Upgrading an Octane 250MHz CPU to a 300MHz CPU

SGI hardware problems, solutions, tips, hacks, etc.
Forum rules
Any posts concerning pirated software or offering to buy/sell/trade commercial software are subject to removal.
User avatar
chicago-joe
Posts: 324
Joined: Fri May 09, 2003 9:01 am
Location: Chicago, IL
Contact:

Upgrading an Octane 250MHz CPU to a 300MHz CPU

Unread postby chicago-joe » Mon Nov 17, 2003 8:00 pm

Hi All,

I deciphered the resistor string to setup the CPU on the Octane CPU board. I upgraded my 250MHz 2M cache to a 300MHz 2M cache CPU board, it's a fairly simple upgrade once you know what to do. :wink:

If you try this be careful, the chips on this board are very static sensitive, I ruined a perfectly good spare 250MHz CPU board because I was not properly grounded and had a static discharge to the CPU board!

CPU board pictures:
top - http://users.rcn.com/joepage/250_top.jpg
bottom - http://users.rcn.com/joepage/250_bottom.jpg

Here we go:

Octane CPU modules (at least 195 - 400MHz) use a series of 32 sets of surface contacts to setup the CPU at boot time. Each set of contacts has a 4.65K resistor to either CPU board (V+) or (V-) to set a 1 or 0 for the boot-time mode data stream. These bits are read sequentially in groups of 8 by the IC chips U2-U5 (on the top of the CPU board) and then back to the CPU chip through a buffer. Bit numbers 9, 10, 11 and 12 set the SysClock (frequency of the CPU board crystal Y1) to Pclock (speed of the CPU core) multiplier, it is important to check the actual frequency of the crystal (printed on top) as all the CPU boards do NOT use a 100MHz crystal. The multiplier is read in the order of bit 9 to bit 12 where bit 9 is the most significant and bit 12 is the least significant bit of the series.

The table of binary string to clock mutiplier is:

0 0 0 0 - 0 - reserved
0 0 0 1 - 1 - reserved
0 0 1 0 - 2 - reserved
0 0 1 1 - 3 - x2
0 1 0 0 - 4 - x2.5
0 1 0 1 - 5 - x3
0 1 1 0 - 6 - x3.5
0 1 1 1 - 7 - x4

bit 9 is contact set marked - A - http://users.rcn.com/joepage/250_top_crop.jpg
bit 10 is contact set marked - B - http://users.rcn.com/joepage/250_bottom_crop.jpg
bit 11 is contact set marked - C
bit 12 is contact set marked - D

As you look at the pictures, when the resistor is across the "top" (of the picture) set of contacts the bit value is a 0 and across the "bottom" set of contacts is a 1. So in the case of this 250MHz board, the resistors are:
bit 9 - top - 0
bit 10 - bottom -1
bit 11 - top - 0
bit 12 - top -0
when read left to right the bin number is - 0100 - when compared to the chart above that is a multiplier of 2.5 times the crystal frequency (in this case) of 100MHz for a CPU frequency of 250MHz. This is true of at least the 195 -400 MHz boards (dual cpu boards are setup exactly the same); I have not seen a real (or even a picture of) a 175 MHz CPU board but I guess they are the same.

So get your 300MHz CPUchips, your grounding straps, your fine tipped soldering irons and your magnifying glasses and start modding! :)

Joe

User avatar
The Keeper
Posts: 1632
Joined: Mon Nov 17, 2003 9:23 pm
Location: Buffalo, NY
Contact:

Cool, thanks!

Unread postby The Keeper » Mon Nov 17, 2003 9:29 pm

Any idea if the fact that there's only 1MB of L2 on most of the sub 300MHz Octane (and O2 for that matter) CPU modules, might have any adverse affect on anything? Obviously, a modded 300MHz module with 1MB of L2 will be slightly lower performing in certain tasks than a true 300MHz module with 2MB, but I'm wondering if firmware/PROM/etc. is expecting to see 2MB?

Speaking of 175MHz Octane modules, here are a few pics of a dual 175MHz module I just took apart:

http://www.kalisiak.com/chris/octane_dual_175mhz_1.JPG
http://www.kalisiak.com/chris/octane_dual_175mhz_2.JPG
http://www.kalisiak.com/chris/octane_dual_175mhz_3.JPG

And by the way, how did you figure out which resistor pads were which bits? I see on your 250MHz R10k 2MB L2, the resistor pads R550, R552, R554, R583, R16, R13, (and two others obscured by "C") indicate the multiplier. Is there a pattern of some sort?

