Performance vs real A1200 + 030

[NovaCoder]

Hi,

Just got my virtual Amiga A1200 up and running

First observations compared to my old A1200 running various accelerators.

Running SysInfo gives 13.69 MIPS which is faster than my old ACA 1230 @ 56 Mhz (12.5 MIPS) and my old Blizzard 1230 @ 50 Mhz (9.9 MIPS)

The thing is that SysInfo also reports the CPU speed at 99 Mhz which is obviously not accurate and the MIPS reading should really be around 6. This is based on my experience with a real A1200, the performance is actually a bit slower than my old Blizzard 1230 @ 50 Mhz and certainly much slower than my old ACA 1230 @ 56 Mhz. Not sure why SysInfo is getting the MIPS reading wrong?

More performance is always nice of course, hopefully this core will get improved to be at least as fast as a real Blizzard 1230

I personally don't think there is much point adding FPU support to the current core unless it's integer performance is increased first.

I did read about some plans to use the ARM CPU to the calculations and then use the FPGA as a 'bridge', this would certainly make it a performance beast

Also I've noticed that it is pretty slow loading large amounts of data, has anyone got PFS3 running and if so did that help with disk performance? I know on a real Amiga that PFS3 can make a big difference over FFS.

[NightShadowPT]

Well, the speed in the MiSTer core is achieved by brute force, not engineering.

What you are witnessing is effectively a 68020 overclocked, and to to the best of my knowledge it's doesn't even have all the new instructions fully implemented (I'll let others correct me on this).

The ACA1230 and Blizzard 1230 both use a Motorola 68030, the following generation of processors, and MHZ to MHZ they are not comparable.

Maybe one day we'll have a 68030 processor in FPGA to be integrated into the Minimig MiSTer core, but for now, this is all you'll get.

[NovaCoder]

I think an 030 has the same instructions as an 020 from memory, I think the only difference is that it has different levels of Cache (and higher possible clock speeds).

[NightShadowPT]

According to the Wikipedia, the 68030 had over the 68020:

"additional 256 byte data cache and a process shrink and an added burst mode for the caches, where four longwords can be placed in the cache without further CPU intervention." (this) "increases performance by ≈5% while reducing power draw by ≈25% compared to the 68020."

I stand corrected, they are comparable from a MHZ perspective.

[kolla]

I could be wrong, but where I notice most of the "sluggishness" on both MiST and MiSTer, compared to "real" A1200/030 systems, is in the "graphics", that is, in the AGA-chipset, and not so much in the CPU.

[akeley]

Personally, I'd rather see this core being first made as close to vanilla A1200 as possible, enabling cycle-accuracy for example.

[kolla]

Personally, I'd rather see this core being first made as close to vanilla A1200 as possible, enabling cycle-accuracy for example.

That's not something to just "enable", that requires total rewrites of both the chipset and the CPU cores.

And for what? The number of software titles that require cycle exact 020+AGA is... zero

[akeley]

And for what? The number of software titles that require cycle exact 020+AGA is... zero

For what? The fact we're aiming to replicate the original machines. Isn't that the main idea behing the whole FPGA angle?

If you're 100% sure that there really is no software which is affected by that then okay, it's a good point - though I know that on A500 it's different. I've heard that there are lots of demos which have problems on AGA. I don't know if this is related to cycle accuracy but it's one of possible reasons. In any case my point is that I'd rather see A1200 brought as close to 1:1 as possible, since extra speed does not really interest me that much (most games which require it are available on other platforms).

[chaos]

I could be wrong, but where I notice most of the "sluggishness" on both MiST and MiSTer, compared to "real" A1200/030 systems, is in the "graphics", that is, in the AGA-chipset, and not so much in the CPU.

Can you describe your feeling of 'sluggishness'? Where / when do you notice it?

[bazza_12]

Can you describe your feeling of 'sluggishness'? Where / when do you notice it?

usually first thing in a morning.. before breakfast.. do you think i'll live, doc?

[grizzly]

Can you describe your feeling of 'sluggishness'? Where / when do you notice it?

From the time i wake up until i get to sleep.
Yeah i´m a goner i know

[kolla]

Can you describe your feeling of 'sluggishness'? Where / when do you notice it?

usually first thing in a morning.. before breakfast.. do you think i'll live, doc?

Funny thing is, I am reading this while eating breakfast, in front of my MiST and MiSTer, hehe

The most obvious "sluggishness" (I keep using quotes here, this is very unscientific) is when dragging screens up and down. I just compared with real ECS (A600) and AGA (CD32), and MiST compares better with ECS than with the AGA. MiSTer is faster, but still feels like it is "dragging behind" the mouse a bit, as if the screen has actual mass, heh

[kolla]

And for what? The number of software titles that require cycle exact 020+AGA is... zero

For what? The fact we're aiming to replicate the original machines. Isn't that the main idea behing the whole FPGA angle?

