Minimig (Amiga) core discussion

https://github.com/MiSTer-devel/Main_MiSTer/wiki

Moderators: Mug UK, Zorro 2, Greenious, spiny, Sorgelig, Moderator Team

kolla
Captain Atari
Captain Atari
Posts: 232
Joined: Thu Sep 17, 2015 11:39 pm
Contact:

Re: Minimig (Amiga) core discussion

Postby kolla » Sun Feb 17, 2019 5:49 am

fpgaarcade wrote:
JimDrew wrote:
kolla wrote:So, is this "new core" Jim mentioned made by MikeJ or not??


No, it is not. It is being created by a group of guys who can't stand Gunnar's attitude.


This core is my attempt at cleaning up the TG68K and fixing the last few bugs. I called it M68K to differentiate.


When I see M68K I think real Motorola 68k chip.
And I am just as wise regarding the new core, is it you, Mike, or some other guys? I get mixed signals when Jim says "no" and you comment as if it is you...
-- kolla

JimDrew
Atari Super Hero
Atari Super Hero
Posts: 815
Joined: Mon Nov 04, 2013 5:23 pm

Re: Minimig (Amiga) core discussion

Postby JimDrew » Sun Feb 17, 2019 8:30 am

The core I am referring to has absolutely nothing to do with Mikej or Replay 1 (or 2). It is a core that has been worked on for the last year. It is a 68040 w/MMU and FPU. It has a 8-way SA cache, TLB, etc. just like a real 68040. The 040 is the simplest of the full (MMU/FPU) Motorola CPUs to emulate. The 68020 w/68851 and 68881/2 and 68030 are rather involved with the number of page sizes that the MMU can use, elaborate FPU instructions, etc. The 040 was a popular CPU (Macs used them as well as the A4000).

This core is being developed by a group of guys that were once involved with helping debug the original Vampire board, but they had a some severe falling out with Gunnar. They are creating a 68040 core that will out perform the Apollo core, it will 100% compatible, and it will be free.
I am the flux ninja

Sorgelig
Ultimate Atarian
Ultimate Atarian
Posts: 5070
Joined: Mon Dec 14, 2015 10:51 am
Location: Russia/Taiwan

Re: Minimig (Amiga) core discussion

Postby Sorgelig » Sun Feb 17, 2019 8:44 am

Revenge is not a good (and not long) motivation. Hope they will release it before they will lose interest.

I remember i've read somewhere that 68060 was rather simplification of 68040's FPU/MMU.

kolla
Captain Atari
Captain Atari
Posts: 232
Joined: Thu Sep 17, 2015 11:39 pm
Contact:

Re: Minimig (Amiga) core discussion

Postby kolla » Sun Feb 17, 2019 9:19 am

I see. Well, without details about who, what plans and status are... that's vaporware as good as any.
-- kolla

witchmaster
Atari maniac
Atari maniac
Posts: 94
Joined: Wed Jun 15, 2011 10:50 am

Re: Minimig (Amiga) core discussion

Postby witchmaster » Sun Feb 17, 2019 9:33 am

kolla wrote:I see. Well, without details about who, what plans and status are... that's vaporware as good as any.

Hopefully it will see the light of day! :)

kolla
Captain Atari
Captain Atari
Posts: 232
Joined: Thu Sep 17, 2015 11:39 pm
Contact:

Re: Minimig (Amiga) core discussion

Postby kolla » Sun Feb 17, 2019 10:02 am

Yes, indeed.

