PPro on a BX?

[Rob Stow]

Not sure if the L2 is on-die, but it definitely runs at full bus speed:

http://processorfinder.intel.com/scripts/details.asp?sSpec=SL259

That link gives the following:
CPU speed = 200 MHz
Bus speed = 66 MHz
L2 speed = 66 MHz

Not something that supports your contention - but
it doesn't support mine either.

 

[David Maynard]

That link gives the following:
CPU speed = 200 MHz
Bus speed = 66 MHz
L2 speed = 66 MHz

Not something that supports your contention - but
it doesn't support mine either.

Yeah, that doesn't make it clear but the intro page for the 200 Mhz, 1 meg
L2 cache data sheet download does.

http://support.intel.com/design/archives/processors/pro/docs/243570.htm

The Pentium® Pro processor with 1 MB L2 cache is a multichip module
targeted for use in high-end 4-way multiprocessor capable server systems.
The component package contains an Intel Pentium Pro processor core, and 1
MB of L2 cache. The 1 MB cache is built using two of the 512 KB SRAM die
found in the 512 KB version of the Pentium Pro processor. While the 512 K
version uses a conventional ceramic package, the Pentium Pro processor with
1 MB L2 cache integrates the three die in a plastic package with an
aluminum heat spreader. This 387-pin package is compatible with the current
Pentium Pro processor footprint. The Pentium Pro processor with 1 MB L2
cache routes all of the processor's high-speed cache interface bus through
balanced nets on a thin film interconnect substrate to the two L2 SRAMs.
This allows for internal component operation speeds of 200 MHz between the
Pentium Pro processor and the L2 cache die.

 

[Rob Stow]

Yeah, that doesn't make it clear but the intro page for the 200 Mhz, 1
meg L2 cache data sheet download does.

http://support.intel.com/design/archives/processors/pro/docs/243570.htm

The Pentium® Pro processor with 1 MB L2 cache is a multichip module
targeted for use in high-end 4-way multiprocessor capable server
systems. The component package contains an Intel Pentium Pro processor
core, and 1 MB of L2 cache. The 1 MB cache is built using two of the 512
KB SRAM die found in the 512 KB version of the Pentium Pro processor.
While the 512 K version uses a conventional ceramic package, the Pentium
Pro processor with 1 MB L2 cache integrates the three die in a plastic
package with an aluminum heat spreader. This 387-pin package is
compatible with the current Pentium Pro processor footprint. The Pentium
Pro processor with 1 MB L2 cache routes all of the processor's
high-speed cache interface bus through balanced nets on a thin film
interconnect substrate to the two L2 SRAMs. This allows for internal
component operation speeds of 200 MHz between the Pentium Pro processor
and the L2 cache die.

Ta-dah! And we have a winner folks.

 

[David Maynard]

Ta-dah! And we have a winner folks.

Btw, from that you can see that a P-II is essentially a repackaged, for
lower cost, Pentium Pro. (The P-II data sheets even give a 'side by side'
package comparison.)

'Typical' ICs on a 'conventional' PCB, the slot cartridge, rather than a
ceramic (or plastic) chip carrier and the cache lowered to half speed,
although you're back to 200 MHz cache speed with the P-II 400 (and back to
the Pentium Pro 150's cache speed as early as a P-II 300).

The P-II also dropped some of the 'mission critical' features of the
Pentium Pro but added MMX.

'P-III' added SSE, but stayed with the same slot configuration.

'Coppermine' was simply the next logical progression, moving the cache onto
the die rather than separate components; achieving what the Pentium Pro's
chip carrier package had been 'simulating'.


Actually, the P-II mobile went to on-die cache before the P-III did,
because the space constraints are obviously tighter and mobiles can sustain
a higher cost premium.

 

[~misfit~]

David Maynard wrote:

Hi David,

Btw, from that you can see that a P-II is essentially a repackaged,
for lower cost, Pentium Pro. (The P-II data sheets even give a 'side
by side' package comparison.)

'Typical' ICs on a 'conventional' PCB, the slot cartridge, rather
than a ceramic (or plastic) chip carrier and the cache lowered to
half speed, although you're back to 200 MHz cache speed with the P-II
400 (and back to the Pentium Pro 150's cache speed as early as a P-II
300).

The P-II also dropped some of the 'mission critical' features of the
Pentium Pro but added MMX.

