Overview Features Instructions Performance Forum Downloads Products Reseller Contact

Welcome to the Apollo Forum

This forum is for people interested in the APOLLO CPU.
Please read the forum usage manual.
VISIT APOLLO IRC CHANNEL



All TopicsNewsPerformanceGamesDemosApolloVampireCoffinReleasesLogin
Documentation about the Vampire hardware

Lock-ups On Floating Point Renders?page  1 2 

Rod March

Posts 105
30 Jun 2019 14:46


I have been doing a lot of Lightwave rendering on my V500/A2000 and and I've started getting lock-ups during renders where the whole system (including mouse) freezes and only a hard-boot with 10secs downtime will get the system up and running again (at which point it's fine). It doesn't always happen at the same point (and doesn't happen during general use), but always when rendering within the first few frames.

If I render the same frames on the integer version of Lightwave it can render for hundreds of hours without issue (albeit four times slower).

Happens with 2.11 and 2.10 cores. (No idea if it's related but I can't flash the cores from the exe - the system runs the update but bricks on reboot - have to use the USB blaster)

Anyone else get this? Any suggestions how to trouble-shoot?


Daniel Sevo

Posts 270
30 Jun 2019 17:57


How many different cores have you tried this with and can you name any that don't have this issue. I imagine isolating the behavour to pa certain core or some specific version change, would be useful info for the devs.


Roy Gillotti

Posts 350
30 Jun 2019 18:13


Rod March wrote:

I have been doing a lot of Lightwave rendering on my V500/A2000 and and I've started getting lock-ups during renders where the whole system (including mouse) freezes and only a hard-boot with 10secs downtime will get the system up and running again (at which point it's fine). It doesn't always happen at the same point (and doesn't happen during general use), but always when rendering within the first few frames.
 
  If I render the same frames on the integer version of Lightwave it can render for hundreds of hours without issue (albeit four times slower).
 
  Happens with 2.11 and 2.10 cores. (No idea if it's related but I can't flash the cores from the exe - the system runs the update but bricks on reboot - have to use the USB blaster)
 
  Anyone else get this? Any suggestions how to trouble-shoot?

Maybe share what you're rendering and the setup to see if it's repeatable on other systems.



Rod March

Posts 105
30 Jun 2019 23:55


Daniel Sevo wrote:

  How many different cores have you tried this with and can you name any that don't have this issue. I imagine isolating the behavour to pa certain core or some specific version change, would be useful info for the devs.
 

 
  Thanks Guys. I intended to start flashing the core back to 2.7, but when I started getting failures when attempting that I got a little gun-shy, so I only got as far as 2.10. I'll try 2.9 and 2.7 this evening (if I can!)
 
  A couple of times I tried to flash the core from the USB blaster and it failed, but locked up my Amiga in a very similar fashion to the renders (that may be completely irrelevant, of course, I just don't know).
 
  The classic Blade Runner example scene freezes pretty reliably on my system: EXTERNAL LINK but renders fine in integer. Sometimes I get a dozen frames in, sometimes not even one.
 
  The problem comes up running Lighwave from the switcher with VT attached and also in standalone. I'm using v5 for FP and v3.5 for INT.
 
EDIT: tested all FPU cores (2.7 - 2.11) same result all-round. 


Mike Kopack

Posts 257
01 Jul 2019 14:47


Hmmm, I'd also consider your PSU and caps on your board. What you're doing is a pretty CPU power intensive operation and it's quite possible that your board and/or PSU aren't up to the loads you're putting it under.

Do you have access to another Amiga that you can put the Vampire into, or another PSU you can swap onto your Amiga to see if you get different behavior?


Rod March

Posts 105
01 Jul 2019 23:54


Having tried every conceivable combination in my A2000, I'll try in an A500 and see what happens.

I guess I had assumed that the A2000 PSU was my most likely to be over-spec, but it's all old tech so it's entirely possible it's just wearing out.


Rod March

Posts 105
02 Jul 2019 15:05


Ok, so I'm trying to exclude IDE as the source of the problem (unlikely I know).
 
  I used to boot my Vampire system from the SCSI HDD card no problem (Oktagon 2008 card). But now whenever I disconnect the CF card from the Vamp the machine sits with a black screen for 45 secs (searching for IDE I guess) then starts booting from SCSI but shortly after gives a guru.
 
  I'm kind of banging my head against a wall trying to think what I've done differently that means the Vamp will ONLY boot off it's IDE - can't get boot menu with both mouse buttons pressed, and can't boot off SCSI HDD. (Can boot from floppy and then access SCSI HDD.)
 
  Just what do I need to tweak or change to stop the Vampire searching for it's internal IDE all the time?

Thanks in advance anyone who can offer advice - Amiga people are the best people! (and Vampire Amiga people are just that little bit better again ;) )


Santiago Gutierrez

Posts 40
04 Jul 2019 13:33


Ok, finally I did a test in Lightwave using LW 5.0r FP and yes, it freeze.
   
    Using LW 3.5 with the same scene, I can render an animation HAM6.anim without any issue.
   
    I can make a video if you guys want to see the issue.
   
    Let me know.
 
 
  Edit here........... after a few hours doing more tests.
 
  More tests.
 
  WinUAE running CoffinOS and same lightwave configuration.
  I can render animations here too.
 
  MorphOS....... same lightwave configuration.
  Running smoothly here too.
 