I know that the 68k Linux and *BSD communities are interested in such a CPU core, and other 68k retro platforms have shown interest in FPGA based drop-in replacements for old CPUs. An FPGA board that could be dropped into an existing 040 socket, and offer a decent boost to any 040 system... that could be popular far beyond Amiga sphere.
-- kolla

Sorgelig
Ultimate Atarian
Ultimate Atarian
Posts: 5070
Joined: Mon Dec 14, 2015 10:51 am
Location: Russia/Taiwan

Re: Minimig (Amiga) core discussion

Postby Sorgelig » Sun Feb 17, 2019 11:12 am

It won't be easy. 68040 AFAIK uses 5V while all modern FPGAs are 3.3V. Probably it can be used as a drop-in replacement for 68060 as a downgrade :) At least they share the same 3.3V.
I think making physical 68060/040 replacement is also not easy as it has a lot of pins. These pins are round and thin. It's not like soldering a simple 1-row header for 68000 or SID drop-in board. Not sure after all it will worth the money as 68040 is not expensive and most likely much cheaper than such fpga board.

kolla
Captain Atari
Captain Atari
Posts: 232
Joined: Thu Sep 17, 2015 11:39 pm
Contact:

Re: Minimig (Amiga) core discussion

Postby kolla » Sun Feb 17, 2019 11:37 am

The voltage problem is AFAIK already taken care of in various 060-adapters that can be dropped into 040 sockets. Ditto for the pin issue, it's a solved problem.

The main problem for any 040 system is heat - the 040 can fry eggs even at 40MHz, replacing the CPU with an FPGA would also solve this.
-- kolla

ijor
Hardware Guru
Hardware Guru
Posts: 3796
Joined: Sat May 29, 2004 7:52 pm
Contact:

Re: Minimig (Amiga) core discussion

Postby ijor » Sun Feb 17, 2019 2:02 pm

JimDrew wrote:This core is being developed by a group of guys that were once involved with helping debug the original Vampire board, but they had a some severe falling out with Gunnar. They are creating a 68040 core that will out perform the Apollo core, it will 100% compatible, and it will be free.


Can't believe the animosity. So they are making it public domain (and not just open source), only to bother the Vampire guy!? LOL
Fx Cast: Atari St cycle accurate fpga core

Sorgelig
Ultimate Atarian
Ultimate Atarian
Posts: 5070
Joined: Mon Dec 14, 2015 10:51 am
Location: Russia/Taiwan

Re: Minimig (Amiga) core discussion

Postby Sorgelig » Sun Feb 17, 2019 2:21 pm

kolla wrote:The voltage problem is AFAIK already taken care of in various 060-adapters that can be dropped into 040 sockets. Ditto for the pin issue, it's a solved problem.

i know about power adapters for 060, but never checked the circuit in details. It looks like these adaptors just convert the power supply while 060 is 5V tolerant on bus. FPGA is not 5V tolerant. If the rest board is traditional TTL where logic 1 is bellow 3.3V then it will be fine. If it's CMOS TTL then logic 1 is almost 5V and may damage the FPGA.

kolla
Captain Atari
Captain Atari
Posts: 232
Joined: Thu Sep 17, 2015 11:39 pm
Contact:

Re: Minimig (Amiga) core discussion

Postby kolla » Sun Feb 17, 2019 2:43 pm

ijor wrote:Can't believe the animosity. So they are making it public domain (and not just open source), only to bother the Vampire guy!? LOL

He's not the Vampire guy, that would Majsta. Gunnar is the Apollo Core guy.
-- kolla

kolla
Captain Atari
Captain Atari
Posts: 232
Joined: Thu Sep 17, 2015 11:39 pm
Contact:

Re: Minimig (Amiga) core discussion

Postby kolla » Sun Feb 17, 2019 2:47 pm

Sorgelig wrote:
kolla wrote:The voltage problem is AFAIK already taken care of in various 060-adapters that can be dropped into 040 sockets. Ditto for the pin issue, it's a solved problem.

i know about power adapters for 060, but never checked the circuit in details. It looks like these adaptors just convert the power supply while 060 is 5V tolerant on bus. FPGA is not 5V tolerant. If the rest board is traditional TTL where logic 1 is bellow 3.3V then it will be fine. If it's CMOS TTL then logic 1 is almost 5V and may damage the FPGA.


