MiSTer FPGA Forum

[EDIT]
I renamed the thread, since It's more about the 100mhz mode now, it has been stable for weeks for me, in dos, windows 98 and even windows Me. It would be great if people could try and see if it stable for them as well. Maybe some recent optimisations made by the devs made it stable, or maybe it depends on the Mister's setup, or maybe it's just a luck of the draw regarding the Cyclone V.
[EDIT]

I would like to have access to some debug features, but I have absolutely no idea how to access this mode without recompiling the core.

I was thinking about doing the change and recompiling the core with my modification, but that is way more complicated than I expected.
If anyone has this info, because I couldn't figure it out by myself. I hope you all had a great weekend, cheers!

Brilliant, I would like to know too

Since no good soul could help me with that I recompiled the core myself with the debug menu activated, it wasn't as simple as I wanted it to be, but it works ^^ If some people not very knowledgeable with Quartus want me to do a tutorial about it, let me know.

What does debug mode give you?

Neocaron wrote: ↑Sun Nov 07, 2021 3:33 pm Since no good soul could help me with that I recompiled the core myself with the debug menu activated, it wasn't as simple as I wanted it to be, but it works ^^ If some people not very knowledgeable with Quartus want me to do a tutorial about it, let me know.

Yes please

If all you are doing is trying to set the core to Debug so you can bump it up to 100MHz you can just use this /config/ file and swap them out. Note, 100MHZ is unstable outside of DOS4GW protected game modes and not worth it as a daily driver...

flynnsbit wrote: ↑Sun Nov 07, 2021 9:19 pm If all you are doing is trying to set the core to Debug so you can bump it up to 100MHz you can just use this /config/ file and swap them out. Note, 100MHZ is unstable outside of DOS4GW protected game modes and not worth it as a daily driver...
AO486.zip

Where do you put the config file? Thanks for that though can you enter any number, or is it 100 only? (Was thinking 95,96etc...)

I actually just wanted to do incremental upgrades on the CPU up until it crashes in the games I use on a regular basis, Like Conquest of the New World or Alien Trilogy, which are almost playable but not quite there yet, and I can confirm that in my case with the active cooling I have, It does give improve performance on those 2 games. I wouldn't recommand it to anyone though, it's just a fun experiment and I force my fan at full speed to prevent any overheating problem.

I'm also looking at a way inside the core to use a custom refresh rate for the screen at 15fps, 20fps or 30fps.
It's probably not even compatible with my screen anyway, but I'm exploring the code and trying to figure things out, it's pretty fun actually

flynnsbit wrote: ↑Sun Nov 07, 2021 9:19 pm If all you are doing is trying to set the core to Debug so you can bump it up to 100MHz you can just use this /config/ file and swap them out. Note, 100MHZ is unstable outside of DOS4GW protected game modes and not worth it as a daily driver...
AO486.zip

Neocaron wrote: ↑Sun Nov 07, 2021 11:36 pm I actually just wanted to do incremental upgrades on the CPU up until it crashes in the games I use on a regular basis, Like Conquest of the New World or Alien Trilogy, which are almost playable but not quite there yet, and I can confirm that in my case with the active cooling I have, It does give improve performance on those 2 games. I wouldn't recommand it to anyone though, it's just a fun experiment and I force my fan at full speed to prevent any overheating problem.

You can try and compile the core in Quartus and try different seed values and perhaps you'll get one that is stable at 100mhz.

The thing you need to ask yourself is if 90mhz vs 100mhz is going to make any significant difference worth all the compiling and testing effort.

Maybe the answer is yes if you have a fast enough machine, but on my fastest computer (10th gen I7) it takes 25min to compile...

Maybe throw multiple machines at it

bbond007 wrote: ↑Tue Nov 09, 2021 1:01 am

flynnsbit wrote: ↑Sun Nov 07, 2021 9:19 pm If all you are doing is trying to set the core to Debug so you can bump it up to 100MHz you can just use this /config/ file and swap them out. Note, 100MHZ is unstable outside of DOS4GW protected game modes and not worth it as a daily driver...
AO486.zip

Neocaron wrote: ↑Sun Nov 07, 2021 11:36 pm I actually just wanted to do incremental upgrades on the CPU up until it crashes in the games I use on a regular basis, Like Conquest of the New World or Alien Trilogy, which are almost playable but not quite there yet, and I can confirm that in my case with the active cooling I have, It does give improve performance on those 2 games. I wouldn't recommand it to anyone though, it's just a fun experiment and I force my fan at full speed to prevent any overheating problem.

You can try and compile the core in Quartus and try different seed values and perhaps you'll get one that is stable at 100mhz.

