Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.sys.transputer > #474 > unrolled thread

stuff on eBay and an update after many years on large T425 system

Back to article view | Back to comp.sys.transputer

Contents

  stuff on eBay and an update after many years on large T425 system James Wilson <macihenroomfiddling@gmail.com> - 2023-02-26 10:25 -0800
    Re: stuff on eBay and an update after many years on large T425 system "Mike B." <michael_bruestle@yahoo.com> - 2023-02-28 16:08 -0800
      Re: stuff on eBay and an update after many years on large T425 system "Mike B." <michael_bruestle@yahoo.com> - 2023-03-01 13:28 -0800
      Re: stuff on eBay and an update after many years on large T425 system Raymond Wiker <rwiker@gmail.com> - 2023-03-11 22:49 +0100
        Re: stuff on eBay and an update after many years on large T425 system "Mike B." <michael_bruestle@yahoo.com> - 2023-03-12 16:00 -0700
          Re: stuff on eBay and an update after many years on large T425 system Raymond Wiker <rwiker@gmail.com> - 2023-03-13 06:45 +0100
            Re: stuff on eBay and an update after many years on large T425 system "Mike B." <michael_bruestle@yahoo.com> - 2023-03-13 01:03 -0700
              Re: stuff on eBay and an update after many years on large T425 system James Wilson <macihenroomfiddling@gmail.com> - 2023-03-13 05:59 -0700
          Re: stuff on eBay and an update after many years on large T425 system "Mike B." <michael_bruestle@yahoo.com> - 2023-03-14 00:57 -0700

#474 — stuff on eBay and an update after many years on large T425 system

Hi all, firstly I'm very glad to see this group still active. I lost touch with most of the community being on/off with all things Transputer related over the years.

Back in 2006 I acquired the guts of a Whitecross WX9020 data mining system that has over 400 T425s, each with 16MB of ECC RAM. It's been in and out of storage and survived several house moves but finally in the last couple of years I've found time to start doing "serious" things with it and have T-Mandel and the Inmos Ray Tracer running on it. See https://github.com/machineroom/T-Mandel & https://github.com/machineroom/oc-ray for the code and rendering performance. The most nodes I've had working so far is 354 and you can see this here https://www.youtube.com/watch?v=wnAxSMvCssQ

The original "route in" to the network was via a system controller card, which has an optical 10Mb link to a Solaris workstation equipped with a custom interface. Sadly in storage the batteries on these cards leaked (lesson learned!) and destroyed most of the PCB. I've been busy building my own system controller replacement that provides 20Mb links over a normal copper pair. These are hooked up to my rpi based host interface. 

I have a running problem with fuses... each Transputer has all 4 links connected into the backplane as ECL pairs using 10H124/125 chips, each fed by fused +5 & -5V2 supplies. For some reason the 1A SMD fuses on the -5V2 line tend to give way, but not in the way a fuse normally goes - these drop out for very short periods (ms) which makes debugging an absolute nightmare! I guess they're breaking down. Anyway, my current activity is to replace all 400+ of the damn things which takes time and I think I'm buying the worlds remaining stock of this particular old stock fuse :)

I also have a few rather special bits that I'm not using on eBay @ https://www.ebay.co.uk/usr/machineroom including an ST20450 eval board and Transtech SCSI host interfaces. Proceeds will find more NOS fuses :-)

If folks are interested in large T425 networks I'll make an updated video with the new system controller and rpi link interface and am always happy to provide more info! 

Cheers
James

[toc] | [next] | [standalone]

#475

Hi James

The Whitecross stuff looks great. And yes I like to hear more about it.
About the INMOS occam ray-tracer - somehow I forgot about it. 

