TF1260 New Firmware Alpha Released!

News & updates for the upcoming 68060 accelerator

Moderators: terriblefire, Terriblefire Moderator

alenppc
Moderator Team
Moderator Team
Posts: 906
Joined: Thu Nov 08, 2018 12:59 pm

Re: TF1260 New Firmware Alpha Released!

Post by alenppc »

terriblefire wrote: Sat Jan 01, 2022 5:55 pm
nzo wrote: Sat Jan 01, 2022 5:19 pm Hello!
First post, I didn't find a Q/A topic for TF.
I have a rev 1D1 A1200 with KS 3.0. Is it OK for a TF1260 (with full 060)? Chucky told me that under KS3.0 the fast mem doesn't recognized. Will it be solved later? A local Amigo told me that his TF1230 won't work in his rev 1D1. Is this also the situation for the TF 1260 or it will work 100%?
The KS3.0 issue is a design decision. It’s not a “problem”. The choice was either KS3.0 support or mediator support. I chose to support the latter. This should have been disclosed to you before you bought the card. Just upgrade.
You don’t even need to upgrade, just soft kick 3.1.

In fact everyone who buys a 1260 from me will get a disk image that does this automatically.
ccapublic
Posts: 8
Joined: Sun Dec 19, 2021 2:58 pm

Re: TF1260 New Firmware Alpha Released!

Post by ccapublic »

alenppc wrote: Sat Jan 01, 2022 5:59 pm
You don’t even need to upgrade, just soft kick 3.1.

In fact everyone who buys a 1260 from me will get a disk image that does this automatically.
Hi Alenppc. How do you softkick 3 1 ? MuMapROM or another method ? A little off topic but that would be very helpful. Thanks ! (And happy new year BTW !)
alenppc
Moderator Team
Moderator Team
Posts: 906
Joined: Thu Nov 08, 2018 12:59 pm

Re: TF1260 New Firmware Alpha Released!

Post by alenppc »

ccapublic wrote: Sat Jan 01, 2022 10:03 pm
alenppc wrote: Sat Jan 01, 2022 5:59 pm
You don’t even need to upgrade, just soft kick 3.1.

In fact everyone who buys a 1260 from me will get a disk image that does this automatically.
Hi Alenppc. How do you softkick 3 1 ? MuMapROM or another method ? A little off topic but that would be very helpful. Thanks ! (And happy new year BTW !)
Use addmem to add the fast ram, then load 3.1 via Skick (available on aminet), and lastly just use MuFastrom Protect command like you would on a physical rom.

You can create a script in the startup-sequence to check the rom version and do this automatically at every boot. It will make the boot time slightly longer but ultimately it will be indistinguishable from a physical 3.1 system from a software point of view.
ccapublic
Posts: 8
Joined: Sun Dec 19, 2021 2:58 pm

Re: TF1260 New Firmware Alpha Released!

Post by ccapublic »

alenppc wrote: Sun Jan 02, 2022 4:05 am
Use addmem to add the fast ram, then load 3.1 via Skick (available on aminet), and lastly just use MuFastrom Protect command like you would on a physical rom.
Thanks.
gsoravil
Posts: 8
Joined: Fri Nov 19, 2021 3:44 pm

Re: TF1260 New Firmware Alpha Released!

Post by gsoravil »

TerribleFire/Stephen,

I'm hoping you take this serious as I think this has a high probability of being useful to you (not just me). I also very much desire success of this project and hope to make this my go-to card. Please fully read to understand what I'm saying as there is some nuance. (And thank you for all the hard work!)

I have a modded-per-your-instructions 1200 (caps, edge connector, etc.). It is dead stable w/my ACA 68030 card and my Blizzard PPC+060 card. This is despite it being quite expanded. This is my main 1200, used nearly daily.

The original firmware on the TF worked for 80% of what I do, but the other 20% is clearly being addressed in the newest alpha. (THANK YOU!)

To my main point:
In the original firmware (80% useful for me) I HAD to use a custom pre-computed phase value in my heavily expanded 1200. Doing so allowed me to run STABLE at 80MHz (or any other speed below that). The default values ALWAYS caused immediate lockups except for at 50MHz where it "just squeaked by" for the 80% I could do on the TF at that time.

In the alpha (100% useful for me IF I can address this), the phase value for 50MHz is similarly marginal on my machine, however the new tool prevents me from using phase values beyond the first few values. So I cannot set the phase for what I need. Speeds > 50MHz on the alpha cause the same lockups I would receive from the default phase values on the original firmware.

I believe this phase stuff may be a clue to some issues people's experience. I can tell you the #1 core problem for me is the timing and complexity of using an Elbox Fast ATA IV controller -- or at least the way I'm using it -- with a fast modern SATA SSD (via adapter). It is heavily DMA'd and adds quite a lot of tightness to the bus access timings.

In short, are weirdo phase values possible on alpha if the tool is unlocked? This is a serious question and I'm genuinely trying to help both of us.

For reference, here's the phase value calculation I used on original firmware. I derived this from my own analysis:

round(<Desired MHz>/(28.636363/4)) ... for convenience I just use round(<Desired MHz>/7.16)

This yields the following phase values:
50 MHz = 7
63 MHz = 9
75 MHz = 10
80 MHz = 11

All of those were completely stable for me on original firmware and were a requirement for stability for me.