 


Pedro Cotter
(Apollo Team Member)
Posts 213
04 Jul 2019 20:59


Which Scene and with what Camera settings?
Just finished to render the first Blade frame without issues.

I use LW5.0r from time to time and sometimes i have lockups too but one can always continue from the last saved frame.

EXTERNAL LINK


Kamelito Loveless

Posts 57
04 Jul 2019 21:10


If you have an action replay or a low level debugger like hrtmon you could freeze the system and see which instruction is causing the problem, that way we could  narrow down the issue and try to code a small test case to fix the possible bug. Since I donít à Vampire yet I canít help more. This is of course valid if the issue is software related.


Santiago Gutierrez

Posts 40
04 Jul 2019 23:17


Kamelito Loveless wrote:

If you have an action replay or a low level debugger......

Great. I will try later. Thanks :)

Pedro, one frame is fine........Could you try to render the whole animation in low res, anim format ??


Rod March

Posts 105
04 Jul 2019 23:48


So glad you guys can reproduce the same bug.
 
  Over the week I have tried the Vamp in the A2000 in several different configurations- all cores from 2.7 - 2.11, with/without IDE/SCSI, different chip RAM boards, as well as in my A500 with a couple of different power supplies.
 
Freeze usually happens during first couple of frames (although my A500 got through 6 frames before locking up last night).


Michael R

Posts 272
05 Jul 2019 20:23


I have a similar situation. I'm building a 3d modeler from Amiga source code for RayStorm. In order to get a good feel for the original program I have to run RayStorm on Amiga. I tried the 68040 version in WinUAE but it freezes while rendering. It requires FPU.
 
  I decided to try it on my A500 with my V2 accelerator but I had to flash the FPGA core to Gold 2.11 first. I couldn't get it to flash the core using the 3.x executable. So I updated my SAGA driver to the latest. Then my mouse cursor disappeared so I can't do anything.
 
  I have a raytracer test program called MiniRay that I can use on the Amiga to test FPU rendering. It uses C libs and it works great on AROS 32bit. It's a raytracer that auto-generates 3d objects then renders the scene using floating points. It renders a 640x480 rgb image of glass sheres in about 3 sec. If it works with Vampire then the FPU is ok. If it doesn't render on Vampire then there's a problem with the fpu.
 
Since I'm not at a point where I can flash the core on my Vampire till I get my mouse cursor back I can post the MiniRay test program and source code later this evening. Then you may use it to test the FPU rendering on Vampire.
 
 
 


Kamelito Loveless

Posts 57
05 Jul 2019 21:26


Did you try WinUAE soft float it is slower but more compatible.



Michael R

Posts 272
05 Jul 2019 22:40


Kamelito Loveless wrote:

Did you try WinUAE soft float it is slower but more compatible.
 

I will try softfloat in WinUAE. I'll compile the raytracer test program for Amiga os 3.x and test that in WinUAE as well. When I get my Vampire v2 working Ill try it on Vampire also. Thanks.


Rod March

Posts 105
08 Aug 2019 00:54


Noticed this lock-up still happens with 2.12.

To the people more math-brained than I:

Is this likely due to rounding from the 64-bit precision in the V2 FPU? If so is it likely to be fixed by the full 80bit precision FPU in the V4?

Sorry if this is an illogical question mathematically speaking, I don't pretend to understand the specifics, I draw pictures for a living...


Simo Koivukoski
(Apollo Team Member)
Posts 468
08 Aug 2019 05:19


Rod March wrote:

Noticed this lock-up still happens with 2.12.living...

Did you test with the JIC or EXE?


Stephen A.

Posts 21
08 Aug 2019 05:39


i've noticed some odd behaviors of the fpu before: EXTERNAL LINK   
 
my code didn't result in machine lockups, but perhaps in the case of lightwave it does.
 
i'm just guessing, but gunnar and vojin's comments in that thread might be worth reading.
 
 
 
 


Rod March

Posts 105
08 Aug 2019 13:35


Simo Koivukoski wrote:

  Did you test with the JIC or EXE?

I think 2.12 only has a JIC.

The last couple of times I've tried to flash using EXE it hasn't worked anyway, so always JIC now.

Stephen A. wrote:

i've noticed some odd behaviors of the fpu before: EXTERNAL LINK     
  my code didn't result in machine lockups, but perhaps in the case of lightwave it does.
   
  i'm just guessing, but gunnar and vojin's comments in that thread might be worth reading.
       

The vibe I get from that thread is that things might appear a little different with 'game precision' FPU but shouldn't crash/lock-up. But yeah, I don't really know.

I'm super excited about the V4, but guess I'm just a little nervous it may not run my fave Lightwave reliably.

This could of course, be a software bug for all I know, apart from the Vamp I only really only have emu and 020 machines to test on, neither of which is in the same class as the 080.

Anyhow, have a great week people, exciting times V4 is nearly here...



Simo Koivukoski
(Apollo Team Member)
Posts 468
08 Aug 2019 13:52


Rod March wrote:

Simo Koivukoski wrote:

  Did you test with the JIC or EXE?

I think 2.12 only has a JIC.
   
The last couple of times I've tried to flash using EXE it hasn't worked anyway, so always JIC now.

Could you please test with RC2 EXE, it has fpsp080 related updates (end of page 2 on thread). Flasher EXE should work without any issues, if your file does not get corrupted. e.g invalid max transfer value with HDD.

posts 40page  1 2