Running it on a 16 x T805G 120MHz board implemented on an Acorn CLE-215+ (XC7A200T-3 Artix-7 FPGA) the performance is really great :-)

RayTracer Image Selection
=========================

<1>  10 Sphere Image
<2>  14 Sphere Image
<3>  Cone, Ellipsoid Image
<4>  Cylinder, Sphere Image

<1> DONE, took      7.03 seconds on 16 processors.
<2> DONE, took      7.87 seconds on 16 processors.
<3> DONE, took      2.12 seconds on 16 processors.
<4> DONE, took      1.55 seconds on 16 processors.

When I run the "14 Sphere Image" the performance counters shows an FPU usage of ~43% and an CPU usage of ~75%. Not so bad.

A few more comments on my part:

To generate the network map (hardware description) for my systems I use 'rspy /OC /M'
To compile I use the transputer based toolset files from the D7205A. It's a matter of honor ;-)
To create an .mtv file and convert it to .ppm is only waste of time. The format is almost identical if not even simpler. That's why I create the ppm format right away in occam.

-Mike

[toc] | [prev] | [next] | [standalone]

#476

Find my stuff here.

https://bitbucket.org/transputer/raytracer.git

-Mike

[toc] | [prev] | [next] | [standalone]

#477

"Mike B." <michael_bruestle@yahoo.com> writes:

> Hi James
>
> The Whitecross stuff looks great. And yes I like to hear more about it.
> About the INMOS occam ray-tracer - somehow I forgot about it. 
>
> Running it on a 16 x T805G 120MHz board implemented on an Acorn
> CLE-215+ (XC7A200T-3 Artix-7 FPGA) the performance is really great :-)

This souds intriguing! Do you have any information to share about the
setup for the CLE-215+? Is the code (FPGA code for the T805G) available
anywhere?

[toc] | [prev] | [next] | [standalone]

#478

Hi Raymond

The CLE-215+ is an full populated and much faster B012 in an smaller size.

C:\Transputer\d72uni>rspy -e -f -t -zm -c4
   # Part  rt Link0 Link1 Link2 Link3 E.Test T.Test F.Test RAM@cycle
   0 T805G120  HOST   1-0   ...  16-0 passed passed passed 4K@1,16380K@2.
   1 T805G120   0-1   2-0   ...   ... passed passed passed 4K@1,16380K@2.
   2 T805G120   1-1   3-0   ...   ... passed passed passed 4K@1,16380K@2.
   3 T805G120   2-1   4-0   ...   ... passed passed passed 4K@1,16380K@2.
   4 T805G120   3-1   5-0   ...   ... passed passed passed 4K@1,16380K@2.
   5 T805G120   4-1   6-0   ...   ... passed passed passed 4K@1,16380K@2.
   6 T805G120   5-1   7-0   ...   ... passed passed passed 4K@1,16380K@2.
   7 T805G120   6-1   8-0   ...   ... passed passed passed 4K@1,16380K@2.
   8 T805G120   7-1   9-0   ...   ... passed passed passed 4K@1,16380K@2.
   9 T805G120   8-1  10-0   ...   ... passed passed passed 4K@1,16380K@2.
  10 T805G120   9-1  11-0   ...   ... passed passed passed 4K@1,16380K@2.
  11 T805G120  10-1  12-0   ...   ... passed passed passed 4K@1,16380K@2.
  12 T805G120  11-1  13-0   ...   ... passed passed passed 4K@1,16380K@2.
  13 T805G120  12-1  14-0   ...   ... passed passed passed 4K@1,16380K@2.
  14 T805G120  13-1  15-0   ...   ... passed passed passed 4K@1,16380K@2.
  15 T805G120  14-1   ...   ...   ... passed passed passed 4K@1,16380K@2.
  16 T225C120   0-3  17-C   ...   ...
  17 C004  (................................) board.c004[0]

