Lightwave benchmark for users with PPC, 060, PiStorm, Vampire accelerators

[LaBodilsen]

Originally Posted by klx300r

Q over at Hold and Modify on YouTube just posted a Lightwave Raytrace Scene benchmark and I absolutely love this type of stuff so if you're interested please follow the link below to follow the simple steps to run the benchmark and post your results & system specs

[ Show youtube player ]

For those that don't want to watch the whole vid here's the quick setup:

1- File-Load Scene-Benchmarks folder-choose Raytrace scene
2- Camera Panel- set to low resolution- set to square pixels
3- click Render or hit F9

So ran the youtube benchmark in WinUAE on i7 9750H 1m 47s (107 seconds)

[Solskogen]

Originally Posted by LaBodilsen

So ran the youtube benchmark in WinUAE on i7 9750H 1m 47s (107 seconds)

It should do it a lot faster, I can do 95sec on a Raspberry Pi 4B+.

[LaBodilsen]

Originally Posted by Solskogen

It should do it a lot faster, I can do 95sec on a Raspberry Pi 4B+.

You where right! Apparently I had set it to -50% cpu.

New test yielded 46sec

Quite impressive speed for the Pi 4B+ though.

EDIT: Newer version of WinUAE gave a small speedboost, and now render the scene at 40sec

[Korban]

Originally Posted by Solskogen

It should do it a lot faster, I can do 95sec on a Raspberry Pi 4B+.

How?
The absolute fastest 68k emulation a pi4 can achieve is using emu68 on a pistorm, which gives about 5 minutes and 50 seconds on a pi4b+ overclocked to 2.2ghz.

[Solskogen]

I just use Amiberry and overclock the Pi to 2.2GHz. Jit is on, ofc.

[Korban]

Even emu68 on a pi4@2.2ghz is only ~%25 as fast as your result, but emu68 is significantly faster than the cpu emulation core in Amiberry, and Amiberry emulates an entire system as well, which eats into cpu resources, so I still dont understand how you're getting these results.
At about 700mips and 225mflops (which is what a pi4@2.2ghz gives on Amiberry according to sysinfo) a result of 95 seconds is impossible.

[Solskogen]

According to sysinfo I get 930Mips and 237 Mflops.

[thomassd]

Big Thax !
Try Later...

[ShK]

i9-13900KS

[klx300r]

Thanks to all that posted their times/ system info so far & keep em coming

[jkdsteve]

Just for fun:

A1200/Vamp v2 Coffin56/Gold2.12fw - 1526s (25m 26s)
Amiga4000D 3.2.1 WarpEngine060@96MHz - 1477s (24m 37s)
Vampire V4SA+ Coffin57/SA_8435.jic (x14) - 1405s (23m 25s)
Amiga4000T 3.2.1, BFG9060@100MHz - 1358s (22m 38s)
A1200/PiStorm32 w/Pi4b+ @2.2GHz.CaffeinOS - 166secs
X5000/20 2GHz MorphOS 3.18 - 119secs

MACPRO5,1 2010 2x3.5GHz 6core. FSUAE/Coffin on BigSur - 80secs
SGI FUEL 800MHz / IRIX 6.5.30f - 65s (62 without render preview turned on)
SGI 4x1GHz (1 THREAD) - 53s (50s no preview)
SGI 4x1GHZ (MAX THREADS) - 15s (14s no preview)

RAYTRACING is really heavy on FPU load, also that scene has a lot of 'edges' for antialiasing, especially with the Adaptive Samples set to 8. (I usually use 16 or 32)

[klx300r]

Amiga 1200, PiStorm32 with Raspberry Pi3a+ (March 10/23 firmware), Amiga OS3.2.2

5m59sec

[trixster]

Interesting Fuel and Tezro results @jkdsteve, thanks for including

[hammer]

Originally Posted by Karlos

There's no such thing as CISC, really. It's just a term invented by early adopters of RISC to explain how their architecture was different.

When you look at, for example, the PowerPC, it's called RISC but it's in no way a Reduced Instruction Set. It has many instructions and a lot of them are pretty complicated. What it adopts of the RISC manifesto (something I just made up but probably exists in some form somewhere) is that it essentially exports the complexity of optimum instruction scheduling to the compiler. It's also a load store architecture which is typical of RISC machines.

There was a point to all this, but it's been a very long day and I've forgotten, so I'll just finish with the closing statement, "bollocks".