Thanks!
Chris

User avatar
The Keeper
Posts: 1632
Joined: Mon Nov 17, 2003 9:23 pm
Location: Buffalo, NY
Contact:

Another update on R10k Octane CPU modules...

Unread postby The Keeper » Tue Nov 18, 2003 9:01 pm

I just disassembled a 195MHz R10k Octane CPU, and noticed that the oscillator is 97.5MHz. The dual 175MHz module had an 87.5MHz oscillator, so things might be a little interesting.

Looks like the 195MHz module will turn out to be 292MHz (3x97.5), and the dual 175MHz will be 306MHz (3.5x87.5).

Any idea if there might be any "front-side bus" (for lack of a better description) issues by using an oscillator frequency that the R12k isn't expecting?

Chris

User avatar
chicago-joe
Posts: 324
Joined: Fri May 09, 2003 9:01 am
Location: Chicago, IL
Contact:

Unread postby chicago-joe » Wed Nov 19, 2003 12:33 pm

The Keeper Wrote
...but I'm wondering if firmware/PROM/etc. is expecting to see 2MB?

Irix reads the data from the CPU board for the information the system needs. I had the upgraded 300MHz 2M module overclocked to 350MHz for a couple of days and Irix reported the CPU as a 350MHz with 2M of cache. It doesn't seem to matter to the Octane or Irix what the CPU speed is. The overclocked chip would run for about 20 minutes then have cache faults and stop the system, I took the CPU board out immediately after the system stopped and the heatsink was to hot to touch. After the CPU cooled off, I put it back in the Octane and it would run for about 20 min and stop the system with the same cache faults.

FYI - Octane, 512M memory, 9G 10Krpm IBM HDD, IRIX 6.5.21f

MediaConvert
250MHz - 2M - 2.58min (stock CPU)
300MHz - 2M - 2.20min (upgraded w/ a 300MHz chip)
350MHz - 2M - 1.90min (overclocked 300MHz chip)
360MHz - 2M - 1.83min (stock CPU)

Any idea if there might be any "front-side bus" (for lack of a better description) issues by using an oscillator frequency that the R12k isn't expecting?

I called NEC and their thoughts (since they don't make these chips anymore) are that the R12K CPU chips would be fine up to the frequency rating plus 3% or 4%, above that, the RISC chips draw more current at a faster and faster rate (vs frequency) until they melt the fine wires between the ceramic package and the actual chip. The chips can run at any frequency less than their rating, so you should be OK up to about 310MHz. I would make sure the inside of the Octane is clean and dust free (of course you should be doing this anyway :wink: ) which will help keep good air flow.

Joe

SparcV
Posts: 51
Joined: Thu May 22, 2003 5:31 am
Location: Providence, RI
Contact:

Re: Another update on R10k Octane CPU modules...

Unread postby SparcV » Thu Nov 20, 2003 7:16 am

The Keeper wrote:I just disassembled a 195MHz R10k Octane CPU, and noticed that the oscillator is 97.5MHz. The dual 175MHz module had an 87.5MHz oscillator, so things might be a little interesting.

Looks like the 195MHz module will turn out to be 292MHz (3x97.5), and the dual 175MHz will be 306MHz (3.5x87.5).

Any idea if there might be any "front-side bus" (for lack of a better description) issues by using an oscillator frequency that the R12k isn't expecting?

Chris


Couldn't you replace the oscillator with a 100Mhz one to get the proper speed? Tho I think the extra effort might not be worth it.

User avatar
The Keeper
Posts: 1632
Joined: Mon Nov 17, 2003 9:23 pm
Location: Buffalo, NY
Contact:

Re: Another update on R10k Octane CPU modules...

Unread postby The Keeper » Thu Nov 20, 2003 10:24 am

SparcV wrote:
The Keeper wrote:I just disassembled a 195MHz R10k Octane CPU, and noticed that the oscillator is 97.5MHz. The dual 175MHz module had an 87.5MHz oscillator, so things might be a little interesting.

Looks like the 195MHz module will turn out to be 292MHz (3x97.5), and the dual 175MHz will be 306MHz (3.5x87.5).


