lepidotós wrote on 2022-11-09, 08:52:

Anybody know some things I should do with it?

I guess it's quite a rare system and I wonder how well my SIV utility would report it, it was OK for a TM8000 on 2010-03-11 @ 15:12:40, but since then SIV has changed a lot.

Please will you try SIV32X and post the two Menu->File->Save Local files?

Oh, for sure, here you are. Hope it helps!
I'm confident that it's not 100% accurate, but it seems good enough, and I'd need to spend more time with it to really know for sure.

Transmeta hardware is certainly not something you play with every day. Should be fun.

I would probably try to figure out if the CPU has any interesting strengths. 32MB of RAM goes to translation cache according to Wikipedia. Does it have any curious lag to it, indicating the code morphing's initial pass? A slow hard drive probably hides that a bit though.

I bought a second hand Nexus 9 years ago because it's another code translator CPU. That uses 256MB RAM as a cache. The CPU is very capable but it's a pretty awful tablet really. Not enough RAM and so it can not really multitask and the EMMC storage is the usual poor performing 2014 EMMC with write latency issues.

Actually yes, for some reason it seems there's a noticeable 2-3 second lag when opening up stuff like dialog boxes, but after that it runs alright. Fast enough that I've gotten the first round of benchmarks done with 3DMark2001. 640x480x16 gets 967 marks, while 1024x768x16 gets 697. I tried 32-bit, but that uses too much texture memory.

I think that might vary based on what version of CMS you have installed -- mine reports that it's only allocating 24MB, leaving 232MB to Windows XP. It's also reporting that there's two RAM slots, but looking at an Actius MM20 board, I only see one, so either of three options are possible: I am bad at searching, the MM20 isn't an exact PC-MM2, or the RAM reserved for CMS is being seen by SIV as a second RAM socket. Any places you think I should start in looking for any hidden strengths?

Honestly, I can relate even with normal ARM hardware, my phone (Moto G Play 2021) is worse off than an LG flagship from 6 years ago. At least with old PCs, you can easily find matching old software.

lepidotós wrote on 2022-11-09, 17:32:

I'm confident that it's not 100% accurate, but it seems good enough, and I'd need to spend more time with it to really know for sure.

lepidotós wrote on 2022-11-09, 20:07:

It's also reporting that there's two RAM slots, but looking at an Actius MM20 board, I only see one, so either of three options are possible: I am bad at searching, the MM20 isn't an exact PC-MM2, or the RAM reserved for CMS is being seen by SIV as a second RAM socket.

Thank you, SIV did OK, but was slightly confused by the RAM. It thought there was both a DIMM and on-board RAM which is why it had space for two DIMMs on the initial screen. When I looked at [ Machine ] I could see there is only one DIMM so I have just adjusted things and 5.67 Beta-09 should only have one DIMM. I also tidied up a few loose ends.

-LAME MP3 encoding
-video playback tests. Can it handle even 480p H.264? It has SSE2 so maybe. I would probably try Media Player Classic with KLite Codec Pack Basic for XP, or maybe VLC.
-Quake 2 or 3 at a variety of resolutions. Start as low as possible to test the CPU primarily and go up and see how things scale.

Will do tomorrow. For now here's a bunch of synthetic benchmark results. Nowhere near exhaustive but it covers a not insignificant range of the benchmark suite for 9x that Phil compiled together. No, I didn't know how to save the 3DMark2000 results in more detail, unfortunately.

I recently learned the TM8800 is SSE3, and the Actius MP30/50/70s came with them. I don't have the money, and I have way more than enough laptops already, but if anyone else would like to get in on the Transmeta action, that's a good place to start since Crusoe is very plentiful and pretty well documented by now I assume. At 1.6GHz, and with a Radeon 7500 instead of a Radeon DDR and definitely at least getting up to 1GB RAM, it seems like a much better gaming machine.

To be honest, the one machine I really wish they made with the Efficeon was a standard-sized laptop, like 12" or 14". Stuff it with a big, conventional-for-'04 sized battery like the one on my Inspiron 2200, and it would likely last a good 10+ hours with zero modifications. I wouldn't complain about a Micro ATX sized board (or even a second Mini ITX one that can actually be found) though either.

lepidotós wrote on 2022-11-09, 20:07:

being seen by SIV as a second RAM socket

I have now made all the adjustments, please will you try the attached SIV32X 5.67 Test-09 and generate new save files so I can see how well the changes work.

It seems that the DIMM SPD is within the PCI config space of device [ 0 - 00 - 1 ] so I have adjusted [ SPD ] in light of this so it should now work.

Here you go, hope it helps. It seems good to me.