'P-III' added SSE, but stayed with the same slot configuration.

'Coppermine' was simply the next logical progression, moving the
cache onto the die rather than separate components; achieving what
the Pentium Pro's chip carrier package had been 'simulating'.

Thanks for that. Maybe I should have said 'on-CPU' rather than on-die. I
don't know if there's actually much of a difference in performance. I knew
that the L2 cache was definately full-speed. And also had ECC.(?)

Actually, the P-II mobile went to on-die cache before the P-III did,

What about the Mendicino Celerons?

 

[~misfit~]

It's a second chip on the same chip carrier; not on CPU die. It does
run full speed.

Thanks. I thought about it after I posted and realised I should have said
that. Too casual, where I meant 'on CPU (or "chip-carrier")' I said 'on
die'.

Was the L2 ECC?

 

[David Maynard]

David Maynard wrote:

Hi David,




Thanks for that. Maybe I should have said 'on-CPU' rather than on-die. I
don't know if there's actually much of a difference in performance. I knew
that the L2 cache was definately full-speed. And also had ECC.(?)

Depends on how you define 'on-cpu'. The P-II has it on the cart but it's
half speed.

What about the Mendicino Celerons?

True.

It's obvious what the idea behind the original celerons was: P-II with the
cache chips dropped off. People complained there was no L2 so, one theory
goes, the 128K cache celerons were originally half cache P-II mobiles
(MMC-2 mobiles were first 512k cache chips like the desktop P-II and then
256k on-die like the coppermines. PGA mobiles were 512k on-die cache).

Or maybe the celeron was first and they increased it for the mobiles.

 

[~misfit~]

Depends on how you define 'on-cpu'. The P-II has it on the cart but
it's half speed.

It does indeed. I don't think of the slot CPU's as CPUs really, more of a
daughter-board. Like a slocket with an embedded CPU (and half-speed L2
sometimes). Real CPUs have pins.

True.

It's obvious what the idea behind the original celerons was: P-II
with the cache chips dropped off. People complained there was no L2

More like people complained because they were dog-slow. I once ran a P-1 200
with no L2 cache and it was slower (subjectively) than a P-120 with 256KB
on-board. At the time I first used it I didn't realise it had no L2 and
couldn't work out why it was so slow.

so, one theory goes, the 128K cache celerons were originally half
cache P-II mobiles (MMC-2 mobiles were first 512k cache chips like
the desktop P-II and then 256k on-die like the coppermines.

Were they the same physical pagkaging as the Mendicino? That would shed some
light on that theory.

PGA
mobiles were 512k on-die cache).

Or maybe the celeron was first and they increased it for the mobiles.

I don't know how the time-frame went. All I know is I really liked the
Mendicino. Heaps better performance than the desktop PIIs (in a slocket for
the PGA ones). Good to OC too, (up to about 565Mhz stable for me) and
out-performed the PIIs easilly. That one was significantly better (on
benchmarks) than a P-III Katmai 500Mhz too.

I've collected a bunch of P-Pros over the last few years but they're all
256KB L2 cache ones. Believe it or not the 512/1024KB L2 versions still
fetch good money. At least in this neck of the woods. Also, there were a lot
less of them made. I only ever had one P Pro system working, and then only
for a day or two. It was a dual board and I had a matched pair of 200Mhz
CPUs in it and it worked perfectly. I had an OS installed, XP Pro (don't
laugh, it was slow but it was the only multi-CPU capable OS I had access to
and was familiar with). Then I noticed on the mobo that I could OC it to
233Mhz. How could I resist? I moved the jumpers and it booted fine. Once.
Ran for a few hours, seemed nice and cool. Next time I went to boot it it
was as dead as a parrot in a Monty Python sketch. I never got that board to
work again. Tried clocking it back, underclocking it, just one, then the
other CPU, nothing. It was an ASUS board (Just noticed the crosspost).

I'd still like to have one running and am always on the look-out for a
Socket 8 board. Make a good firewall with linux coyote or something similar
methinks. Gotta build me a firewall box now I have ADSL on and a fixed
IP/speed upgrade soon. I'm not confident just with software firewalls. A
mate's talked me into Coyote. A P1, two PCI 10/base-T NICs, a graphics card
(to set it up) minimum 32MB RAM and a floppy. No HDD needed. He's set up a
couple for friends. Use 'em in a mini-tower and he reckons you can use
passive cooling on the CPU and remove the fan from the PSU and cut out the
grilles a bit.. Heat rising takes care of the minimal heat-transfer needed.
he set up a box like this over two years ago and it hasn't been turned off
in all that time. Costs next to nothing to run, no heavy processing and all
done in RAM, just uses the floppy to boot from and to store settings if he
tel-nets (?) into it to update it/set permissions.

