectively" 333, but I don't remember what it was running at with the
3200+.

Thanks...

 

[kony]

I'll overclock the 3200+ too, once I actually get one.

Then you're buying the wrong CPU.
Get a mobile Barton XP2500. Manually set the multiplier to
around 11X, the FSB to around 220 (if your memory will
tolerate it), and voltage roughly 1.7V... give or take 10%
on all variables.

With the 3200+/400 working as expected, then overclocked, all those
marginal improvements should add up to me being the terror of the online
shoot-em-ups.

Ok, so long as you're happy with it. Frankly I'd consider
an Athlon 64 plus new motherboard, then selling the old
board or building a 2nd system.

"Auto" is working flawlessly and as expected with the 3000+/333, but
I'll definitely adjust this if the new 3200+ gives me problems. And
when I overclock it, of course.

Maybe it's flawless, or maybe it just does 166 ok and it's
coincidence that your present CPU is 166 FSB.

Interesting! I'd never run into this before on prior motherboards, nor
gotten around to researching it until just now. I doubt my non-windowed
aluminum case needs help holding back EMI, so I'll take your advice.

Quite a few have the setting, it's not just for overclocking
but anyone/everyone would be as well off disabling it,
except as I mentioned previously that a very rare
interference might occur, but intermittent interference is
actually greater with it enabled.

I gather the "Enable" setting is just fine for non-overclocked machines,
at some discernable benefit to monitors, etc, else ASUS wouldn't make it
the default. Or maybe that's too trusting of me.

OEMs like "Enabled", they have to meet EMI emission
mandates.

The problem I have with your theory is that it wasn't just CPU-Z, but
also AMD CPUInfo and the BIOS itself that all believed it was a "AMD
Athlon(tm) 2500+". So they would all have to be using the same trick
to come up with a dynamic value rather than the "real" model number of
the cpu.

Not really a trick, just same method. This is normal,
expected. Having a motherbaord default to too low a FSB is
an extremely common situation. In fact many many boards
default to the lowest FSB they support regardless of what
CPU is installed, and the user or technician setting up the
system must then set the FSB, IF the board doesn't do so
correctly, automatically.

But you could be right... I'm no expert in the PC world, just a
hobbyist, and too many years in the big iron unix field may have made me
too optimistic in my hardware diagnostics. I'm also surprised that
"Auto" is not to be trusted on these ASUS boards, based on your advice.
My current cpu works fine with the "Auto" values and reports correct
info about its model no., so I'll be surprised if it turns out that the
3200+/400 doesn't behave as politely. (Why would it lie? Is it evil?)

In any case, I'll post about how the replacement works out.

Thanks...

A later bios update might address it. Considering all the
things that could be wrong with a board, that's a pretty
minor thing, usually.

 

[chocolatemalt]

At this point I believe you (and kony) are right, the "AMD Athlon(tm)
2500+" (or whatever) string is dynamically determined by the BIOS, or
any other utility that wants to determine the processor type. There in
fact appears to be no way to find this info from within any OS, only the
current operating speed and whatever that corresponds to in a table.

I was playing with the multipliers and FSB speeds earlier tonight,
managing to hose up my box a few times and needing a hard reboot to get
the BIOS back, and with all settings wiped clean. In the process, it
would often come up as "AMD Athlon(tm) 1800+", clearly not in tune with
the "3000+" that it is. So, the number doesn't reflect the actual
version.

The cpu utilities (and probably the BIOS) use the CPUID and RDTSC
instructions to figure out all the processor attributes and real-time
running speed. Here are some details:

http://www.sandpile.org/ia32/cpuid.htm
http://www.paradicesoftware.com/specs/cpuid/

As shown by all the fields buried in the CPUID specs, there is ample
room for tons of info, including strings such as "AuthenticAMD" and
"It's Hammer Time!", but AMD doesn't store the marked speed anywhere. I
thought for a moment that the stepping code would have this info, but
it's only obliquely related.

So, that's an education. It's also disappointing that the 3200+ choked
on the Auto settings in the BIOS, but I'll figure that out soon enough.



The AI Overclock was/is set to "Auto", like most everything else in the
BIOS... the default settings. Currently the FSB is at 166,
"effectively" 333, but I don't remember what it was running at with the
3200+.