Right. In case of Amiga, my understanding is that due to pretty much pure luck, CBM chose logic 1 to be below 3.3V.
-- kolla

Sorgelig
Ultimate Atarian
Ultimate Atarian
Posts: 5070
Joined: Mon Dec 14, 2015 10:51 am
Location: Russia/Taiwan

Re: Minimig (Amiga) core discussion

Postby Sorgelig » Sun Feb 17, 2019 3:00 pm

kolla wrote:In case of Amiga, my understanding is that due to pretty much pure luck, CBM chose logic 1 to be below 3.3V.

some time ago i was planing (and still planing) the scaler board for Amiga and had the same question about logic levels in Amiga. If i remember correct, Amiga uses exactly CMOS levels, so FPGA cannot be used directly.
Quick googling revealed discussions of different forums where people say that 68060 is simply 5V tolerant. That's why simple power supply conversion works on adapters.
And i believe 68040 core itself should be very big, so only expensive FPGA will fit it. I don't think it will be a good alternative for drop-in replacement. At least while real 68040 are available.

ijor
Hardware Guru
Hardware Guru
Posts: 3796
Joined: Sat May 29, 2004 7:52 pm
Contact:

Re: Minimig (Amiga) core discussion

Postby ijor » Sun Feb 17, 2019 3:11 pm

kolla wrote:
ijor wrote:Can't believe the animosity. So they are making it public domain (and not just open source), only to bother the Vampire guy!? LOL

He's not the Vampire guy, that would Majsta. Gunnar is the Apollo Core guy.


Ok, sorry, I stand corrected. But still doesn't change the main point :)
Fx Cast: Atari St cycle accurate fpga core

JimDrew
Atari Super Hero
Atari Super Hero
Posts: 815
Joined: Mon Nov 04, 2013 5:23 pm

Re: Minimig (Amiga) core discussion

Postby JimDrew » Sun Feb 17, 2019 8:52 pm

Phase 5's 060 board for the Amiga could use an 040 by just changing the main crystal and moving a jumper (3.3v->5v). They are pin compatible, just not voltage compatible.

As someone who worked on Motorola's LSPLD project (doing the microcode for the 040 and 060 cores) as well as having written a software 68040 w/FPU & MMU emulation (in assembly), I can tell you that the 060 is definitely not a simplified version of the 040. Having dual pipelines really changes how everything works.
I am the flux ninja

ericgus
Captain Atari
Captain Atari
Posts: 455
Joined: Fri Apr 08, 2016 4:53 am
Location: Boston MA - USA

Re: Minimig (Amiga) core discussion

Postby ericgus » Sun Feb 17, 2019 9:36 pm

JimDrew wrote: They are creating a 68040 core that will out perform the Apollo core,.


HA..

I wish them the best of luck (we can only all benefit from a free 68060 class core, but I highly doubt it will outperform the Vampire/Apollo core.)

Gunnar and the apollo core team have put a lot of work into optimizing, tweaking and incorporating modern techniques and methods and tricks to try to get the most performance out of a 68K class processor.. they will have a number of years of catch up .. and even then, if they eschew incorporating modern microprocessor advancements in their 060 core (to keep with a pure core) its doubtful it will perform at the same or comparable levels of the vampire .. but as I said I am sure we will all benefit from their work in trying to catch up to what the Apollo team has already proven they can do and brought to the market.

ericgus
Captain Atari
Captain Atari
Posts: 455
Joined: Fri Apr 08, 2016 4:53 am
Location: Boston MA - USA

Re: Minimig (Amiga) core discussion

Postby ericgus » Sun Feb 17, 2019 9:42 pm

ijor wrote:Can't believe the animosity. So they are making it public domain (and not just open source), only to bother the Vampire guy!? LOL