I'm in the process of finding a nice case now, sorting through my back room.
I want a case that will have a good air-path for a thermal gradient
flow-through. I have some large passive HS's for socket 7. I suggested to
him that I might use a PSU fan on 5 or 7 volts and he said totally not
needed. It works fine without and with no moving parts (or fans to suck dust
in) they run for years without being touched. He said the usual place to put
them is in a wardrobe, (big) cupboard or even in the crawl-space. He tried
with desktop cases but the thermal gradient wasn't there and they failed
often. He hasn't had a single one fail in a good tower.

Oopps, got a bit off track. You get that at 3:15am.

 

[David Maynard]

It does indeed. I don't think of the slot CPU's as CPUs really, more of a
daughter-board. Like a slocket with an embedded CPU (and half-speed L2
sometimes). Real CPUs have pins.

hehe. It's 'packaging', when you think about it. Even 'on-die' is a
packaging choice.

More like people complained because they were dog-slow. I once ran a P-1 200
with no L2 cache and it was slower (subjectively) than a P-120 with 256KB
on-board. At the time I first used it I didn't realise it had no L2 and
couldn't work out why it was so slow.

Running no L2 wasn't really all that uncommon back then.

Were they the same physical pagkaging as the Mendicino? That would shed some
light on that theory.

I never paid that much attention to it but I don't see it would reveal
much. It's the die that's 'the thing'.

I don't know how the time-frame went. All I know is I really liked the
Mendicino. Heaps better performance than the desktop PIIs (in a slocket for
the PGA ones). Good to OC too, (up to about 565Mhz stable for me) and
out-performed the PIIs easilly. That one was significantly better (on
benchmarks) than a P-III Katmai 500Mhz too.

Well, they only matched, or exceeded, the P-II when overlcocking. When used
'as specified' the P-II was always faster.

I've collected a bunch of P-Pros over the last few years but they're all
256KB L2 cache ones. Believe it or not the 512/1024KB L2 versions still
fetch good money. At least in this neck of the woods. Also, there were a lot
less of them made. I only ever had one P Pro system working, and then only
for a day or two. It was a dual board and I had a matched pair of 200Mhz
CPUs in it and it worked perfectly. I had an OS installed, XP Pro (don't
laugh, it was slow but it was the only multi-CPU capable OS I had access to
and was familiar with). Then I noticed on the mobo that I could OC it to
233Mhz. How could I resist? I moved the jumpers and it booted fine. Once.
Ran for a few hours, seemed nice and cool. Next time I went to boot it it
was as dead as a parrot in a Monty Python sketch. I never got that board to
work again. Tried clocking it back, underclocking it, just one, then the
other CPU, nothing. It was an ASUS board (Just noticed the crosspost).

I'd still like to have one running and am always on the look-out for a
Socket 8 board. Make a good firewall with linux coyote or something similar
methinks. Gotta build me a firewall box now I have ADSL on and a fixed
IP/speed upgrade soon. I'm not confident just with software firewalls. A
mate's talked me into Coyote. A P1, two PCI 10/base-T NICs, a graphics card
(to set it up) minimum 32MB RAM and a floppy. No HDD needed. He's set up a
couple for friends. Use 'em in a mini-tower and he reckons you can use
passive cooling on the CPU and remove the fan from the PSU and cut out the
grilles a bit.. Heat rising takes care of the minimal heat-transfer needed.
he set up a box like this over two years ago and it hasn't been turned off
in all that time. Costs next to nothing to run, no heavy processing and all
done in RAM, just uses the floppy to boot from and to store settings if he
tel-nets (?) into it to update it/set permissions.

Right. telnet. command line kind of interface.

 

[P2B]

Interesting, because according to the P-II data sheets, while codes up
to 3.5 are defined, only up to 2.8 is 'required' for P-II motherboards.

MIght be worth breaking out the DVM just to make sure it's really doing it.

The VRM datasheet says 1.3 -> 3.5 - and the board wouldn't POST if the
CPU wasn't getting Vcore.

