Is 030 overated, and 040 underrated?

[grond]

Perhaps the 030 is held in such high esteem because it was the last 68k generation that was easily up to par with the x86 alternative. The 040 was a good processor on the outside but heat problems and the comparatively slow progress on the clock rates wasted most of its qualities.

[Old_Bob]

Originally Posted by d4rk3lf

Personally, for 3D scenes, I think it's more of a RAM issue, then processor speed. I was very impressed with result my Aca 500 plus (68000 on 14 Mhz) processor did in only 7 minutes, in Real 3D, with default scene (table with teapot), where I just removed bump from teapot (it was ugly, anyway, and only slowing the rendering), and it calculated raytrace reflections and all shadows, so I got very nice 3D render in only 7 minutes.

I did the very same test with the candle scene in Real3D on my machines when I finally got my 1230MKIV. I may even still have the sheet of A4 somewhere with the numbers scribbled on it...

B

[eXeler0]

Originally Posted by d4rk3lf

It would be very interesting if someone could make a Lightwave benchmark, and compare the speed of rendering (the same scene) on 030 vs 040 vs 060.
Personally, for 3D scenes, I think it's more of a RAM issue, then processor speed. I was very impressed with result my Aca 500 plus (68000 on 14 Mhz) processor did in only 7 minutes, in Real 3D, with default scene (table with teapot), where I just removed bump from teapot (it was ugly, anyway, and only slowing the rendering), and it calculated raytrace reflections and all shadows, so I got very nice 3D render in only 7 minutes.

Even now, and especially back in the day, people were not really capable of optimizing 3D scenes, and they were placing raytrace reflections pretty much anywhere (even in non reflective materials like ground and dirt), and they had enormous render times, for no real reason.

People do that even today.
Because, how can you explain, enormous crying of people that they don't have enough processing power to render something (i see comments on youtube, all the time), and Jurassic Park (1994) was rendered on machines slower then processor on low end mobile phone today, and these renders looks absolutely amazing, even today.

I was a Imagine3d user, (Lightwave not so much). Off the top of my head, 50MHz 060 render times were 5x faster than 50Mhz 030 (Both Blizzards).
Here's a fun little story for ya.
My last big project I ever did on Amiga was a scene with over 330,000 polygons.
You can see screenshots here, including wireframe view from Imagine:
https://www.exretro.com/galleries/co...ers/index.html
I had 80MB RAM on my Blizzard 1260. I bought the SCSI kit just so that I could get another RAM slot. So I had 64MB on the accelerator and 16MB on the SCSI kit. The scene would run out of RAM if I tried to render it with scanline renderer, but for some reason the raytrace mode worked. Render time was almost 11 days! Yes, thats's on a 060. This was a rev 1 which was quite hot but it didn't overheat. Funny thing was that on the night between day 9 and 10 there was a thunderstorm and the electricity vanished for a fraction of a second and my PC sitting next to the Amiga, rebooted, but the Amiga didn't, it rendered on like a trooper.
Anyway, If I had tried to render this on a 030 it would take roughly 2 months, give or take a few days.
After that I switched to 3dsmax on PC and never did any ambitious 3d work on the Amiga ever again.

[Overmann]

I recently acquired an A4000 and later upgraded with an 040@40mhz and I'm hugely impressed with the performance of the CPU. I was originally planning to upgrade the accelerator to an 060 but I see no reason to, even if prices hadn't been insane. I've thrown a lot at this machine now and it handles everything with ease. The Zorro III RTG that I have installed helps of course, but I'm hugely impressed with the CPU.

[Old_Bob]

Originally Posted by Overmann

I recently acquired an A4000 and later upgraded with an 040@40mhz and I'm hugely impressed with the performance of the CPU. I was originally planning to upgrade the accelerator to an 060 but I see no reason to, even if prices hadn't been insane. I've thrown a lot at this machine now and it handles everything with ease. The Zorro III RTG that I have installed helps of course, but I'm hugely impressed with the CPU.

I don't suppose anybody has ever tried it but, I wonder how 040 and 060 would compare if both were clocked at 40mhz?

B

[Overmann]

Originally Posted by Old_Bob

I don't suppose anybody has ever tried it but, I wonder how 040 and 060 would compare if both were clocked at 40mhz?

B

I was looking for that when I was freshly blown away by the 040 but found nothing. It really seems to perform extremely close to the 060 to me. I have an 060 clocked at 66mhz and even then the difference is very small.
Not only does Doom play well but Quake is playable as well, and Shapeshifter absolutely flies with an RTG-card. Marathon, Warcraft II, Links LS and SimCity2000 all play really nicely.

[eXeler0]

Originally Posted by Overmann

I was looking for that when I was freshly blown away by the 040 but found nothing. It really seems to perform extremely close to the 060 to me. I have an 060 clocked at 66mhz and even then the difference is very small.
Not only does Doom play well but Quake is playable as well, and Shapeshifter absolutely flies with an RTG-card. Marathon, Warcraft II, Links LS and SimCity2000 all play really nicely.