Thanks...

This is a little old now, but I finally got the replacement cpu back
from newegg.com, and popped it in last night. To be sure all variables
were identical to the original trial, I set the cpu multiplier, dram
settings, etc, back to "Auto", except for forcing the memory bus to
"dual channel", as it was before. This was the only non-auto setting.
I even set the spread spectrum back to enabled, to eliminate all
differences.

CPU comes up... 2500+, 1.8 GHz, 166 MHz FSB, just as before. Bummer.

Set "Spread Spectrum" back to "disabled", as per kony. Reboot...

CPU comes up... 3200+, 2.2 GHz, 200 MHz FSB!

So, the very first suggestion by kony was the winner here, so it seems.
Just that one setting made it work as expected.

Just to be sure, I re-enable spread spectrum and give it another reboot.

CPU comes up... 3200+, 2.2 GHz, 200 MHz FSB.

Huh.

So there goes that theory. Maybe the original disabling of spread
spectrum was enough to kick the cpu in the butt and the bios/mobo/cpu
"remembered" that it could handle the higher speeds even after the
setting was re-enabled. But I'm just speculating. Clearly there is
some non-derministic voodoo going on, but mainly I'm just happy it's
working.

10 hours of memtest86 worked without issue. Sandra's arithmetic test
shows it working at about equivalent to 3100+... i.e. halfway between
3000+ and 3200+ on-file benchmarks. Kinda strange. Haven't run any
other benchmarks yet, but cpu-z shows the expected info, matching the
bios.


Lessons learned:

1) The internally reported cpu model "xx00+" is bogus -- it is
calculated from architecture parameters reported by CPUID and the speed
reported by RTDSC, and is therefore not reliable for determining the
tested-at and sold-at speed from the manufacturer. This is true for all
AMD cpu's (in the last decade or so, supporting CPUID and RTDSC) and may
be true of Intel as well, but I didn't check.

This one caught me by surprise because it's the opposite of what you
find on mid- and upper-end (i.e. not Intel-based) Sun/HP/IBM hardware,
where having certain knowledge of the cpu models is required for
licensing and inventory. I believe the encroachment of Intel/AMD
hardware into the server market is a good thing, but I have to wonder
how these issues are handled.

It appears that, on further investigation, that Athlon 64's and Opterons
support a "brand ID" that definitely reports the real model number,
unlike the Athlon XP or prior. I suspect AMD did this to fix the
server-market admin issues, and all future cpu's will behave the same
way. Can anyone with a 64 or Opteron confirm this?

2) ASUS motherboard default settings can't be trusted to run a cpu at
spec. Maybe this is common knowledge to overclockers, but it's the
first time I've seen it after building several systems from scratch.
I'm surprised at this as well, since ASUS can expect unnecessary support
calls (and therefore extra costs) from ordinary non-overclocking
customers. Then again, maybe my board is glitched.

3) Give the usenet boards a couple days before sending anything back to
newegg.


Thanks for all the help.

 

[chocolatemalt]

Then you're buying the wrong CPU.
Get a mobile Barton XP2500. Manually set the multiplier to
around 11X, the FSB to around 220 (if your memory will
tolerate it), and voltage roughly 1.7V... give or take 10%
on all variables.

You're probably right. Overclocking is a low priority for me, just
something to experiment with when I get the time, so I did no research
on it.

Ok, so long as you're happy with it. Frankly I'd consider
an Athlon 64 plus new motherboard, then selling the old
board or building a 2nd system.

Amazing how many things can be improved when you start throwing more and
more money at them. I'll probably build a new workstation in a year
or so... right now, too much of a splurge, and current system is fine.

Maybe it's flawless, or maybe it just does 166 ok and it's
coincidence that your present CPU is 166 FSB.

As my other post indicated, 200 FSB worked fine. I'd've been bummed for
sure if the A7V880 turned out to be a turkey, several months after I
bought it and never tested the full speed.

Quite a few have the setting, it's not just for overclocking
but anyone/everyone would be as well off disabling it,
except as I mentioned previously that a very rare
interference might occur, but intermittent interference is
actually greater with it enabled.