The thing you need to ask yourself is if 90mhz vs 100mhz is going to make any significant difference worth all the compiling and testing effort.

Maybe the answer is yes if you have a fast enough machine, but on my fastest computer (10th gen I7) it takes 25min to compile...

Maybe throw multiple machines at it

I am actually haha
I have a 3900x, and it takes 40min to compile everytime To add some nuisance to this process, for some reason I have to uninstall Quartus every time I do it... otherwise it fails to compile at the 29min mark with a routine congestion error. I already recompiled it 8 times trying different stuff and trying to understand some part of the code, but it's not easy. Still I enjoy it very much, and like I said in my case the 100mhz oc does work with Alien Trilogy, Conquest of the new world and Caesar III. I'm gonna keep experimenting, and share how it goes. ^^

Neocaron wrote: ↑Tue Nov 09, 2021 3:10 am I am actually haha
I have a 3900x, and it takes 40min to compile everytime To add some nuisance to this process, for some reason I have to uninstall Quartus every time I do it... otherwise it fails to compile at the 29min mark with a routine congestion error. I already recompiled it 8 times trying different stuff and trying to understand some part of the code, but it's not easy. Still I enjoy it very much, and like I said in my case the 100mhz oc does work with Alien Trilogy, Conquest of the new world and Caesar III. I'm gonna keep experimenting, and share how it goes. ^^

I have no idea if this will work, but Quartus does not seem to be particularly well threaded, so maybe try to do multiple compiles simultaneously on a VMs running on that 3900x... maybe you'll find the golden 100mhz core.

So I fixed my issue by downloading Quartus 17, instead of 21 and now it works flawlessly and it takes only 30min to compile and I don't need to reinstall it evry time, yeah me! ^^
I'm stress testing the core at 100mhz using some heavy tasks in windows 98 (game install, zip files, gaming on Civ2, Ceasar III and so far still not a single crash after hours of tests.
To make the 100mhz mode actually work I had to swap it with the 15mhz mode, since I couldn't figure out how to add a new preset. Not optimal, but that work as an experiment. I also decided to swap it with the 15mhz mode, because it's very easy to tell if you are running at 15 or 100, that' for sure. XD

I also made sure it was working in dxdiag by doing the buffer test, and checking the difference between every mode and it does work and was faster than 90mhz. I also managed to launch scummvm, and while it does work in win 98 with sound! it's 10 time slower than Dos even at 100mhz, so absolutely not usable unless there is a magic box to tic in the sofware somewhere. Anyway it's fun, I should also mention that I have Antonio Villena's version of the Mister and that the fan is running maxed out as a safety measure.

I just realized that ScumVM is a thing by itself on the Mister, I'm gonna see how it performs. I'm very curious ^^

bbond007 wrote: ↑Tue Nov 09, 2021 4:10 am I have no idea if this will work, but Quartus does not seem to be particularly well threaded, so maybe try to do multiple compiles simultaneously on a VMs running on that 3900x... maybe you'll find the golden 100mhz core.

I wish I had known about your amazing script before, I would not have tried so hard to make Versailles 1685 run on the Ao486. It runs flawlessly on your script, and I have a big library of this kind of games, so it makes me very happy to be able to play them on the Mister. So thanks again for making this a reality on the Mister! Do you know how to fix the tearing though? It pretty bad on Versailles, even with vsync on it's there. And also I want to be able to use the integer scale LCD filter that I use on the Ao486, any idea? Can I inject it in Scummvm somehow?

Thanks again, and don't listen to people casting doubt on your work, what matters is how good it is at the end, and for what it does, your port of Scummvm is amazing, because you just can't play any of those games at this speed on the Mister.

Neocaron wrote: ↑Wed Nov 10, 2021 12:08 am I wish I had known about your amazing script before, I would not have tried so hard to make Versailles 1685 run on the Ao486.

Then you would have not learned what you did about Quartus and the ao486 core

With any luck, that knowledge will be come in handy for you in the future...

Neocaron wrote: ↑Wed Nov 10, 2021 12:08 amDo you know how to fix the tearing though? It pretty bad on Versailles, even with vsync on it's there. And also I want to be able to use the integer scale LCD filter that I use on the Ao486, any idea? Can I inject it in Scummvm somehow?

I think (just a guess) for the tearing issue, multiple pages would need to be added to the framebuffer implementation. I imagine the menu.rbf would need changed as well as the framebuffer driver in the Linux kernel. For filters (again just a guess) I would imagine the menu.rbf would need some work.

bbond007 wrote: ↑Wed Nov 10, 2021 12:40 am

Neocaron wrote: ↑Wed Nov 10, 2021 12:08 am I wish I had known about your amazing script before, I would not have tried so hard to make Versailles 1685 run on the Ao486.