The one metric 040 wins clearly these days is performance / dollar ;-)
But one important thing to understand is that the 060 is a dual instruction pipeline design, so in theory, running perfectly optimized code, it could retire double the amount of instructions/clock cycle compared to the 040. But that's in theory..
Software that doesn't do anything to utilize a dual instruction pipeline will run similarly at same clock on 040 and 060. This is still noticable today.. I have a 12 core, 24-core thread Ryzen in my PC running at ~4GHz. But if you run a single threadded benchmark it will lose to a quadcore Ryzen (of the same generation) that runs at a higher clock.

The other thing to remember is that getting the CPU to run at a higher clock is also an option when chasing more performance. 486 was technically worse / clock than 040 but would clock much higher. There's a reason there was never any 040 faster than 40MHz. In comparison, even the rev 1 060 could run at 50MHz without cooler. And as we know, last revision clocks to 90-100MHz.

So these are things to consider when you compare 040 to 060.
That being said, I'm pretty certain that a 40MHz 040 is what would give you the best bang for the buck in terms of overall processing power (CPU+FPU) on your Amiga.

[Overmann]

Originally Posted by eXeler0

The one metric 040 wins clearly these days is performance / dollar ;-)
But one important thing to understand is that the 060 is a dual instruction pipeline design, so in theory, running perfectly optimized code, it could retire double the amount of instructions/clock cycle compared to the 040. But that's in theory..

I've read this before but I'm yet to come across a program or game that runs noticeably faster on my 060 then my 040. Are there any? Because if the 060 could really be twice as fast when running optimized code that is pretty damn impressive!

[eXeler0]

Originally Posted by Overmann

I've read this before but I'm yet to come across a program or game that runs noticeably faster on my 060 then my 040. Are there any? Because if the 060 could really be twice as fast when running optimized code that is pretty damn impressive!

I would imagine a lot of demos run heavily 060 optimized code. 060 arrived after Commodore went bankrupt so probably not a whole lot of commercial software that was carefully optimized for 060 unfortunately.

[grond]

The 060 can also take heavy hits in comparison to the 040 if long MUL and DIV instructions are used. It's a pity there never was an 070 with the missing instructions.

[meynaf]

Originally Posted by grond

The 060 can also take heavy hits in comparison to the 040 if long MUL and DIV instructions are used. It's a pity there never was an 070 with the missing instructions.

I have a program that is 2.5 times faster on 030 than on similarly clocked 060 because of this.

[Thomas Richter]

In such cases, MuRedox will definitely help.

[Thomas Richter]

Originally Posted by Overmann

I've read this before but I'm yet to come across a program or game that runs noticeably faster on my 060 then my 040.

This is pretty much my experience as well. The 040 was about twice as fast as a 030, and the 060 was again about twice as fast as the 040. I had a 40Mhz GVP 030 first, then a 33Mhz GVP 040 later, and then (due to income from Os 3.9) a B2060 with a 50Mhz 060.

[Torti-the-Smurf]

The 1230 Blizzard was the finest card i ever used.
I still can remember that no card impressed me like the 030 Card from Phase5.
When i got later an 040 i was kind a dissapointed, even the 060 did not impressed me that much.
Back then, for the money and the "bang you get for youre buck" the 030 was the perfect sweetspot.

Well, today i dont care for AGA no more and for 030 and beyond.
Strangely enough, the FURIA card for the A600 reminds me alot of the 030 for the A1200.

The sweetest sweetspot

[Nightshft]

Had my first A1200 in modern times and my first expansion was a Blizzard 1220 (020@28Mhz and 4 MB Fastram). It already is a real fine and noticeable boost. I was really surprised how much from the speed-up doesn't come from the cpu but just from having some fastram.

Now I have an Aca1233 (030@40 and more Fastram that I need) and I really like it. Not only because it makes the system faster but also it needs nearly no power (AFAIK <4 W) and thus doesnt need active cooling or get hot. From watt what I know about other 030 cards it is +- a few Watts the same. So for me it's the sweet spot too.

[KONEY]

Back in the days I exchanged an A4000 for an A3000 because '040 with IDE is slower than '030 with SCSI and never regretted

[eXeler0]

Originally Posted by grond

The 060 can also take heavy hits in comparison to the 040 if long MUL and DIV instructions are used. It's a pity there never was an 070 with the missing instructions.

Well Philips took a dump on the 070 name with their custom 68000 variant in the CDi. ;-)
Anyway, I think its a good idea to look at the situation from several perspectives.. Motorola CPU designers didn't just dump FPU instructions for the sake of it. The 040 was their first design to include an FPU on the same die as the CPU. It also hade significantly (for the time) increased caches compared to 030. You could say 68881/2 was over engineered feature wise :-) so it was always going to be tough to implement it 100% in this new unified design. The engineers hit some transistor budgets limits along the way and hade to make choices.. They decided to simplify the FPU (and fixed the "holes" with software patches). Remember, even with a simplified FPU this design was not great from a clock-speed perspective which became obvious with ever increasing clocks on the 486. 40MHz is all we ever got from this design. Even if they hadn't hit some transistor count limit, with full 68882 functionality, I don't exactly see it would have been clocked higher.