Yep believe it or not. Some would say the Vampire/Apollo core stand to crash the insanely overpriced 060 accelerator board market, I know personally I have zero desire to spend thousands of US dollars on a 25+ year old board, when I can spend a few hundred on a 100% brand new board made in 2019 and outperform those antique boards.. while adding other features (rtg etc) literally saving hundreds of dollars more on addon parts, RTG cards, network cards, ram board, etc.. So it presents a serious financial threat to some, hence it undercuts their finances.. so you can see where some of that annomoisity comes from and why those people would want to trash talk/bad mouth the Vampire its represents a direct financial threat.. for others its simply seen as a "perversion" of the old defined hardware/standards and thats offensive to others.. so you have a number of groups of people who are getting offended .. thats where a lot of this originates from.. And in a few rare cases there has been direct confrontation between the Vampire/Apollo team and individuals (so thats another source).

kolla
Captain Atari
Captain Atari
Posts: 232
Joined: Thu Sep 17, 2015 11:39 pm
Contact:

Re: Minimig (Amiga) core discussion

Postby kolla » Mon Feb 18, 2019 1:45 am

Nobody ever had anything against the Vampire cards, rather the contrary - absolutely _ALL_ critics have been towards the CPU core and Gunnar himself.
-- kolla

Sorgelig
Ultimate Atarian
Ultimate Atarian
Posts: 5070
Joined: Mon Dec 14, 2015 10:51 am
Location: Russia/Taiwan

Re: Minimig (Amiga) core discussion

Postby Sorgelig » Mon Feb 18, 2019 4:57 am

I think faster means original 68K instructions not including 68080's MMX-like additions.
Basically no one need those MMX additions besides the creator. For preservation purpose a close replica to 68K is more valuable than any fictional additions.
Who need an insanely fast Amiga - those can run WinUAE on modern Core i7/i9 PC. Amiga has no tasks for such speed. So i vote for proper 68040/060 replica.

Hope it's not vapour project. Sometimes between "release soon" and "released" years may pass (if it will ever happen).

JimDrew
Atari Super Hero
Atari Super Hero
Posts: 815
Joined: Mon Nov 04, 2013 5:23 pm

Re: Minimig (Amiga) core discussion

Postby JimDrew » Mon Feb 18, 2019 6:04 am

kolla wrote:Nobody ever had anything against the Vampire cards, rather the contrary - absolutely _ALL_ critics have been towards the CPU core and Gunnar himself.


That's not entirely true. The head of intellectual security for the HDMI group contacted me when they learned (from a YouTube video) that I was going to import the Vampire boards for the U.S. market. The Vampire boards have a HDMI connector, and their spec sheet (since changed, but archived) states the video output is HDMI. Why does this matter? Because any product being sold as HDMI compatible requires licensing the use of the HDMI technology. The Vampire boards are not licensed products so the HDMI group was coming after me to collect the royalties. At that point I had not gotten any boards so I was in the clear, but they are still researching who to go after for the license/patent infringement and periodically contact me about this matter.

I agree with you, Sorgelig. It makes no sense to have the fastest core if it is not 100% instruction compatible. You can have both. When I was helping Gunnar debug the Apollo core I told him this frequently, but he was more interested in being the fastest and any instructions not supported were not really needed. :\
I am the flux ninja

ericgus
Captain Atari
Captain Atari
Posts: 455
Joined: Fri Apr 08, 2016 4:53 am
Location: Boston MA - USA

Re: Minimig (Amiga) core discussion

Postby ericgus » Mon Feb 18, 2019 7:15 am

Sorgelig wrote:I think faster means original 68K instructions not including 68080's MMX-like additions.
Basically no one need those MMX additions besides the creator. For preservation purpose a close replica to 68K is more valuable than any fictional additions.
Who need an insanely fast Amiga - those can run WinUAE on modern Core i7/i9 PC. Amiga has no tasks for such speed. So i vote for proper 68040/060 replica.

Hope it's not vapour project. Sometimes between "release soon" and "released" years may pass (if it will ever happen).


My comments were about not about the non MMX etc extensions that were added .. im talking the base set of 68K "standard" opcodes that the apollo core does support run significantly faster .. because in part of the behind the scenes under the cover optimizations he's done.