Something I noticed is that my system is reporting as AGP disabled, and WinBench 2000 says the Radeon is PCI. I've made sure the AGP driver is installed and am re-running the benches to see if that did anything now.

Edit: yeah, it's absolutely flying now that AGP is enabled. Good 40+ FPS in demos that used to be 10FPS.

lepidotós wrote on 2022-11-10, 20:27:

Something I noticed is that my system is reporting as AGP disabled

Thank you, all the changes worked almost ideally and the only minor issue is that on [ SPD ] then 0_50 should be PCI [ 0 - 00 - 1 ]

SIV reports AGP as the Transmeta TM8000 Northbridge has an AGP Capability as does the AMD Radeon Mobility M6 LY [GPU-0]. It's quite common to have these of GPUs that can be either AGP or PCI, they are always disabled the the GPU is PCI. Press [ PCI Bus ] , then press [ 0 - 00 - 0 ] and/or [ 1 - 00 - 0 ] and you will see the decoded Standard Capabilities

lepidotós wrote on 2022-11-10, 20:27:

Edit: yeah, it's absolutely flying now that AGP is enabled.

I have just uploaded SIV 5.67 Beta-09 which should fix [ SPD ] to say PCI [ 0 - 00 - 1], does it please?

It should also now report the AGP enabled state.

Looks like it says so. However, I noticed the Machine tab suggests it's a Crusoe, and the main page says it's 90nm, but TM8600 is supposed to have been fabbed at 130nm; TM8800 is 90nm.

lepidotós wrote on 2022-11-11, 20:42:

I noticed the Machine tab suggests it's a Crusoe, and the main page says it's 90nm, but TM8600 is supposed to have been fabbed at 130nm; TM8800 is 90nm.

It's a case of Garbage In -> Garbage Out, SIV is simply reporting what the BIOS setup in the SMBIOS. To fix this the BIOS would need to be fixed.

The code is as below, so it's stepping < 4 that is 130nm

1      if( cpu->stepping < 4 )                                                   //
2      {                                                                         //
3        cpu->technology = sgb->slb.tec_130;                                     //
4        cpu->skt_def    = 1;                                                    //
5      }                                                                         //
6      else                                                                      //
7      {                                                                         //
8        cpu->technology = sgb->slb.tec_090;                                     //
9        cpu->skt_def    = 2;                                                    //
10      }                                                                         //

Fair enough, that makes sense.

Anyway, I'm currently heading out for something but am currently in the process of testing a few games. I'll finish them when I get back, and get to doing the CPU benchmarks suggested, plus a compiling test with the Digital Mars D compiler.

Half-Life Timedemo -- Software (Sept. 1997)

c2a4 ----- c2a4a ----- c2a4b ----- avg. ----- 320x200
77.9 76.2 74.9 76.333
c2a4 ----- c2a4a ----- c2a4b ----- avg. ----- 640x400
37.5 38.2 38.7 38.133
c2a4 ----- c2a4a ----- c2a4b ----- avg. ----- 800x600
25.2 24.0 25.5 24.9
c2a4 ----- c2a4a ----- c2a4b ----- avg. ----- 1024x768
15.5 16.2 16.2 15.967

Half-Life Timedemo -- nGlide (Sept. 1997)

c2a4 ----- c2a4a ----- c2a4b ----- avg. ----- 320x200
65.8 58.2 66.6 63.533
c2a4 ----- c2a4a ----- c2a4b ----- avg. ----- 640x400
62.9 57.0 64.3 61.4
c2a4 ----- c2a4a ----- c2a4b ----- avg. ----- 800x600
47.1 49.0 52.1 49.4
c2a4 ----- c2a4a ----- c2a4b ----- avg. ----- 1024x768
32.5 35.4 37.8 35.233

Quake III Arena Timedemo -- OpenGL -- default settings

demo001 ----- demo002 ----- avg. ----- 320x240
50.8 48.2 46.8
demo001 ----- demo002 ----- avg. ----- 640x480
46.0 45.8 45.9
demo001 ----- demo002 ----- avg. ----- 800x600
44.1 41.9 43..0
demo001 ----- demo002 ----- avg. ----- 1024x768 (with significant fluctuation)
32.6 32.9 32.7

Quake III Arena Timedemo -- OpenGL -- lowest settings

demo001 ----- demo002 ----- avg. ----- 320x240
56.3 52.5 54.4
demo001 ----- demo002 ----- avg. ----- 640x480
53.3 50.0 51.6
demo001 ----- demo002 ----- avg. ----- 800x600
51.2 50.0 50.6
demo001 ----- demo002 ----- avg. ----- 1024x768 (with significant fluctuation)
39.3 41.3 40.3

Quake III Arena Gameplay -- OpenGL -- q3dm17 -- default settings

320x240 ----- 640x480 ----- 800x600 ----- 1024x768
45 - 60 30 - 55 25 - 50 20 - 50
There were some moments of a good 10-20 extra fps per resolution, but just moments.