OEMs like "Enabled", they have to meet EMI emission
mandates.

Maybe this half-hearted support of this "feature" is why the ASUS bios
doesn't have any info on it whatsoever, not even a five-word blurb. In
any case, I've disabled it and will never touch it again, since it
seemed to be the culprit in the downclocking of my 3200+.

Not really a trick, just same method. This is normal,
expected. Having a motherboard default to too low a FSB is
an extremely common situation. In fact many many boards
default to the lowest FSB they support regardless of what
CPU is installed, and the user or technician setting up the
system must then set the FSB, IF the board doesn't do so
correctly, automatically.

As discussed in my other post, your info on this is correct. But I'm
still curious to see if the later Athlon 64 and Opteron cpu's no longer
use this method. With the "Brand ID" they ought to be able to show the
model regardless of running core speed or FSB.

Clearly this would be a good thing, in the sense that it might prevent
people like me from returning cpu's that appear to be futzed.

 

[kony]

You're probably right. Overclocking is a low priority for me, just
something to experiment with when I get the time, so I did no research
on it.

It's quite a performance increase and a great cost savings
too. Frankly I'd not spend the $$$ for an Athlon XP3200
today either way.

Amazing how many things can be improved when you start throwing more and
more money at them. I'll probably build a new workstation in a year
or so... right now, too much of a splurge, and current system is fine.

Well relatively speaking, the XP3200 is more expense with no
gain, so it stands to reason that if you want to spend more,
you ought to get more?

Maybe this half-hearted support of this "feature" is why the ASUS bios
doesn't have any info on it whatsoever, not even a five-word blurb. In
any case, I've disabled it and will never touch it again, since it
seemed to be the culprit in the downclocking of my 3200+.

They don't need to have any info on it. Generally if
someone doesn't know what a setting does already, they
shouldn't be told so they can change it, they should leave
it alone. There's where the problem comes in, whether the
motherboard manufacturer did the work of setting the bios
defaults correctly or not, and whether they really care
about optimal settings or just that it works "well enough".

As discussed in my other post, your info on this is correct. But I'm
still curious to see if the later Athlon 64 and Opteron cpu's no longer
use this method. With the "Brand ID" they ought to be able to show the
model regardless of running core speed or FSB.

There are several reasons it's good to have a system first
POST at the slowest speed. For one, the CPU generates less
heat and if the heatsink wasn't on good, the odds are
highest of a warning and shutdown in time to save the CPU.
A LOT of things on modern systems are tailored to help
people get them running when things aren't right.
Considering that someone who has never built a system before
has a good chance of getting one running, they're doing OK
by that mark considering how short product cycles are.

Clearly this would be a good thing, in the sense that it might prevent
people like me from returning cpu's that appear to be futzed.

Not to be condecending, but you should've checked the FSB
speed and if it wasn't what you *knew* it should be, you
should've changed it. This is fairly standard, common
knowledge for those who build PCs. Everyone has to start
somewhere though.

 

[chocolatemalt]

Clearly this would be a good thing, in the sense that it might prevent
people like me from returning cpu's that appear to be futzed.

Not to be condecending, ...[/QUOTE]

Didn't work. Real nice try though.

but you should've checked the FSB
speed and if it wasn't what you *knew* it should be, you
should've changed it. This is fairly standard, common
knowledge for those who build PCs. Everyone has to start
somewhere though.

I would certainly have done this, hacked around with timings, etc, had I
known that the "AMD Athlon XP (tm) 2500+" so boldly advertised by the
cpu in the BIOS and numerous utilities was in fact false.

I still wonder: Has AMD changed this behavior in the Athlon 64 and
Opteron? Or do they still report variable model numbers when speeds are
changed in the BIOS? I don't have either cpu at hand to test.

 

[kony]

I still wonder: Has AMD changed this behavior in the Athlon 64 and
Opteron? Or do they still report variable model numbers when speeds are
changed in the BIOS? I don't have either cpu at hand to test.

??

For the past dozen years motherboards have reported a
higher-speed CPU when the FSB was raised... to that similar
to another CPU that the bios recognizes. It's been that way
long before Athlon XP existed.