C:\Transputer\d72uni>rspy -l
   # Part  rt Link0 Link1 Link2 Link3
   0 T805G120   48M   10M   ...   10M
   1 T805G120   10M   10M   ...   ...
   2 T805G120   10M   10M   ...   ...
   3 T805G120   10M   10M   ...   ...
   4 T805G120   10M   10M   ...   ...
   5 T805G120   10M   10M   ...   ...
   6 T805G120   10M   10M   ...   ...
   7 T805G120   10M   10M   ...   ...
   8 T805G120   10M   10M   ...   ...
   9 T805G120   10M   10M   ...   ...
  10 T805G120   10M   10M   ...   ...
  11 T805G120   10M   10M   ...   ...
  12 T805G120   10M   10M   ...   ...
  13 T805G120   10M   10M   ...   ...
  14 T805G120   10M   10M   ...   ...
  15 T805G120   10M   ...   ...   ...
  16 T225C120   10M   ...   ...   ...

The FPGA code is written and maintained by Claus Meder and not available.
My part is more the software side and the testing.

We use the CLE215 under Windows and Linux.

But the Hardware is rare, the primary setup requires JTAG+Vivado, so maybe the Kria is the easier platform (12 T805G@230MHz).
http://transputer.net/ktm/ktm.asp

-Mike

[toc] | [prev] | [next] | [standalone]

#479

"Mike B." <michael_bruestle@yahoo.com> writes:

...

> The FPGA code is written and maintained by Claus Meder and not available.
> My part is more the software side and the testing.
>
> We use the CLE215 under Windows and Linux.
>
> But the Hardware is rare, the primary setup requires JTAG+Vivado, so maybe the Kria is the easier platform (12 T805G@230MHz).
> http://transputer.net/ktm/ktm.asp
>
> -Mike

Thanks! This looks like a great option for (re-)exploring the
transputer. Looks like the CLE215 supports more nodes, but the Kria has
a higher operating frequency. Anyway, great stuff!

[toc] | [prev] | [next] | [standalone]

#480

Yes, it's really an pleasure to work on modern 64bit hardware and have pleasant compile times for greater projects.
There are some points which maybe not everyone noticed:

The Linkspeed to the host is 48MB/s. On the Kria the internal links even 61MB/s.
Of course there is now a cache in the Transputer and many performance enhancements like pfix/nfix elimination.
There are also some performance counters to analyze FPU/CPU usage.
And last but not least - It is currently being actively operated, maintained and further developed. 

If anyone wants to try the Kria, I can only repeat my offer:
Send my an email (http://transputer.net/contact.asp) and I will create an account (on the Ubuntu 22.04 os on the kria) and send you the connection infos back. The INMOS toolset C and occam compiler is installed and ready to run. 

-Mike

[toc] | [prev] | [next] | [standalone]

#481

On Monday, 13 March 2023 at 08:03:27 UTC, Mike B. wrote:
> Yes, it's really an pleasure to work on modern 64bit hardware and have pleasant compile times for greater projects. 
> There are some points which maybe not everyone noticed: 
> 
> The Linkspeed to the host is 48MB/s. On the Kria the internal links even 61MB/s. 
> Of course there is now a cache in the Transputer and many performance enhancements like pfix/nfix elimination. 
> There are also some performance counters to analyze FPU/CPU usage. 
> And last but not least - It is currently being actively operated, maintained and further developed. 
> 
> If anyone wants to try the Kria, I can only repeat my offer: 
> Send my an email (http://transputer.net/contact.asp) and I will create an account (on the Ubuntu 22.04 os on the kria) and send you the connection infos back. The INMOS toolset C and occam compiler is installed and ready to run. 
> 
> -

Hi Mike, I need rspy in my life :) Where might I find the code?

[toc] | [prev] | [next] | [standalone]

#483

On Monday, March 13, 2023 at 12:00:09 AM UTC+1, Mike B. wrote:
> We use the CLE215 under Windows and Linux. 
> But the Hardware is rare, ...

https://github.com/RHSResearchLLC/NiteFury-and-LiteFury

-Mike

[toc] | [prev] | [standalone]

Back to top | Article view | comp.sys.transputer

csiph-web