Note my 060 is a Rev6, but the same values applied for my Rev 5 though it could not run at 80.

My reasoning for the weird values was that clocks are continuous and this allowed for effectively "half clocks". As I figured out the values, I found it very telling that they aligned to what is effectively a skew of the base amiga frequencies. I don't know if the TF's clock is routed through a CPLD or not but I suspect a phase alignment issue or possibly inversion (high vs low and/or rising edge vs falling edge) vs the motherboard's clock. Once again, I'm just trying to throw out my thoughts not as any kind of fact but rather as a possibility. My goal is to help the project.

Anyway, I'd like to help and also ask if an unlocked cpuspeed tool for alpha is available.

I am an old heavy overclocker and tinker in building electronics (I have a scope too) so I'm basing a lot of this on a "sixth sense" of what I suspect is happening. I admit that's not scientific though. :-)

Also, feel free to PM if that's preferred for whatever reason.

Appreciated,
Greg
terriblefire
Moderator Team
Moderator Team
Posts: 5368
Joined: Mon Aug 28, 2017 10:56 pm
Location: Glasgow, UK

Re: TF1260 New Firmware Alpha Released!

Post by terriblefire »

Thanks...

With the ~70ns cycle the phases are effectively modulo 7 at 100Mhz. ie. phase 7 = phase 0 @ 100mhz.
———
"It is not necessarily a supply voltage at no load, but the amount of current it can provide when touched that
indicates how much hurting you shall receive."
gsoravil
Posts: 8
Joined: Fri Nov 19, 2021 3:44 pm

Re: TF1260 New Firmware Alpha Released!

Post by gsoravil »

terriblefire wrote: Mon Jan 03, 2022 5:23 pm Thanks...

With the ~70ns cycle the phases are effectively modulo 7 at 100Mhz. ie. phase 7 = phase 0 @ 100mhz.
Well that is helpful info for me.

Please be aware (FYI) on my machine at least, the new cpuspeed on alpha doesn't seem to let me set the desired phase, so my testing is being severely limited (unless I missed a post somewhere). If I do phase=2 or phase=3 it might set it to 4, or 0. It does whatever it feels like. I'm not sure what's up with it. I am one of the people where the original old cpuspeed tool and this new one shows garbage for the MHz. For the old one, Alper Sonmez (I believe) did a fix. I don't know if he's part of this project or has access to the source for the new cpuspeed.

On original firmware I swear I tried all the low phase values first (I didn't know not to) and I didn't get the same stability. Is it possible that tiny bit of skew is making a difference in my config? I agree this is very strange as when I do the math, that's effectively a phase error of only 0.02 at 50MHz which is well under half of a percent (talking NTSC machine here -- error is even less for PAL). That sure sounds like it shouldn't have mattered at all, yet for me with the FastATA on NTSC it did (which seems crazy). I can tell you my crashes were specifically related to timing with the FastATA as my crashes were always disk driver related and utterly predictable. And I'm certain it uses DMA so the clocks are probably critically timed. Freaking bizarre. I'm always a damn edge case but for debugging that's probably a good thing.

(Again, thank you for all the hard work. It's very exciting stuff!)

Greg
terriblefire
Moderator Team
Moderator Team
Posts: 5368
Joined: Mon Aug 28, 2017 10:56 pm
Location: Glasgow, UK

Re: TF1260 New Firmware Alpha Released!

Post by terriblefire »

Phase is no longer a direct phase. You only get 0-3 with this one. Also CPU clock timings are reworked entirely so nothing from the old version is the same.
———
"It is not necessarily a supply voltage at no load, but the amount of current it can provide when touched that
indicates how much hurting you shall receive."
terriblefire
Moderator Team
Moderator Team
Posts: 5368
Joined: Mon Aug 28, 2017 10:56 pm
Location: Glasgow, UK

Re: TF1260 New Firmware Alpha Released!

Post by terriblefire »

Possibly i should explain what the phase does in the new firmware.

Traditionally Amiga accelerators work by using a fast clock domain to sample signals in the slow clock domain. When you do this you need to pass the signal through 2 D-Type flip flops to avoid something called metastability. The downside is this introduces latency each time you cross the clock domains. There is another way...

You can reduce the clock domain crossings by having your 14Mhz clock in the same clock domain as the CPU clock. Some people do this by multipliying the 14Mhz clock but this limits your available speeds.

I do this by sampling the 14Mhz clock in the faster clock domain through a chain of 9 flip flops and allowing the phase parameter to pick up which one of those 9 flip flops will be the output for the synced 14Mhz clock. If the A1200 chipset didnt suck the timing on this wouldnt be so critical.
———
"It is not necessarily a supply voltage at no load, but the amount of current it can provide when touched that
indicates how much hurting you shall receive."
terriblefire
Moderator Team
Moderator Team
Posts: 5368
Joined: Mon Aug 28, 2017 10:56 pm
Location: Glasgow, UK

Re: TF1260 New Firmware Alpha Released!

Post by terriblefire »

There was also a bug in the original CPU speed that would ignore phase=0
———
"It is not necessarily a supply voltage at no load, but the amount of current it can provide when touched that
indicates how much hurting you shall receive."
Post Reply

Return to “TF360”