I have a vampire V2 in my A500 and its not "vapor" I can assure you. its a real thing that works as described.

I sorta get a bit what he's doing, he's trying to make a "what if" fictional processor that never existed since hes in there anyway .. eg what if they had continued the 68K series and what sort of features it would have.. that said, yes no existing software will take advantage of some features like MMX unless coded for it, but other enhancements impact the performance across all instructions including the original defacto standard .. One other curious thing is, the 060 lacks some instructions the previous 040 had (hence the need to load special libraries under Amiga OS, curiously with Apollo core this isn't required since the core supports the 040 and 060 instruction sets.. so in some ways, depending on how you look at it, its actually more "compatible" than the 060 actually is.

Sorgelig
Ultimate Atarian
Ultimate Atarian
Posts: 5070
Joined: Mon Dec 14, 2015 10:51 am
Location: Russia/Taiwan

Re: Minimig (Amiga) core discussion

Postby Sorgelig » Mon Feb 18, 2019 7:48 am

ericgus wrote:I have a vampire V2 in my A500 and its not "vapor" I can assure you. its a real thing that works as described.

I'm talking about 68040 core. I'm not talking about Vampire which is selling several years already.
No need to put some "magic" into Apollo CPU. The same magic applies to current TG68K core. It's just instructions are executed in less clock cycles than real CPU. Pretty much standard thing for FPGA re-implementations.
Less clock cycles per instruction with well written HDL code allowing to run the CPU on higher clock gives high speed. It's like NextZ80 core where instruction is executed in one clock cycle.

ericgus
Captain Atari
Captain Atari
Posts: 455
Joined: Fri Apr 08, 2016 4:53 am
Location: Boston MA - USA

Re: Minimig (Amiga) core discussion

Postby ericgus » Mon Feb 18, 2019 6:07 pm

Sorgelig wrote:
ericgus wrote:I have a vampire V2 in my A500 and its not "vapor" I can assure you. its a real thing that works as described.

I'm talking about 68040 core. I'm not talking about Vampire which is selling several years already.
No need to put some "magic" into Apollo CPU. The same magic applies to current TG68K core. It's just instructions are executed in less clock cycles than real CPU. Pretty much standard thing for FPGA re-implementations.
Less clock cycles per instruction with well written HDL code allowing to run the CPU on higher clock gives high speed. It's like NextZ80 core where instruction is executed in one clock cycle.


Ah..


Well Gunnar is also doing some multi threaded stuff in the background to enhance performance with the entire core.. as I said hes doing a lot of interesting things with getting the most performance out of the fpga as he can.. anyway its an interesting read (across several posts on the apollo team forums).

JimDrew
Atari Super Hero
Atari Super Hero
Posts: 815
Joined: Mon Nov 04, 2013 5:23 pm

Re: Minimig (Amiga) core discussion

Postby JimDrew » Mon Feb 18, 2019 11:49 pm

The Apollo core is much like the 060 in that it has dual pipelines, so instructions using different registers can occur within the same cycle. If instructions are ordered correctly you can have a massive speed improvement. Gunnar added some special instructions for bit manipulation (reordering of bits) for my PCx program. It turns out that the cache and pipeline mechanism works so well that the special instructions made no difference in the speed of PCx. I have always tried to do proper instruction order in all of my assembly code so that it runs as fast as possible on an 060.
I am the flux ninja

kolla
Captain Atari
Captain Atari
Posts: 232
Joined: Thu Sep 17, 2015 11:39 pm
Contact:

Re: Minimig (Amiga) core discussion

Postby kolla » Tue Feb 19, 2019 5:19 am

@ericgus
And I have two V2 cards. They always offer what is described...ish. But what's described may of course change at Gunnar's whim.
-- kolla


Return to “MiSTer”

Who is online

Users browsing this forum: No registered users and 6 guests