Then you would have not learned what you did about Quartus and the ao486 core

With any luck, that knowledge will be come in handy for you in the future...

Neocaron wrote: ↑Wed Nov 10, 2021 12:08 amDo you know how to fix the tearing though? It pretty bad on Versailles, even with vsync on it's there. And also I want to be able to use the integer scale LCD filter that I use on the Ao486, any idea? Can I inject it in Scummvm somehow?

I think (just a guess) for the tearing issue, multiple pages would need to be added to the framebuffer implementation. I imagine the menu.rbf would need changed as well as the framebuffer driver in the Linux kernel. For filters (again just a guess) I would imagine the menu.rbf would need some work.

I do have a dirty solution for the filters... I'll use my Ossc as a scanline pass if there isn't any easy fix for that at the moment.
Oh so the tearing options are just for show currently in Scummvm?

And you are absolutely right, I'm still very happy about having a 100mhz ao486, Alien trilogy is in the low 10s now, before it was around 7, and Conquest of the New World has way smoother combat and scrolling thanks to the boost. My devil inside wants to push it even furster to 105 or 110mhz, but I'll restrain myself, I don't want to kill my Mister in a ship shortage world, that would be hmm... wrong

And now I know how to compile and modify a core.

So I've been experimenting pretty hard with the ao486 core at 100mhz and I don't know if there is any form of "luck of the draw" regarding the OC of the Cyclone V but on Antonio Villena's Mister, I didn't get a single crash playing over 20 games for hours. And this is true in windows 98, windows me, and Dos.
So maybe this should be visible for people to try out in a experimental section, this way people know it might not work for everyone but at least the option is there. For me on some of those 95/96 games, the small increase was the difference between playble and not playable. Thoughts?

Neocaron wrote: ↑Tue Nov 23, 2021 1:26 pm So I've been experimenting pretty hard with the ao486 core at 100mhz and I don't know if there is any form of "luck of the draw" regarding the OC of the Cyclone V but on Antonio Villena's Mister, I didn't get a single crash playing over 20 games for hours. And this is true in windows 98, windows me, and Dos.
So maybe this should be visible for people to try out in a experimental section, this way people know it might not work for everyone but at least the option is there. For me on some of those 95/96 games, the small increase was the difference between playble and not playable. Thoughts?

Post your compile and we can test it across misters. Maybe you hit a stable build.

flynnsbit wrote: ↑Tue Nov 23, 2021 1:52 pm

Neocaron wrote: ↑Tue Nov 23, 2021 1:26 pm

So I've been experimenting pretty hard with the ao486 core at 100mhz and I don't know if there is any form of "luck of the draw" regarding the OC of the Cyclone V but on Antonio Villena's Mister, I didn't get a single crash playing over 20 games for hours. And this is true in windows 98, windows me, and Dos.
So maybe this should be visible for people to try out in a experimental section, this way people know it might not work for everyone but at least the option is there. For me on some of those 95/96 games, the small increase was the difference between playble and not playable. Thoughts?

Post your compile and we can test it across misters. Maybe you hit a stable build.

I don't think that I did, it's sloppy I didn't know how to add it back properly so I just modified this:

(* romstyle = "logic" ) wire [27:0] clk_rate[8] = '{90000000, 15000000, 30000000, 56250000, 100000000, 100000000, 100000000, 100000000 };
( romstyle = "logic" *) wire [17:0] speed_div[8] = '{ 'h0505, 'h1e1e, 'h0f0f, 'h0808, 'h20504, 'h20504, 'h20504, 'h20504 };

To this

(* romstyle = "logic" ) wire [27:0] clk_rate[8] = '{90000000, 100000000, 30000000, 56250000, 100000000, 100000000, 100000000, 100000000 };
( romstyle = "logic" *) wire [17:0] speed_div[8] = '{ 'h0505, 'h20504, 'h0f0f, 'h0808, 'h20504, 'h20504, 'h20504, 'h20504 };

And in the gui
"D2D1P2O56,CPU Clock,90MHz,15MHz,30MHz,56MHz;",
to
"D2D1P2O56,CPU Clock,90MHz,100MHz,30MHz,56MHz;",

I replaced the lowest one to be sure it wasn't placebo, this way I can tell the difference easily between 15 and 100mhz.

So basically I just used the value that was reserved for the debug 100mhz and I did no change to anything else. If it was unstable, it should be unstable for me unless some recent changes to the core made it stable now.
How often would it crash for you? Like could you play a game for 3hours straight? Could the amp of the power supply affect it?