Not really, which should be obvious - the original machines and these FPGA systems don't have so much in common. The idea behind the whole "FPGA angle" is to have new hardware to replace old hardware, and to avoid certain issues with software emulation (latency and more).

If you're 100% sure that there really is no software which is affected by that then okay, it's a good point - though I know that on A500 it's different.

Correct, on the A500 is _is_ different, lots of games need cycle exact replication to play correctly - that is also why MiSTer uses the cycle exact Fx68k core in 68000 mode, and the original Minimig core is OCS/ECS. With the AGA systems, the situation was very different than it was with A500 - people had learnt a thing or two, expansions were commonplace, software and games were not written with "cycle exact" bare A1200 in mind.

I've heard that there are lots of demos which have problems on AGA. I don't know if this is related to cycle accuracy but it's one of possible reasons.

Perhaps of the way Minimig-AGA is implemented (Chaos can elaborate on this, if he feels like it).

In any case my point is that I'd rather see A1200 brought as close to 1:1 as possible, since extra speed does not really interest me that much (most games which require it are available on other platforms).

And I'd rather see A4000 brought as close to 1:1 as possible - after all, that is more like what the MiST and MiSTer resemble, lacking PCMCIA, clock ports, and with tons of zorro ram stuck in there...

[bazza_12]

A4000 would be amazing.. is it possible?

[Chris23235]

A4000 would be amazing.. is it possible?

Unlikely, right now there is now 68030 implementation and I think it would bring the FPGA of the MiSTer to its limits speedwise.

[kolla]

Well, what is really the big difference between a 68020 and a 68EC030 soldered on the motherboard of the A4000D/CR in terms of speed and software compatibility, and how does the TG68 in the MiSTer compare? My guess is that the TG68 in the MiSTer easily outperforms the 25MHz EC030 on the A4000D/CR motherboard.

[akeley]

Not really, which should be obvious - the original machines and these FPGA systems don't have so much in common. The idea behind the whole "FPGA angle" is to have new hardware to replace old hardware, and to avoid certain issues with software emulation (latency and more).

One of the big hype factors behind FPGA promotion is "hardware simulation", the "feel", and how we are sort of replicating the old machines' logic in Verilog. Perceived like that they have much more in common with OG HW than emulation. That's what I've heard countless times out there, before I got MiSTer, and even if I was quite skeptical about it - it's a very subjective approach - I must now admit that it kinda works, even if "it's all in the mind". Yes, Minimig "feels" like Amiga, much more than WinUAE.

It's a quite big angle, because if you take away the latency (another very big selling point), there's not much left in favour of FPGA way - emulation is more versatile and in many cases faster and more accurate (eg Amiga).

This is perhaps a meta-subject for another thread, but I like to know for sure where we are in regard to core development. It also helps with conversations with skeptics/people on the fence about MiSTer. If cycle accuracy is really not important than indeed I can mentally tick it off of the to-do list.

[chaos]

Can you describe your feeling of 'sluggishness'? Where / when do you notice it?

usually first thing in a morning.. before breakfast.. do you think i'll live, doc?

I LOLed now that I've re-read it

[chaos]

The most obvious "sluggishness" (I keep using quotes here, this is very unscientific) is when dragging screens up and down. I just compared with real ECS (A600) and AGA (CD32), and MiST compares better with ECS than with the AGA. MiSTer is faster, but still feels like it is "dragging behind" the mouse a bit, as if the screen has actual mass, heh

This can only be the effect of the CPU slowness / latency, as the chipset, both OCS/ECS and AGA, is hard real-time - that is, there is no backpressure control, so either it works as fast as it should (and as it does in real Amigas), or there would be visible glitches in the display.

BTW, do you see the same issues on analog video? And do you have turbo enabled?

[NovaCoder]

Mike has made some performance improvements to the TG68 core (about 10%), be great if these updates could find their way into the MiSTer one day....


https://www.mike-stirling.com/2020/02/f ... amiga-500/

[kolla]

This can only be the effect of the CPU slowness / latency, as the chipset, both OCS/ECS and AGA, is hard real-time - that is, there is no backpressure control, so either it works as fast as it should (and as it does in real Amigas), or there would be visible glitches in the display.

I see, and yes that does make sense, all my AGA systems have faster CPUs, while the A600 I just tried with is plan 7MHz 68000.

BTW, do you see the same issues on analog video? And do you have turbo enabled?

The MiST has only analogue, and I see less of this with MiSTer's HDMI output, perhaps because its CPU is a bit faster. Turbo always enabled.

Anyways, not a big issue