The 060 fixed a bunch of these issues but also had the benefit of improving manufacturing tech. Early 060 are 0,6um (or 650 nm), but last revision was almost half of that (0.32um or 320nm).
060 has about double the amount of transistors of the 040 and we got dual instruction pipeline and again much improved caches. At this point, most software using FPU had been recompiled for 040.. So 060 running 68882 code maybe was an issue only for a select few.

The fact that some old software runs poorly on 040 /060 because of the stuff that was removed is something that was known by the designers the second they decided for certain compromises, but something had to go. In the real world, we can be sure we wouldn't have had the same CPU with all characteristics intact but with all features of the 68882. If they had fully implemented FPU something else would have been compromised.
What counts in the end is the performance you get in the software you use. For me, the jump from 50MHz 030 to 50MHz 060 gave me 500% real-world performance increase so didn't exactly care what instructions they removed ;-)

[d4rk3lf]

Originally Posted by eXeler0

You can see screenshots here, including wireframe view from Imagine:
https://www.exretro.com/galleries/co...ers/index.html
I had 80MB RAM on my Blizzard 1260. I bought the SCSI kit just so that I could get another RAM slot. So I had 64MB on the accelerator and 16MB on the SCSI kit. The scene would run out of RAM if I tried to render it with scanline renderer, but for some reason the raytrace mode worked.

Wow - that's amazing render mate - congrats!

I guess "scanline render" meant using the shadow maps, instead of raytrace, and that makes sense, because shadows maps needs to be stored in Ram like bitmaps, while raytrace is just math.
I assume, if you could make it run with shadow maps (if you had more ram, or made shadow maps les res), render would be waaaaay faster.

Anyway, awesome renders and work.

I've played a little bit with Lightwave in Winuae few months a go, but unfortunately, didn't had tiome to play more (same as you - becxause of regular fx 3D work), but I was impressed what various options Lightwave offered back then.
Hopefully, I?ll have more time in the future, because I really wanna put it to the limit, and see what kind of shot Amiga can pull of.

I asume Amiga could do shots in quality of Independence day, or Starship Troopers... just someone needs to make them

[grond]

Originally Posted by eXeler0

Motorola CPU designers didn't just dump FPU instructions for the sake of it. ... The engineers hit some transistor budgets limits along the way and hade to make choices.. They decided to simplify the FPU (and fixed the "holes" with software patches).

Yes, I know. For all I know the 040 FPU is so much faster than the 881/882 that even the instructions that were dropped are faster on the 040 than in the old FPUs. Moreover, the 060 supports the same range of FPU instructions as the 040 so there was nothing missing from the 060 that didn't already miss in the 040. However, I was referring to the integer instructions that were dropped in the 060 for pretty much the same reasons, die size/transistor count/clock cycle constraints. Contrary to the FPU, the 060 does NOT support all integer instructions the 040 still had. The most prominent case are the 64bit MUL and DIV instructions (MOVEP is rarely used and usually not time critical but wouldn't cost a lot of transistors either). Here emulation via the exception vector is painfully slow in comparison even to the 030 as pointed out above by meynaf. If the 68k line of processors hadn't been discontinued, I'm sure these instructions would have been re-added with better manufacturing processes allowing for a higher transistor count. The 080 has support for all integer and FPU instructions and I would expect the same decision be made if the 68k had survived into today's billions of transistors CPU era. (The 080-FPU in its Vampire-incarnations has different limitations such as not supporting some rounding modes, rarely used number representation schemes and limited precision of calculations which would go away in an ASIC implementation).

[VladR]

Originally Posted by eXeler0

I was a Imagine3d user, (Lightwave not so much). Off the top of my head, 50MHz 060 render times were 5x faster than 50Mhz 030 (Both Blizzards).
Here's a fun little story for ya.
My last big project I ever did on Amiga was a scene with over 330,000 polygons.

Interesting project. I plugged this number to my excel sheet where I have benchmark data from Vampire (for all 3d pipeline stages) and it would render the scene in ~4 minutes using simple flatshading.

The 3D Mesh would use about 16 MB of RAM, so plenty of RAM is free.

Out of 15,000 frames (/60 = 4.1 minute), the 13,000 are used for 3D transform and quad set-up and 2,000 frames for scanline traversal.

Only 103 frames (1.5 second out of over 4 minutes) would be spent on pixel fill (actual drawing). Texturing would take slightly longer, shadows depending on approach also.


I reckon if I used indexing, it would drop to 25-50%, as edges are shared (between 2 polygons) and so are vertices (usually by 4 polygons) so no need to compute them again.


This particular scene - judging from wireframe - appears to have majority of polygons in chains and cobbles, so without them it could totally be real-time and playable [flatshaded] on Vampire as it should be under 30,000 tris easily.