Best AGA implementation on FPGA?

[JimDrew]

Originally Posted by lurch

Anyway, just need to know when the Reply and the 060 board is out in the wild. Thinking strongly about buying both.

Is what yaqube was saying in another thread on here still the case?

"The Replay board is able to host Minimig AGA core (which can be switched into OCS/ECS compatibility modes) and allows all AGA stuff to work. This core comes with a lot of additional goodies: 8MB of FAST RAM, 1.5MB of SLOW RAM, 48MB of extra CHIP RAM, support for 2MB ROMs, 24-bit AHI sound, Picasso96 compatible RTG board capable of 1920x1080 resolution and full colour display (although not at the same time).

The Replay board can be easily expanded with different sorts of daughterboards. For example the one which I have made has a real 68060 CPU running at 106MHz, 128MB of local SDRAM, High Speed USB host controller, Fast Ethernet controller, RTC, micro SD card socket, TOSLINK optical audio output."

Yes, that is still accurate today. Add to that list support for .scp flux images (and the fact that we could now add a real floppy drive and boot anything an Amiga could through a simple drive port interface). I am currently running my Amiga Workbench in AGA mode with 57MB of fast RAM, 2MB chip RAM. I assembled the driver for the RTG support, and now Mike has to add the video modes in the video layer, and then the RTG should work just fine.

[Sandro]

Originally Posted by JimDrew

That's not the case. First of all the benchmarks shown are slower than a 33MHz 040. Perhaps you are not familiar with FUSION. It is far more stable, more compatible, and hands down faster, than Shapeshifter.

the best way to know how fast is the 060 in mac emulators is disabling superscalar cache and make a sysinfo test

ex:

68060 /50 mhz all caches enabled = 38mips

68060 superscalar cache disabled = ??


to compare:

68040 /40mhz : 30 mips

68040/33 mhz : 25 mips

68040/25 mhz : 18 mips

[Megol]

Originally Posted by JimDrew

If you have a base clock of 50MHz and change that base clock to 250MHz, you would see a 5x speed increase, no? Faster part = faster speed.

Let's assume the shifter in a one stage implementation requires 8ns worst case, then the maximum clock rate is 125MHz.
But one can split the shifter into a two pipeline stage design and then one could perhaps have a 4ns latency per stage enabling 250MHz.
One can't separate the latency of operations and the clock rate - they are expressions of the same thing: frequency=1/latency, latency=1/frequency

[matthey]

Originally Posted by Sandro

the best way to know how fast is the 060 in mac emulators is disabling superscalar cache and make a sysinfo test

ex:

68060 /50 mhz all caches enabled = 38mips

68060 superscalar cache disabled = ??

It's not quite so simple as the 68060 branch cache may need to be disabled also (according to Jim). I would expect the 68060 to be severely handicapped with both superscalar and the branch cache disabled, perhaps even to the point of the 68040 outperforming at the same clock speed. In this configuration,

68060 has better instruction timings
68060 has twice the I and D caches
68040 has better branch performance
68040 has bigger instruction fetch (and in some cases better cache performance)

If we look at the results from the 68060 daughter board video I linked above, the CPU (integer) performance in the MacOS benchmark is close between the 68040 and 68060 on a per clock basis. The Mac 840AV with 68040@40MHz has a CPU rating of 158 which should be about 198 with a 68040@50MHz. A little less than half the 460 CPU score for the 68060 would outperform the 68040 by a relatively small amount. The better CPU score by the 68060 could be because of faster memory and cache performance of the overclocked 68060 and improved memory performance of the 68060 daughterboard. The 68060 does clearly win in performance vs the 68040 by out clocking the 68040. The FPU performance of the 68060 is substantially better than the 68040. It looks to me like it is at least 75% better at the same clock speed. I don't believe turning superscalar off (I believe FPU instructions still execute in parallel as with the 68040) or the branch cache off (I don't believe the FBcc instruction uses the branch cache) affects the FPU performance. Even a 68060@50MHz looks to me like a nice performance gain over a 68040@40MHz for MacOS emulation. Most 68060 processors will allow 60-65MHz for a substantial performance gain over a 68040@40MHz. The few blessed Amiga users with rev 6 68060 will have more than twice the performance of a 68040@40MHz for MacOS emulation.