I can share it here but please be careful, It was literally the first time I recompiled something, I have no experience with coding at all and to even find the right things to change, it was just guess work. My point is more about the fact that it has been stable for me for almost 2 weeks, so I think people should at least be allowed to have the option to try it, with a warning not to do it without a fan for exemple.

And in case some people wonder where you do the changes, it's only in the ao486.sv file. Didn't touch anything else.

Do you still need a paid version of Quartus to compile? I thought I remembered that Ao486 included some IPs that weren't included in the free version.

I think you can use the free version. The IP was for the UART but an open source one is in use now.

dave18 wrote: ↑Tue Nov 23, 2021 8:34 pm Do you still need a paid version of Quartus to compile? I thought I remembered that Ao486 included some IPs that weren't included in the free version.

I only used the free 17 version of Quartus with the cyclonev-17.1.0.590.qdz and the model sim setup
For some reasons I had fitting errors during the compilation, so that's maybe because of the free version? Anyway, after trying and trying again, I was able to compile successfully.

So anyone tried it yet? Curious to know if it comes down to my Mister or if some recent changes to the core made 100mhz stable.

Neocaron wrote: ↑Tue Nov 09, 2021 3:10 am

bbond007 wrote: ↑Tue Nov 09, 2021 1:01 am

flynnsbit wrote: ↑Sun Nov 07, 2021 9:19 pm If all you are doing is trying to set the core to Debug so you can bump it up to 100MHz you can just use this /config/ file and swap them out. Note, 100MHZ is unstable outside of DOS4GW protected game modes and not worth it as a daily driver...
AO486.zip

Neocaron wrote: ↑Sun Nov 07, 2021 11:36 pm I actually just wanted to do incremental upgrades on the CPU up until it crashes in the games I use on a regular basis, Like Conquest of the New World or Alien Trilogy, which are almost playable but not quite there yet, and I can confirm that in my case with the active cooling I have, It does give improve performance on those 2 games. I wouldn't recommand it to anyone though, it's just a fun experiment and I force my fan at full speed to prevent any overheating problem.

You can try and compile the core in Quartus and try different seed values and perhaps you'll get one that is stable at 100mhz.

The thing you need to ask yourself is if 90mhz vs 100mhz is going to make any significant difference worth all the compiling and testing effort.

Maybe the answer is yes if you have a fast enough machine, but on my fastest computer (10th gen I7) it takes 25min to compile...

Maybe throw multiple machines at it

I am actually haha
I have a 3900x, and it takes 40min to compile everytime To add some nuisance to this process, for some reason I have to uninstall Quartus every time I do it... otherwise it fails to compile at the 29min mark with a routine congestion error. I already recompiled it 8 times trying different stuff and trying to understand some part of the code, but it's not easy. Still I enjoy it very much, and like I said in my case the 100mhz oc does work with Alien Trilogy, Conquest of the new world and Caesar III. I'm gonna keep experimenting, and share how it goes. ^^

40 minutes on a 3900x?? I'm on a 3700x and I'm well down into the 26-28 minute realm to compile ao486! I'm not sure how well Quartus scales beyond 8c16t CPUs but I'd expect better from your CPU. I'm on Linux which tends to be better with multithreading but even when I was still on windows I was getting about that time if memory serves.

On a different note I must have a good Cyclone chip as it is rock solid at 100mhz in any testing DOS/Windows.

Neocaron wrote: ↑Wed Nov 24, 2021 6:35 am

dave18 wrote: ↑Tue Nov 23, 2021 8:34 pm Do you still need a paid version of Quartus to compile? I thought I remembered that Ao486 included some IPs that weren't included in the free version.

I only used the free 17 version of Quartus with the cyclonev-17.1.0.590.qdz and the model sim setup
For some reasons I had fitting errors during the compilation, so that's maybe because of the free version? Anyway, after trying and trying again, I was able to compile successfully.

No it's a setting in Quartus. The default is to favour performance of the logic over the speed of the compile. Can't remember exactly where in settings the setting for the fitter operation is but it's that one, with that changed it'll compile.

Blitzwing wrote: ↑Mon Nov 29, 2021 12:55 am

Neocaron wrote: ↑Wed Nov 24, 2021 6:35 am

dave18 wrote: ↑Tue Nov 23, 2021 8:34 pm Do you still need a paid version of Quartus to compile? I thought I remembered that Ao486 included some IPs that weren't included in the free version.

I only used the free 17 version of Quartus with the cyclonev-17.1.0.590.qdz and the model sim setup
For some reasons I had fitting errors during the compilation, so that's maybe because of the free version? Anyway, after trying and trying again, I was able to compile successfully.

