SGI: Hardware

It seems that LACE Technology will no longer do the soldering job required for the O2 300 MHz RM5200 -> 600 MHz RM7000C upgrade modification. Has anyone located another rework shop willing to do this work for a reasonable price?

Also, has anyone tried R12K/400 -> 450 MHz overclocking and/or L2 cache upgrade as ChicagoJoe suggested sometime back?

I will have to return my loaner Octane back to its original owner in a few weeks and am looking for a replacement machine. Despite it being slower than Octane2, I am leaning towards a faster CPU for my O2. I had originally hoped to wait for a 700 MHz RM7000C, but it seems as though such a CPU may never be built.

If we could get some inside help from someone at SGI then we could probably make the required PROM modifications for the RM7900 at 835 and 900 MHz.

Anyone have any contacts that might be able to make this a possibility.

I'm still holding off in the vian hope that this may eventuate.

Glenn

ruckusman wrote: If we could get some inside help from someone at SGI then we could probably make the required PROM modifications for the RM7900 at 835 and 900 MHz.

I don't want to offend you but do you believe in Santa Claus too?


Jason

Jason wrote:

ruckusman wrote: If we could get some inside help from someone at SGI then we could probably make the required PROM modifications for the RM7900 at 835 and 900 MHz.

I don't want to offend you but do you believe in Santa Claus too?


Jason

...hehehehe!! I could be a lot happy if simply one company could replace to LACE, taking on a common base the 600MHZ upgrade jobs

Well; I've asked a week ago to somebody in SGI about the PROM modification help. No reply...

Jason wrote:

ruckusman wrote: If we could get some inside help from someone at SGI then we could probably make the required PROM modifications for the RM7900 at 835 and 900 MHz.

I don't want to offend you but do you believe in Santa Claus too?

A few years ago, maybe ... sometimes engineers get attached to their creations. What engineer would want to see his O2 crippled to 200mhz when he knows it can run at 900 ?

Same thing happened at IBM ... the source for OS/2 is now out there, somewhere ....

but is there anyone left at sgi with any kind of attachment to the O2 ? Do they have any Irix engineers at all nowadays ? Curious .....

It was more a case of not wanted to accidentally re-flash the PROM with something which wouldn't boot than not knowing what to change, wasn't it? Did anyone ever find the relevant list of CPU IDs? And would the kernel have to be altered too? Mayhap a huge pile of backup motherboards might be more helpful than inside information :)

lewis wrote: It was more a case of not wanted to accidentally re-flash the PROM with something which wouldn't boot than not knowing what to change, wasn't it?

Yes, probably the case. But the change stills beeing possibly.

lewis wrote: Did anyone ever find the relevant list of CPU IDs? And would the kernel have to be altered too? Mayhap a huge pile of backup motherboards might be more helpful than inside information

I think even that some of ours Neko-Masters are ready as to put their hands on the PROM assembler code. Just matter of time and interest to do the job. Probably these guy that was on the tunning of the Octane Linux kernel (see the thread to see who is the guy!), could be ready... he seems the perfect partner for our Chicago-Joe CPU-Patron-Saint!

Don't tell me Santa Claus doesn't exist, OH MAN, I'm just getting over the loss of the easter bunny.

I'm really bummed now

ruckusman wrote: Don't tell me Santa Claus doesn't exist, OH MAN, I'm just getting over the loss of the easter bunny.

I'm really bummed now

hehehe!!!

Get the 600MHZ Glenn!; There'll be enoughly faster for the major part of the works, and I think can't be so difficult get a lab ready to do the job.

Chicago-Joe says in his 600MHZ thread that he is finding for a new one.

Cheers!

I got a 600Mhz mod from chicagojoe a little while ago, everything works fine and I'm really happy about it, and Joe sure did a great job (btw: thanks Joe!)

If your application needs an O2 (and for instance its cheap video in/out capabilities, or 2D capabilities, or picture editing capabilities -roam or whatever-), or you already have the licenses (Jaleo, whatever) and everything, do it. If you're not sure of short in money, go with an octane, am octane v6+r12k will be cheaper than an O2 mod upgrade. Keep that in mind when you do the check.

However, a 600Mhz O2 is a very nice thing to own, and it surely makes ppl jealous.


Jason

Hi All,

I have found a new shop here in Chicago to do the 600MHz CPU chip rework, I have sent them one set of 4 boards and have been happy with their service. I am going to send them another set of boards (including one I am going to do some cache mods to) this Tuesday, if they do a good job on this set as well, I will post their contact information. I don't want to be hard to get along with by not posting the shop's information, but I want to be sure I'm giving people information for a reliable rework shop.

For the O2, when you start the computer the CPU board reads the data from the boot-mode PROM chip (the Xilinx chip) into specific registers on the CPU chip and does a Power On Self Test (POST). If the POST passes, the ip32PROM then reads that data from the specific CPU chip registers and checks that the CPU module is returning data values the ip32PROM recognizes and passes those data values on to IRIX for the system. The problem is that the data is in different registers and has different values for the RM7900 series of chips than the RM7000 series of chips. The RM7000 chips and the RM5270 chips use the same data registers so the ip32PROM sees the RM7000 chips just fine, not so with the RM7900 chips. When ip32PROM does not see a data value it recognizes (for whatever reason), it halts the system at that point.
So at least three things need to be done to get this project back on track:
1.) a way needs to be found to tell the ip32PROM the correct registers to look in for the data on the RM7900 CPU chip
2.) new values need to be added to the ip32PROM's list of possibe data values for a couple of items
3.) the setup for the L2 cache in the ip32PROM needs to be changed as the RM7900 chips start the cache memory at a different memory location than the RM7000 chips
If there is someone out there who understands the ip32PROM and how to modify the bin file, I will help them any way I can but the software part of this is beyond my understanding.