[JimDrew]

That's not the case though. We had the various 060 boards for the Amiga. My PP040-33 smoked all of them in Speedometer testing. It is also faster than the 840AV's benchmarks, even though it is a slower speed CPU. Norton's test is not very thorough, and certainly does not represent all of the various math tests that typical benchmark programs should perform. Snooper, Speedometer, and several others were commonly used for comparison (never Nortons), but Speedometer 4.0 was the defacto standard for benchmark testing.

The Mac OS does cache clearing everywhere, especially in system traps, and clearing caches on an 060 takes far longer than an 040 so there is a big speed penalty from that. I got tired of typing CacheClearU().

I am still very convinced that a good FPGA 040 core is the way to go for compatibility and speed. I created a complete 68040 emulation (CPU, FPU, MMU) for FUSION-PC, my PC version of FUSION. So, I have all of the simulated microcode necessary for a conversion to FPGA. The FPU would have to be created from scratch as I used the x86's FPU in FUSION-PC, but the MMU was part of the software CPU core. I know microcode well, PALs, GALS, PEELs, etc. but unfortunately, I don't know VHDL or Verilog well enough currently to tackle such a project. I am going to take some time to learn the ins and outs completely and see if I can do a full conversion.

[JimDrew]

Originally Posted by Megol

Let's assume the shifter in a one stage implementation requires 8ns worst case, then the maximum clock rate is 125MHz.
But one can split the shifter into a two pipeline stage design and then one could perhaps have a 4ns latency per stage enabling 250MHz.
One can't separate the latency of operations and the clock rate - they are expressions of the same thing: frequency=1/latency, latency=1/frequency

I am talking about clocking the entire FPGA itself at a higher clock rate, using a newer part that can be clocked at a higher rate.

[Sandro]

Originally Posted by matthey

It's not quite so simple as the 68060 branch cache may need to be disabled also (according to Jim). I would expect the 68060 to be severely handicapped with both superscalar and the branch cache disabled, perhaps even to the point of the 68040 outperforming at the same clock speed. In this configuration,

68060 has better instruction timings
68060 has twice the I and D caches
68040 has better branch performance
68040 has bigger instruction fetch (and in some cases better cache performance)

If we look at the results from the 68060 daughter board video I linked above, the CPU (integer) performance in the MacOS benchmark is close between the 68040 and 68060 on a per clock basis. The Mac 840AV with 68040@40MHz has a CPU rating of 158 which should be about 198 with a 68040@50MHz. A little less than half the 460 CPU score for the 68060 would outperform the 68040 by a relatively small amount. The better CPU score by the 68060 could be because of faster memory and cache performance of the overclocked 68060 and improved memory performance of the 68060 daughterboard. The 68060 does clearly win in performance vs the 68040 by out clocking the 68040. The FPU performance of the 68060 is substantially better than the 68040. It looks to me like it is at least 75% better at the same clock speed. I don't believe turning superscalar off (I believe FPU instructions still execute in parallel as with the 68040) or the branch cache off (I don't believe the FBcc instruction uses the branch cache) affects the FPU performance. Even a 68060@50MHz looks to me like a nice performance gain over a 68040@40MHz for MacOS emulation. Most 68060 processors will allow 60-65MHz for a substantial performance gain over a 68040@40MHz. The few blessed Amiga users with rev 6 68060 will have more than twice the performance of a 68040@40MHz for MacOS emulation.

nope master, the 68060 is not a core i7
the 68040/40mhz is faster than a 68060/50mhz in mac emulation, maybe is faster than a 68060/66mhz (integer performance)

but the FPU of the 68060 is about 2.5x of the one in the 68040, mac fpu programs will run faster in the 608060, ie duken nukem 3d ( requires fpu)