 

[David Maynard]

The VRM datasheet says 1.3 -> 3.5 - and the board wouldn't POST if the
CPU wasn't getting Vcore.

It might post, and then lock up, if it was getting too low a Vcore.

Just thought it might be worth verifying.

 

[P2B]

It might post, and then lock up, if it was getting too low a Vcore.

Just thought it might be worth verifying.

Never let it be said I left a stone unturned

Vcore = 3.31v according to my trusty DMM.

 

[David Maynard]

Never let it be said I left a stone unturned
Hehe


Vcore = 3.31v according to my trusty DMM.

Doesn't look like that's 'the problem'

 

[P2B]

Doesn't look like that's 'the problem'

No.

Disabling the CPU caches in BIOS then installing the PPro didn't make
any difference either.

I'm about out of ideas - I suspect getting this to work would require
disassembling a working LX or FX BIOS and porting the CPU initialisation
code into a BX BIOS, just not sure I'm sufficiently motivated to learn
the requisite skills. Where's apple_rom when you need him?

 

[P2B]

No.

Disabling the CPU caches in BIOS then installing the PPro didn't make
any difference either.

... to booting the PPro - but it had a somewhat surprising side-effect:

I forgot to re-enable the CPU caches when I re-installed the Tualatin-S
today to test some new software. XP Pro wouldn't even boot with the
caches off - it sat at the splash screen for 3-4 minutes, then BSOD. I
assumed the hardware had been disturbed (it's a board-on-the-bench
system) and re-plugged it all, same thing. Then I remembered the caches...

Interesting - NT and W2K systems are dead slow but otherwise unaffected
by lack of CPU cache, however it stops XP Pro cold.

P2B

 

[BananaOfTheNight]

Here's a random idea for you: get a P2 overdrive chip like this one
http://www.valuelist.co.uk/ProductDetailPage.asp?guid=&ticketid=0&productID=2735
and put it on one of those slotkets just for fun...

 

[David Maynard]

.. to booting the PPro - but it had a somewhat surprising side-effect:

I forgot to re-enable the CPU caches when I re-installed the Tualatin-S
today to test some new software. XP Pro wouldn't even boot with the
caches off - it sat at the splash screen for 3-4 minutes, then BSOD. I
assumed the hardware had been disturbed (it's a board-on-the-bench
system) and re-plugged it all, same thing. Then I remembered the caches...

Interesting - NT and W2K systems are dead slow but otherwise unaffected
by lack of CPU cache, however it stops XP Pro cold.

Well, that is strange. Why would XP give a hoot whether cache is enabled or
not?

Hmm. Maybe it's trying to time something and being super slow screws up a
counter somewhere, or tells it crucial things are missing/defective (from
time-out).

 

[P2B]

Here's a random idea for you: get a P2 overdrive chip like this one
http://www.valuelist.co.uk/ProductDetailPage.asp?guid=&ticketid=0&productID=2735
and put it on one of those slotkets just for fun...

If/when one turns up in the dumpster, I'll do just that. Can't see
*buying* one, though, even at that price

 

[P2B]

Well, that is strange. Why would XP give a hoot whether cache is enabled
or not?

Dunno, but it's also the only OS I've used that objects to you removing
one CPU in a dual setup - similar issues result.

Hmm. Maybe it's trying to time something and being super slow screws up
a counter somewhere, or tells it crucial things are missing/defective
(from time-out).

It certainly was sloooow - I didn't know XP displayed the black & white
progress bar (like W2K) before the splash, I guess it's normally not
there long enough to see - but surely if it's timing something it would
use system clock timer-ticks, which are presumably unaffected.

 

[David Maynard]

Dunno, but it's also the only OS I've used that objects to you removing
one CPU in a dual setup - similar issues result.



It certainly was sloooow - I didn't know XP displayed the black & white
progress bar (like W2K) before the splash, I guess it's normally not
there long enough to see

Hehe. Yeah. I caught it going by once too. Don't remember why. Maybe I
blinked just right

- but surely if it's timing something it would
use system clock timer-ticks, which are presumably unaffected.

One would think so but MS has been known to use timing loops. At any rate,
the second half of my 'hmm' doesn't depend on a counter overflow for the
'problem'. It might simply think something critical is not working, hence
'missing', because it failed to respond in a timely manner.