Couldn't you replace the oscillator with a 100Mhz one to get the proper speed? Tho I think the extra effort might not be worth it.


That's a very good question. I pondered that as well, but I haven't really looked at the CPU module PCB to see if it's practical or not. I'll investigate it further this evening. Might have to pick up some oscillators from DigiKey...

Joe, do you have any thoughts on this?


Thanks,
Chris

User avatar
chicago-joe
Posts: 324
Joined: Fri May 09, 2003 9:01 am
Location: Chicago, IL
Contact:

Unread postby chicago-joe » Thu Nov 20, 2003 9:26 pm

The Keeper Wrote
Might have to pick up some oscillators from DigiKey...

I don't know if these are actual crystals (as part of an oscillator circuit) or self contained "oscillators in a can", I'm not at home until this weekend and can't tell without looking more closely. Two things to consider either way:
1) If you change the SysClk frequency (100, 97.5, 87.5MHz etc.) the cache memory is also clocked from the CPU chip (based on the SysClk) so you will have to calculate a new multiplier for that and change more resistors, you could wind up with a slower cache speed and lost performance. I think Intel had this problem with the various core/cache speeds of some of the Pentium CPU chips.
2) I looked at the picture of the Octane CPU board and the crystal/oscillator is soldered through the board, these are a multilayer board (6 or 7) and you would have to be very careful not to short or damage the layers when you unsoldered the can, a mistake here could ruin the board
I would agree with SparkV, I don't think the risk (for the very minor performance gain) would be worth the effort or cost, but that's what modding is all about. :wink:

Joe

User avatar
The Keeper
Posts: 1632
Joined: Mon Nov 17, 2003 9:23 pm
Location: Buffalo, NY
Contact:

Ok, results of my experimentation...

Unread postby The Keeper » Fri Nov 21, 2003 10:53 pm

I've spent about 6 hours on this so far, and here's a brief summary of what I have done. Feedback is appreciated.

1) Verified five D30710RS-300 CPU's using an 030-1355 as a test platform (native 300MHz R12k).
2) Swapped resistors on 195MHz R10k to upgrade it to 300MHz (3x97.5MHz).
3) Dropped in 300MHz IC, dropped onto 030-0887 mobo, plugged in, powered on, frontplane LED's indicate bad CPU.
4) Dropped two 300MHz IC's into dual 175MHz R10k module without swapping resistors, just to see what happens. Same bad CPU.
5) Dropped both upgraded modules into another 030-0887 mobo. Same bad CPU.
6) Dropped both into 030-1467 mobo, just for grins. Same bad CPU.
7) Pulled 300MHz IC from 195MHz, dropped it back into 030-1355, diagnostics pass.
8 ) For grins, I dropped one of the 175MHz IC's (actually 180MHz, like in the Origins) into the upgraded 195MHz module. Same bad CPU. Not sure what's up with that.

So, the problem tracks the 195MHz and 175MHz modules. Dropping 300MHz IC's into otherwise unmodified dual 175MHz module should have resulted in two 175MHz R12k's, but it didn't make it past POST.

Preliminary conclusion is that the 1MB L2 is causing problems. No combination of 300MHz IC and 1xxMHz module tested this evening seems to work. Note that all confirmed reports of successful upgrades in this forum are using 2MB L2 modules as a basis for upgrade.

Looking around the room for things I haven't tried yet, I pull out one last mobo with a 225MHz R10k module, which has 1MB L2. Pop the cover (as a point of reference, it has a 90MHz oscillator, for a 2.5 multiplier), drop a 300MHz IC into it, drop it onto an 030-0887... And it POSTS and passes IDE...

So...

Apparently there is something specific to the 1xxMHz R10k PCB's that don't allow it to successfully POST with a 300MHz IC installed in it and/or clocked at 300MHz. No idea what the problem is.


Comments? Questions? Anything?


(Heh, $50 to the first person that figures out how to make these 300MHz IC's work in a 1xxMHz module, specifically a dual 175MHz... I'd really love to have a working dual 300MHz Octane module, even if there's only 1MB of L2 per CPU...)


Chris


Oh yeah, forgot to mention, I also tried dropping in a 300MHz IC into a 150MHz O2 module, and the same thing happened as the Octanes -- the O2 didn't complete POST.