Quake III Arena Gameplay -- OpenGL -- q3dm17 -- lowest settings

320x240 ----- 640x480 ----- 800x600 ----- 1024x768
45 - 65 45 - 60 30 - 60 25-55
There were some moments of a good 10-20 extra fps per resolution, but just moments.

Quake II runs well up to 800x600. OpenGL runs ever so slightly slower at 320x240 than software, but better at 800x600; 320x240 OpenGL would likely be the ideal way to run Quake II. I'd be using more concrete numbers here but timedemo 1 doesn't work despite the Internet saying otherwise, and not using a source port.

Banzai Bug runs alright, about 45-50 at the lowest res (320x200/240?) and 30s dipping to 25 at the higher (640x480 I assume).

SiN Demo, software renderer: 320x240 and 640x480 work fine, but 800x600 may have severe visual issues if coming to it from 640x480 fullscreen and 1024x768 drops down to windowed mode. 800x600 is rather sluggish.
OpenGL: Runs fine. 800x600 runs much better, and 1024x768 still windows the game.

Jazz Jackrabbit 3D fails to run, as does O.D.T.

Team Fortress 2: Invasion (2003) runs between 10 and 20 fps at 640x480 on software, and much the same at 800x600 -- but drastic changes cause frame dips for a second or two. 1024x768 shows very similar performance, and generally speaking, this version of Source seems remarkably consistent in its software performance. Direct3D seems to be pretty much similar, but with a high average of around 25fps. Generally speaking, it seems to be a fairly consistent game -- either that, or the Efficeon is the bottleneck, and not the Radeon. I would have used a copy of the HL2 leak, but I didn't have one on hand and didn't want to wait the probably 18 hours it would take to download, no exaggeration.

LAME encoded a 512kbps, 13mb copy of Careless Whisper into MP3 in 18 seconds.

480p H.264 video decodes perfectly (using The Molten Salt Reactor Experiment as my test video -- 658x480), at least ~30fps video, I don't have any 60 on hand. It however does struggle very visibly with 720p, displaying a frame every 10 seconds or so, a very steep drop off.

D test has to be put on the backburner for now while I figure out how to get this laptop to actually use the internet connection I'm sharing over the ethernet cable, since it won't see my mobile hotspot despite it working just fine with Mac OS 9. It's unfortunate, but it'll get done sooner or later.

I grabbed Odin and threw it on the net disk (my Acer laptop I'm using), and am using it to image the Windows XP install in case there's anything interesting on it. To image the 7,207 MB that are in use, it took 4:50:53 with an average transfer speed around 450-500KB/s, compressed with BZip2. Verifying started with a climb up to about 1.7MB/s and then gradually dropped to 1.0MB/s before resting at 1.2MB/s, completing in 1:33:20.

Now I'm contemplating what the next OS I should install be. I guess probably ReactOS since I have the media downloaded already. The laptop can boot off USB, thankfully. I'll be reinstalling ROS as its permanent OS (or maybe dual boot it with Haiku) once it's SSD modded.

lepidotós wrote on 2022-11-13, 01:59:

I guess probably ReactOS

Good luck and once ReactOS is installed please will you generate the SIV Menu->File->Save Local files for me? I have never tried SIV on ReactOS and I have often wondered what would happen. Which Windows version is ReactOS the same as?

Update: I felt I was being lazy so booted ReactOS-0.4.14-Live.iso on my P4 system. SIV started OK and reported V5.02, but only one of the two threads, that's as far as I got as when I tried to access the SIV menus I got a BSOD 🙁

red-ray wrote on 2022-11-13, 13:33:

Which Windows version is ReactOS the same as?

It's going for Windows Server 2003/XP x64 once the amd64 port is brought up.

red-ray wrote on 2022-11-13, 13:33:

Update: I felt I was being lazy so booted ReactOS-0.4.14-Live.iso on my P4 system. SIV started OK and reported V5.02, but only one of the two threads, that's as far as I got as when I tried to access the SIV menus I got a BSOD 🙁

It might or might not work better on the 0.4.15 nightlies, and that's what I have downloaded anyway as they're much more stable with live booting as 0.4.14 is a year old now.

lepidotós wrote on 2022-11-13, 18:39:

It's going for Windows Server 2003/XP x64 once the amd64 port is brought up.

On the Transmeta Efficeon? AFAIK it does not have amd64 so assume this must be another system.