No it's a setting in Quartus. The default is to favour performance of the logic over the speed of the compile. Can't remember exactly where in settings the setting for the fitter operation is but it's that one, with that changed it'll compile.

Oh so that's why it would randomly compile after multiple tries? Is it in the preference settings?
Anyone tried to run the 100mhz mode to see if its stable for other people than me?

Neocaron wrote: ↑Mon Nov 29, 2021 11:46 am

Blitzwing wrote: ↑Mon Nov 29, 2021 12:55 am

Neocaron wrote: ↑Wed Nov 24, 2021 6:35 am

I only used the free 17 version of Quartus with the cyclonev-17.1.0.590.qdz and the model sim setup
For some reasons I had fitting errors during the compilation, so that's maybe because of the free version? Anyway, after trying and trying again, I was able to compile successfully.

No it's a setting in Quartus. The default is to favour performance of the logic over the speed of the compile. Can't remember exactly where in settings the setting for the fitter operation is but it's that one, with that changed it'll compile.

Oh so that's why it would randomly compile after multiple tries? Is it in the preference settings?
Anyone tried to run the 100mhz mode to see if its stable for other people than me?

When my fiancée gets off the computer (doing university work) I'll have a look. I'm not a developer so it was all guess work and following what the errors were that lead me to the setting, I'm not sure how the GitHub version is compiled whether they do multiple attempts like you did or they just change the setting like me maybe mine is stable because I change that setting!?

Yes I've tried it multiple times, to the point where whenever a new version of the core gets pushed out that I will compile my own with the 100mhz OC settings menu opened up... I don't have the greatest of testing regimes for it but it seems pretty stable in both DOS and Windows 95 games for me, never had it randomly lock up or working games unexpectedly crash.

I'll edit this once I get the information.

Awesome, glad to see I'm not the only one having some luck with 100mhz mode. Over 3 weeks using it, and still no crash or lock up on my end.
I think we need more people to test it out though and report back.

Neocaron wrote: ↑Tue Nov 30, 2021 1:27 pm Awesome, glad to see I'm not the only one having some luck with 100mhz mode. Over 3 weeks using it, and still no crash or lock up on my end.
I think we need more people to test it out though and report back.

Reporting back. Now I had used ao486 with the OC in DOS and Windows 95/98 games, benchmarks etc, and it was always rock solid. I compiled from the most recent code from the Github repository and it falls over almost instantly now, Freedos crashes straight away, Windows 95 and 98 don't get far into booting before they throw protection errors. Don't know what has changed.

Anyway the settings I was talking about are under... Assignments - Settings - (new box opens) - Compiler Settings - Advanced Settings (Fitter) - Fitter Aggressive Routability Optimizations set it to always.

Probably worth ignoring it for now as it is crashing for me now.

Edit. I also feel like the compile is taking much longer than I remember, do you know if there has been major changes to the core recently?

My question is, how much of a performance improvement do you get going from 90Mhz to 100Mhz? I realize at these speeds every extra Mhz makes a difference, I'm just curious how much of a difference this can possibly be.

Mr. Encyclopedia wrote: ↑Tue Nov 30, 2021 11:42 pm My question is, how much of a performance improvement do you get going from 90Mhz to 100Mhz? I realize at these speeds every extra Mhz makes a difference, I'm just curious how much of a difference this can possibly be.

10%

Blitzwing wrote: ↑Tue Nov 30, 2021 10:22 pm
Anyway the settings I was talking about are under... Assignments - Settings - (new box opens) - Compiler Settings - Advanced Settings (Fitter) - Fitter Aggressive Routability Optimizations set it to always.

Probably worth ignoring it for now as it is crashing for me now.

Edit. I also feel like the compile is taking much longer than I remember, do you know if there has been major changes to the core recently?

Did you try anything with Fitter Initial Placement Seed in that same settings menu? As 42 is the "Answer to the Ultimate Question of Life, the Universe, and Everything", maybe go with that?

It's possible that Fitter Aggressive Routability Optimizations is slowing down the compile down?

Ok now I get it

In the past when I used to try and compile it it would always fail, so I changed the option mentioned above that made it work successfully. Something must have changed in ao486 that now lets it compile successfully with that option set to "Automatically/Never" but because it hasn't "decreased the design speed" the core is now crashing like most peoples, I've changed that setting back to "Always" and the core is working again.

Now it is marginally faster in 'System Info, Top Bench and Doom' as expected but it does beg the question where, if anywhere is it slowing the logic down when that option is set to always!?

Love messing with this stuff when I've got zero clue what I'm doing