User avatar
chicago-joe
Posts: 324
Joined: Fri May 09, 2003 9:01 am
Location: Chicago, IL
Contact:

Unread postby chicago-joe » Sun Nov 23, 2003 12:09 pm

Keeper,

I'm not sure what's going on here, I don't have a 175 or 195 cpu module but I have modded a 225-1M, 250-1M and 2M and a dual 250-1M to 300MHz without problems. I have (quickly) looked back over the spec sheets and I didn't see anything obvious, I will look over the sheets again in more detail this week when I have more time.
Two questions: 1.) what is the part number on the Celestica CPU power supply, these are the small copper boards(s) that screw onto the ends of the finned heatsink assy. that bolts over and under the CPU board? The p/n on all mine is: 9981049 94WT (this number is a white sticker on the outside of the large inductor on the left-hand side of the Celestica board) If these are different between the 175/195 series and 225/250/300 series of CPU boards, that may be the problem. 2.) If you put the 175 CPU back into the original board and put the CPU board back into the Octane, does it get all the way to boot-up?

Joe

SparcV
Posts: 51
Joined: Thu May 22, 2003 5:31 am
Location: Providence, RI
Contact:

Unread postby SparcV » Sun Nov 23, 2003 3:51 pm

chicago-joe wrote:
Two questions: 1.) what is the part number on the Celestica CPU power supply, these are the small copper boards(s) that screw onto the ends of the finned heatsink assy. that bolts over and under the CPU board? The p/n on all mine is: 9981049 94WT (this number is a white sticker on the outside of the large inductor on the left-hand side of the Celestica board) If these are different between the 175/195 series and 225/250/300 series of CPU boards, that may be the problem.


Well, the power supply on my single 195 module is the same: 9981049 94WT

My dual 195 power boards dont have any stickers on them and they look different.

Could the dual 1xxMHz boards be wired for a different voltage?

Here's a picture of both power boards:

http://www.vhdesigns.com/195powersupply.jpg

User avatar
The Keeper
Posts: 1632
Joined: Mon Nov 17, 2003 9:23 pm
Location: Buffalo, NY
Contact:

Unread postby The Keeper » Sun Nov 23, 2003 5:48 pm

chicago-joe wrote:I'm not sure what's going on here, I don't have a 175 or 195 cpu module but I have modded a 225-1M, 250-1M and 2M and a dual 250-1M to 300MHz without problems.


Yes, the 225MHz module that I have here is able to accept a 300MHz IC as well. In fact, the hinv reports a 225MHz R12000 with 1MB of L2.

chicago-joe wrote:I have (quickly) looked back over the spec sheets and I didn't see anything obvious, I will look over the sheets again in more detail this week when I have more time.


Sounds good.

One thing I noticed, by the way, is that the 1xxMHz CPU modules have cache IC's on both sides of the PCB, totalling the 1MB. The 225MHz CPU module only has cache IC's on the top side, totalling 1MB, so the cache density of the 225MHz module is higher. Same holds true for the 300MHz R12k module, with IC's on both sides giving you 2MB of L2.

I could see how that's related, but I'm not exactly sure why.

The next test will be to drop the 225MHz R10k IC into a 1xxMHz CPU module, and see if the same thing happens. If the 225+1xxMHz combo has the same "bad CPU" problem, then that might mean that later R10k CPU's and the R12k CPU's have a different, and incompatible, cache subsystem.

chicago-joe wrote:Two questions: 1.) what is the part number on the Celestica CPU power supply, these are the small copper boards(s) that screw onto the ends of the finned heatsink assy. that bolts over and under the CPU board?
The p/n on all mine is: 9981049 94WT (this number is a white sticker on the outside of the large inductor on the left-hand side of the Celestica board)


I see the sticker you're referring to on the 225, but the 195 doesn't have that same kind of sticker. The inductor has a part number, but I'm sure it's the part number of the inductor itself. The VRM (I'm assuming that's what this board is, anyway) is different between the 225 and the 195, but I can't really tell what's different. They're soldered through-hole boards, so it's not going to be practical to swap them...

http://www.kalisiak.com/chris/PDR_0001.JPG

chicago-joe wrote:If these are different between the 175/195 series and 225/250/300 series of CPU boards, that may be the problem. 2.) If you put the 175 CPU back into the original board and put the CPU board back into the Octane, does it get all the way to boot-up?


Good question. I'll find out shortly. I need to re-verify a few of the CPU's in the original configurations to make sure they still work.



Thanks for your help!