[Sandro]

Originally Posted by JimDrew

I am still very convinced that a good FPGA 040 core is the way to go for compatibility and speed.

the best for compatibility is a FPGA 68020+fpu core

of course your idea of the FPGA 040 is not bad, is better than the 68060 idea, which sucks because the 060 is incompatible with tons of programs

[ransom1122]

I just read about MIST FPGA and would like to know more about AGA compatibility. Ive seen the Lion King aga YouTube clip. Has there been any more progress on the aga core? How many aga games have been tested and what results are they?

At $199 euro I'd like it to work 100% if I was to buy it.

Away from db9 joysticks does any usb joystick work on them? Can we use ps3/360 controllers on it wirelessly?

[britelite]

Originally Posted by ransom1122

I just read about MIST FPGA and would like to know more about AGA compatibility. Ive seen the Lion King aga YouTube clip. Has there been any more progress on the aga core? How many aga games have been tested and what results are they?

The AGA core is still far from perfect, so if that's the only reason to buy the MiST, I'd suggest you wait and see how it matures.

Away from db9 joysticks does any usb joystick work on them?

Haven't tried, but at least USB mice work on the Minimig core.

Can we use ps3/360 controllers on it wirelessly?

Not out of the box, maybe if you have a usb-receiver?

[ransom1122]

It would be the reason to buy it as you can get a minimig for ocs/ecs games already.

I would love if someone would showcase more games. Like top 20 AGA games with it on youtube.

I've seen Breathless running on it. Yes it does slow down a little on fullscreen shrinking the screen size Speeds it up. Also seen body blows galactic aga shown on youtube.

Just need to see more please anyone already with one of these gadgets.

[britelite]

Originally Posted by ransom1122

It would be the reason to buy it as you can get a minimig for ocs/ecs games already.

The MiST is already running the Minimig core (both regular and AGA cores). So, no need to get another device for OCS/ECS stuff.

[ransom1122]

Originally Posted by britelite

The MiST is already running the Minimig core (both regular and AGA cores). So, no need to get another device for OCS/ECS stuff.

Does the speed differ between the mist and fgpa arcade when running either cores?

[britelite]

Originally Posted by ransom1122

Does the speed differ between the mist and fgpa arcade when running either cores?

I haven't tried the FPGA Arcade, so I can't really say. But considering the FPGA Arcade IIRC supports cpu-cards, it can probably achieve higher speeds when emulating AGA machines. When emulating an OCS-machine, I at least prefer it to be represent the A500 speed as close as possible

[ransom1122]

The mist is not upgradable and the fgpa arcade is? Therefore tout can achieve higher cpu settings? Is that correct?

[britelite]

Originally Posted by ransom1122

The mist is not upgradable and the fgpa arcade is?

Yes, the MiST is just an FPGA-board, and as far as I know the FPGA arcade can be expanded.

[Locutus]

Beware though that the 060 card for it is still semi-mythical, it has been demo'd but you cant buy it yet.

And we all know about aftermarket Amiga boards with expansion headers, they all seem to have a expansion planned for them never released :>

So, its the waiting game as usual (-:

[ransom1122]

Would be nice to be seen more AGA games shown on youtube beside the odd 3 or 4 games with the AGA core on either minimig and arcade FPGA

[jezry]

I bought the Mist and its a nice little machine. Still would lile to get the fpga arcade when it becomes available perhaps people will experiment with other cpu cards for it also.

[ransom1122]

Originally Posted by jezry

I bought the Mist and its a nice little machine. Still would lile to get the fpga arcade when it becomes available perhaps people will experiment with other cpu cards for it also.

You can buy the arcade by emailing the guyswho created it. But first I need more results shown before I purchase.

Email here
laurent@amedia-computer.com

They will send information on it first


North American members email here

sales@cbmstuff.com.