First Tabor/OS4 benchmark - Cow3D Warp3D Nova:

Tabor/Radeon HD7750 - 593 FPS (source)
SAM460/Radeon HD7750 - 1131 FPS (source; comment by Spectre660).

I´m surprised performance penalty is so low (and Tabor´s OS4 may be not even fully optimized yet). I´m eager to see real game benchmark (eg. Quake 3).

@pavlor

Quote:

I´m surprised performance penalty is so low...

And I'm surprised it is that high.

Eleven months left...its looking like you might be receiving some chocolate, but it isn't ready yet...

@iggy

Quote:

And I'm surprised it is that high.

High? Assuming Cow3D is not compiled for e500v2, this result looks really good.

@pavlor

Why would you assume that?

@iggy

Cow3D Nova is not compiled for e500v2 SPE code.

_________________
Amiga Kit Amiga Store
Links: www.amigakit.com | New Products | A600GS

@amigakit

Is it possible to run Quake3 benchmark to see real "classic" Warp3D performance? I know I ask too much.

@amigakit

Would be possible to compile cow3d on tabor enabling the SPE optimization?

_________________
retired

Betatesting of OS4 for Tabor will start in next weeks, CD image is ready (source: Trevor at Amiga Ireland Meetup 2017; Q/A 11:10).

@pavlor

Nice, too bad that the video has been cut too soon though.
Kamelito

@pavlor

Nova is designed to use the GPU, most of math is done on GPU for the Cow demo.

I think we need to run a lot more test then just this one to know what applications are effected and what not.

Last edited by NutsAboutAmiga on 22-Jan-2017 at 10:43 PM.

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS

@AmigaBlitter

Quote:

Would be possible to compile cow3d on tabor enabling the SPE optimization?

That would make sense.

@iggy

Quote:

iggy wrote: @pavlor Quote: I´m surprised performance penalty is so low... And I'm surprised it is that high.

The sam is one of the slowest computers on the platform, and in this particular test, the tabor only scores 50% of Sam's figures...

@TRIPOS

Quote:

in this particular test, the tabor only scores 50% of Sam's figures...

This has to be a floating point issue, as the rest of the e500 core should outperform an Applied Micro cpu.

To be fair, a version compiled for the spe fpu ought to be used.

Hopefully we won't see too much of this "shoot yourself in the foot" style of benchmarking.
Let's face it, in apps that make significant use of floating point instructions, applications will need to be compiled for Tabor's specific fpu.
Its not that odious a requirement.

@iggy

Quote:

This has to be a floating point issue, as the rest of the e500 core should outperform an Applied Micro cpu.

It could be floating point performance or maybe Tim was running OS4 in debug mode although that's unlikely. Another factor is that the ASUS 7750 used in this test has a core clock of 820Mhz whereas many of the R7 250 cards in the Sams tested have a core clock of 1Ghz or more., not that it would account for a 50% drop in performance.

@iggy

Quote:

Let's face it, in apps that make significant use of floating point instructions, applications will need to be compiled for Tabor's specific fpu. Its not that odious a requirement.

It wouldn't be if OS4 had a Linux-sized developer base. Good luck hunting down devs to recompile their programs. Then there are many only available in binary form. Up until now binary compatibility was the king, but now suddenly the responsibility is shifted to the developers to provide multiple executables for system they probably don't even own.

_________________
This is just like television, only you can see much further.

Just a quick note.

This was the first real boot of a Tabor beta system which Tim had hacked together himself, as a personal demo for me, using the X5000 pre-release distribution and some beta files provided by Thomas Frieden. Expect to see major performance improvements in the coming weeks and months once the AmigaOS4.1 iso is configured for the Tabor machine.

Personally, I was really pleased having only seen Linux running on the A1222 before and wish to thank Tim for his work. It's been a long time coming but the Tabor beta testers can soon put their Linux distros aside and start testing the OS that the Tabor board was designed for.

TrevorD

PS I must learn not to be excited and show early pre-release info.

Last edited by TrevorDick on 23-Jan-2017 at 09:34 AM.
Last edited by TrevorDick on 23-Jan-2017 at 09:28 AM.

_________________
No, I don't need no reason, I'm just breezin'

@TrevorDick

That is fine, I just hope that it will be release to end users this year.
Regards
Kamelito

@kamelit0

Me too

TrevorD

_________________
No, I don't need no reason, I'm just breezin'

@BSZili
Quote:

It wouldn't be if OS4 had a Linux-sized developer base. Good luck hunting down devs to recompile their programs. Then there are many only available in binary form. Up until now binary compatibility was the king, but now suddenly the responsibility is shifted to the developers to provide multiple executables for system they probably don't even own.

If for os4 on Tabor will have needs to recompile specially for them, then, we meet with facts : no one will do so, and ppls who will own tabor will only be faced with what will works for them and that to be seen what will works and what will not.

But probably they will find some solutions, like add something to kernel, which will software emulate missing stuff, or skip them, but in any case, i think its all will be problem, and make no sense to spend resources on such HW. Even Altivec is not that supported and can be said that forgoten. I even didn't mean that XENA which we don't need and no one use in end :)

_________________
Join us to improve dopus5!
zerohero's mirror of os4/os3 crosscompiler suites

@kas1e

Quote:

add something to kernel, which will software emulate missing stuff

Wasn't that the original idea.

Software which would be better being recompiled are mostly game ports from Linux and people like Huno are still around and active in the community to make multiple binaries.

Another idea is that we need a middle code solution, where the code is compiled to machine code at runtime (like in Java and Dot Net). That way the code is always optimized for the CPU make and model which ever you have in your hardware. (I'm wondering if it would be possible to license that virtual CPU stuff of Amiga Anywhere still.

_________________
Rock lobster bit me. My Workbench has always preferences. X1000 + AmigaOS4.1 FE
"Anyone can build a fast CPU. The trick is to build a fast system." -Seymour Cray