Chris

User avatar
The Keeper
Posts: 1632
Joined: Mon Nov 17, 2003 9:23 pm
Location: Buffalo, NY
Contact:

Ok, spent a few more hours...

Unread postby The Keeper » Mon Nov 24, 2003 11:47 pm

I spent a few more hours, and appear to have ended up trashing a few 1xxMHz modules. I don't know why, but it appears that two single 195MHz modules and one dual 175MHz module don't even work anymore, even if I put them back together with the original CPU's.

On the one 195MHz and the dual 175MHz, I didn't even move the resistors. I just dropped in a 225MHz R10k IC and a pair of 300MHz R12k IC's, respectively, into the modules, and now no matter what I try, I can't get the Octane to make it past the "bad CPU" stage.

(As a point of reference, I tried disassembling the virgin 195MHz module, dropping the 225MHz IC into it, and powering it up. This was to prove out the theory that the 2xxMHz version of the chip introduced some dramatic change. The red LED blinked at me, like there was a memory problem, but not the typical "bad CPU" indicator. No idea what to make of that.)

There is a difference in part numbers between the 1xxMHz and 2xxMHz IC's -- D30700RS-200 and D30700LRS-225. Don't know what the "L" means in the 225MHz IC, could mean anything, but my guess is there's some relationship to the problems I'm having.

I could buy either the explanation that the faster CPU's took out something on the CPU module, like the cache chips, or the explanation that I physically happened to damage all three of the 1xx modules, but not the 2xx module.

As a corroboration to the damage theory, I've pretty much trashed the heat sinks of the 195's -- the heat sink assemblies are likely not designed to be disassembled and reassembled repeatedly. The screws just gave out and snapped. Nothing happened to the CPU PCB, though. I just ended up needing an "L" allen wrench in the "high-torque" position to torque the heads off the screws. No idea why, but four of the eight screws decided they didn't want to play nice.



Summary seems to be:
1) Only 2xxMHz modules can be upgraded.
2) Any attempts to upgrade a 1xxMHz module will kill the module, but not the 300MHz IC. I know it doesn't make sense, but I have three dead 1xxMHz modules to show for my troubles.
3) Be judicious with the disassembly/reassembly of CPU modules. You appear to only have between 6 and 10 tries to get it right before the screws start siezing and snapping.
4) When reassembling, hold the CPU module in one hand while starting the screws in the other hand. If you set the module on the work surface with no support under the bottom part of the heat sink, the CPU module can come slightly apart, and the CPU IC can become misaligned. Once two diagonal screws have started to apply pressure, then it appears to be safe to set down the module to finish driving the screws.


Ok, enough rambling for the night...


I hope to try a dual 2xxMHz Octane module at some point... I'll post once I have acquired a dual 2xxMHz module.

Chris

acronym
Posts: 178
Joined: Sat Sep 27, 2003 8:11 pm

Unread postby acronym » Sat Dec 06, 2003 11:10 am

Hi Joe-
where are you getting your 300 chips?
I've only found 2 places online, one on Yahoo shopping is charging $17.
http://shop.store.yahoo.com/americanmic ... s-300.html

I'm not sure I want to give them my card number, tho'

bigendian
Posts: 35
Joined: Wed Jul 30, 2003 10:06 pm
Location: Winter Park, FL
Contact:

Unread postby bigendian » Sun Dec 07, 2003 6:13 pm

Anybody have any hope of this same trick working with IP25s?

daniel

User avatar
chicago-joe
Posts: 324
Joined: Fri May 09, 2003 9:01 am
Location: Chicago, IL
Contact:

Unread postby chicago-joe » Sun Dec 07, 2003 8:19 pm

acronym wrote
where are you getting your 300 chips?
I've only found 2 places online, one on Yahoo shopping is charging $17.


R12K-300MHz cpu chips are US$133 + S&H per chip, this price is for 5 chips or less, but the chips can be bought 1 at a time. This is the contact person and information:

Circuit Solutions (631)563-1100 http://www.cirsolutions.com
Contact person is: Jon
email is: mailto:jd@cirsolutions.com
the proper P/N is: UPD30710RS-300 Rev 2.4 or Rev 2.3 (either works fine)

If you have found these for $17 each, you should buy all you can for that price and resell them. :)

Joe


Return to “SGI: Hardware”

Who is online

Users browsing this forum: Google [Bot] and 2 guests