The main difference between CISC and RISC is CISC's fused arithmetic load store and variable length instruction set.

RISC has discrete fixed-length instructions for arithmetic, load, and store operations.

ARM has two length instruction set to compete against X86's code density.

Modern X86 CPUs breaks up fused X86 instructions into fixed-length uops to feed the discrete arithmetic/load/store functional units. X86 instructions are effectively being used as compressed (fused) instructions.

Apple M1 has 8 instruction decoders to compete against X86's fused instruction set.

Zen 4 is using AVX-512 instruction as a compressed instruction to feed its multiple 256-bit SIMD units i.e. Zen 4 doesn't have an actual 512-bit SIMD unit and performance gain from AVX-512 usage is working around the frontend decoder bottleneck. AVX-512 instruction set for Zen 4 is to conserve instruction decoder issue slots.

[Dick Dastardly]

WinUAE V5 on an AMD 5900x @ 4.9Ghz, 39 seconds.
WinUAE V5 on an Intel I7 6700K @ 4.5Ghz, 40 seconds.

Lightwave 2020 AMD 5900x @ 4.5Ghz, All core 3.9 seconds.
Lightwave 2020 AMD 5900x @ 4.9Ghz, Single core 45.8 seconds.

AMD 5900x Render
[ Show youtube player ]

[derSammler]

Originally Posted by Bruce Abbott

Yep, even the 060 is stretching it. The Vampire stretched it a bit more, and many Amiga fans got upset about it. But at least it was still hardware, and not too outrageous. Pistorm is a ludicrously fast software emulation of a 68k CPU, RTG and who knows what else in the future.

What's the problem you say? The problem is it makes 'Amiga' a meaningless label.

You are somehow contradicting yourself. So the Vampire only "stretched it a bit more" but was still hardware? The Vampire provides an imaginary 68080 with new instructions that never existed. THAT makes "Amiga" meaningless. Anything compiled for the Vampire is incompatible with any real Amiga and any Amiga emulator. The PiStorm/Emu68 in contrast is still just a 68040.

Also, FPGA lovers seem to hate the word "emulation". But most don't even get how a modern FPGA works. The difference to Emu68 is really not that great. FPGAs are not like PALs/GALs, which got their logic burnt in. When an FPGA is powered on, its programming is read from flash memory and copied to internal SRAM. Then the FPGA behaves according to that. The bare-metal Emu68 does more or less the same: it loads startup code from flash (SD card) into RAM and executes it on the CPU. The main difference is really only how the emulated CPU is described. For FPGA, it's all done by describing the logic, whereas Emu68 describes the CPU by its instructions, registers etc. (so low-level vs. high-level) The end result is the same - the Amiga sees a real CPU on which it is executing code.

The PiStorm with Emu68 is not more or less of real hardware than the Vampire.

[Promilus]

@derSammler - sure, FPGA ain't no "68k in a flesh" but still it is AC68080 synthesized in a form of programmable logic. But to "emulate" there first would have to be real, existing processor it tries to pose as. AC68080 does not pose as any existing 68k processor hence it is not emulation. It does emulate 68k bus signals and cycles but that's pretty much the same as any programmable logic did on Amiga turbo cards as 68060 certainly has different signals and bus timings than 68000. Right? Right. And yet you could've use 68060 on A2000 with Blizzard 2060 right? Right. So Blizz 2060 emulates 68000 CPU timings and signals to work with A2k motherboard. And no. Emulation has no negative meaning in terms of computer system. It's widely used and pretty much has been for decades in one way or another. It's just silly ppl making it look like something bad.

[alcomatt]

Originally Posted by jbenam

Honestly I would’ve expected much more out of the Icedrake, especially since it’s such an expensive piece of hardware and the fact that Vampire users keep boasting about its (supposed) power.

It will never be as fast as pistorm but It also provides a lot of extra functionality, including rudimentary 3D accel and simd extensions. Mind you you will need software specifically written to utilise these.

As a package on the whole, I personally prefer the Icedrake, only because it offers a lot of extras like usb, 16-bit audio, aga over hdmi etc...

I suspect, in time, pistorm might offer a similar functionality but not for now and certainly not with emu86 yet.

[Karlos]

I don't think the hardware capabilities of the Pi are close to being fully leveraged yet, just the raw CPU emulation.

[HornBeamSoft]

FPGA is something real. Emulator is imitation.