The situation with the cache upgrade is this: a rework shop is going to charge about $30 per cache chip (9 or 5 chips) to remove the old chips and another $30 per chip (5 chips) to put the new ones on. The new cache chips are about $25 each (5 chips) so the total to mod one 250MHz 1M cache Octane CPU module to 4M cache will be about $545 plus S&H, double that for a dual processor CPU module. The cost for a 360MHz Octane or 400MHz O2 module modded to 4M cache will be about $425 plus S&H. It goes without saying that this can get to be an expensive hobby very fast. I have been working on a way to do this cache rework in a more reasonable money way, at this point I just can't justify the "out of pocket" expense of several hundred dollars for something I don't know will work. I'm working on a couple ideas on this though, as I would very much like to see the performance difference of a 1 or 2M cache module vs a 4M cache module.

Anyone who has ideas or would like to help, please jump in, there is plenty of room in the O2 pool.

Joe

chicago-joe wrote: Anyone who has ideas or would like to help, please jump in, there is plenty of room in the O2 pool.
Joe

Hello Joe

I've dropped one PM to you the last week about my consult on SGI, to see if anybody could help us, but was no reply at all.

I'm pretty sure that the assembler mod on the PROM can't be so difficult, but this needs to be implemented with a real knowledge of the O2 internals at boot time.

So, the first group of thing neededs:

1) registers to look in for the data on the RM7900 CPU chip
2) new values to be add to the ip32PROM's list
3) RM7900 L2 cache memory location

Second group of thing neededs:

1) A disassembled copy of the IP32 PROM
2) A dissasembled copy of the IP32 Kernel

Third group of thing neededs:

1) An Hex Editor (Easy! ...GHex is your friend!)
2) A test-bed

For the fourth thing; I mean, anyone knowing about reverse engineering of PROM and kernel, maybe you'll find interesting the following thread:
http://forums.nekochan.net/viewtopic.php?t=2443&highlight=octane+linux+kernel+port

I'll try to get disassembled the PROM at least, to see if I can get with the needed strings; but it is more than nothing on a curiosity base!

Diego Wrote

So, the first group of thing neededs:

1) registers to look in for the data on the RM7900 CPU chip
2) new values to be add to the ip32PROM's list
3) RM7900 L2 cache memory location

Items 1 and 2 are not a problem, I know the new string values and which registers they are in. I was thinking that for the time being the L2 cache could be turned off. No L2 would hurt the overall performance of the system, but it might make first steps of the ip32PROM mod easier to get going. Once we had a running system, the cache issue could be addressed (no pun intended).

As far as the test bed goes, I still have the 2 835MHz RM7900 chips and I would be happy to send 1 to someone that wanted to put it on an O2 CPU module to do the testing.

Joe

chicago-joe wrote: Items 1 and 2 are not a problem, I know the new string values and which registers they are in. I was thinking that for the time being the L2 cache could be turned off. No L2 would hurt the overall performance of the system, but it might make first steps of the ip32PROM mod easier to get going. Once we had a running system, the cache issue could be addressed (no pun intended).

I'm very interested on such mod; my only question is:

What is the lastest state reached by an O2 currently with the RM7900 mounted on his motherboard? ...I mean, what is the "last screen", before the die?

Can be used an R5K motherboard, just changing the CPU module by yours to debug the PROM and Kernel?

chicago-joe wrote: As far as the test bed goes, I still have the 2 835MHz RM7900 chips and I would be happy to send 1 to someone that wanted to put it on an O2 CPU module to do the testing.
Joe

I could be interested; even paying any costs that you could have for this. My other question: could this burn on fire my motherboard?

Seriously! Could it be possibly? ...Or it is just a matter of:

"No O2(s) were harmed during this debugging" ???

Diego wrote: I'm very interested on such mod; my only question is:

What is the lastest state reached by an O2 currently with the RM7900 mounted on his motherboard? ...I mean, what is the "last screen", before the die?

I could be interested; even paying any costs that you could have for this. My other question: could this burn on fire my motherboard?

Seriously! Could it be possibly? ...Or it is just a matter of:

"No O2(s) were harmed during this debugging" ???

Diego,

I had the 835MHz chip mounted on a standard (older style) RM5271-300MHz based O2 CPU board. When the computer was started, the CPU module would make it most of the way through the POST and then stop the system with a red blinking LED. I never tried starting the system with a serial console attached, it never crossed my mind. After I ran out of patience with the 835 chip, I had it removed and a 600MHz chip put on and that CPU module is still running today so I don't see any problems with it "killing" the O2 CPU module or the motherboard. If I were going to use the 835 chip again, I would use a socket and header setup with a 600 chip on one header and an 835 chip on another and the PROM chip in a socket so I could have a useful O2 when I wasn't working on modding the ip32PROM program. The socket/header setup would cost more up front, but would be far more practicle in the long run.

FWIW - These are the same issues with the SandCraft CPU chips. If the ip32PROM could be hacked to make the PMC-Sierra RM7900 chips work, then the SandCraft chips could be made to work as well.

Joe