@BSzili

tlosm is right that lack of FPU performance is not unbearable. Amiga software is mostly integer demanding, except 3D games or some applications like Blender. From my experience with WinUAE/OS4, only few software titles are hindered by sloooooooow FPU. However, even some integer speed increase in case of P1022 can´t justify similar price as SAM460 that has no such problems.

@pavlor

Incidentally the performance of 3D games is what matters for me the most.

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

@BSzili

Quote:

floating point on the GPU,

this is the reason why im hoping for the future.
i dont think Trevor and others of aeon make this choice without thinking about consequences. making this choice start made me really thinking about a new smp and new sdk who will support it.
double threads will cover the gap of slower fpu cores.
probably the new radeon driver will born for cover it too

Last edited by tlosm on 10-Jan-2016 at 01:17 PM.

_________________
I love Amiga and new hope by AmigaNG
A 500 + ; CDTV; CD32;
PowerMac G5 Quad 8GB,SSD,SSHD,7800gtx,Radeon R5 230 2GB;
MacBook Pro Retina I7 2.3ghz;
#nomorea-eoninmyhome

@BSzili

If A-EON had commissioned an OpenGL driver instead of faffing about with Warp3D...

But the Tabor could be a big step up for WinUAE-OS4 users looking to get further involved at a reasonable price.

_________________

@tlosm

Quote:

dont think Trevor and others of aeon make this choice without thinking about consequences.

I think their choice was based on similar idea as p-cubed: cheap entry hardware. Let´s hope Tabor´s end user price will be more close to p-cubed too.

@tlosm

It's just silly speculations, until product is ready, and as for SMP and as for programing the GPU using OpenCL, well some has write software. And using OpenCL is not like using FPU on the CPU. the FPU on the CPU allows opcodes to be mixed, with a graphic card, you need to upload something too it (the graphic card), so it's not a replacement.

Is the CPU better then CPU in Sam460, yes..

But you need to remember this is a CPU with different instruction set, and so you need recompile programs to get most out of programs on this CPU, and AltiVec optimization won't work, as the SIMD instruction set is different, so that means a rewrite of optimized code in codecs and so on, who is going to do that exactly?

UVD is not around the corner.
SMP god how long we have waited for that.

Last edited by NutsAboutAmiga on 10-Jan-2016 at 01:36 PM.
Last edited by NutsAboutAmiga on 10-Jan-2016 at 01:33 PM.
Last edited by NutsAboutAmiga on 10-Jan-2016 at 01:31 PM.

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

@CarasGhant

That would have been awesome. I wouldn't have a problem with having our own low level API, but not one that is stuck in the late 90s.

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

@NutsAboutAmiga

Quote:

you need recompile programs to get most out of programs on this CPU

Yes i was sure about , for sure they will make (or is already done) a new sdk for make an universal executable . For sure the old ppc software will run slower on P1022 or will not run at all .
It was the same on G4 and G5 .... the G4 programs run slower or not run at all on G5.
Apple optimized and fixed it only on 3.x and better on 3.1.4 Xcode.

Last edited by tlosm on 10-Jan-2016 at 02:30 PM.

_________________
I love Amiga and new hope by AmigaNG
A 500 + ; CDTV; CD32;
PowerMac G5 Quad 8GB,SSD,SSHD,7800gtx,Radeon R5 230 2GB;
MacBook Pro Retina I7 2.3ghz;
#nomorea-eoninmyhome

@tlosm

Quote:

tlosm wrote: @cdimauro Quote: but with a much worse CPU. for what i had been watch on euae jit performances .... the emulation run about 200x on a1222 compared the sam 460....

Let's wait for some benchmarks. And prices.
Quote:

about your geek infos ... than i remember good what you write about p50xx compared with 970MP 2.5 ghz .... and is for me the proof that you are only writing bla bla bla for spoil the job of peoples who are doing something real and not only like you bla bla bla...

Then post the link, so other people can see how was the discussion about that. Otherwise your bla bla bla remains bla bla bla...

Quote:

tlosm wrote: @pavlor for what i can now of today use integer continue be most impotant thing because all the old floating points are did by the modern gpu.

That shows how much you know about how CPUs and GPUs work, and especially how code get executed.

You continue to talk of things of which you zero knowledge about.
Quote:

the good news of today is the amd make a new opensorce driver , i hope amigaos will take advantage of it... gor now linux ppc start fly on g5 now i start watching the advantage of new gpu in the machine

Then you can show how (E|Win)UAE can offload "all the old floating points" (your words) to the GPU, right?

I can't wait...
Quote:

the real advantage in any way will be when the machine will use smp on os4

Another 2 weeks...


@pavlor

Quote:

pavlor wrote: @BSzili tlosm is right that lack of FPU performance is not unbearable. Amiga software is mostly integer demanding, except 3D games or some applications like Blender. From my experience with WinUAE/OS4, only few software titles are hindered by sloooooooow FPU. However, even some integer speed increase in case of P1022 can´t justify similar price as SAM460 that has no such problems.

Andy wont agree with you.

@cdimauro

Quote:

Andy wont agree with you.

Of course, he is author of Blender port!

@cdimauro

Quote:

Then you can show how (E|Win)UAE can offload "all the old floating points" (your words) to the GPU, right?

No check with new Fs-Uae if you will be able to make first and after running on PPC

_________________
I love Amiga and new hope by AmigaNG
A 500 + ; CDTV; CD32;
PowerMac G5 Quad 8GB,SSD,SSHD,7800gtx,Radeon R5 230 2GB;
MacBook Pro Retina I7 2.3ghz;
#nomorea-eoninmyhome

@pavlor

Quote:

pavlor wrote: @cdimauro Quote: Andy wont agree with you. Of course, he is author of Blender port!

I know, but he's also the author of Sketchblock.


@tlosm

Quote:

tlosm wrote: @cdimauro Quote: Then you can show how (E|Win)UAE can offload "all the old floating points" (your words) to the GPU, right? No check with new Fs-Uae if you will be able to make first and after running on PPC

Sorry, I didn't get it.

I suppose that culture is always good even if not 68k/PPC.

https://www.youtube.com/watch?v=eDmv0sDB1Ak

Kamelito

@NutsAboutAmiga

Quote:

NutsAboutAmiga wrote: @tlosm It's just silly speculations, until product is ready, and as for SMP and as for programing the GPU using OpenCL, well some has write software. And using OpenCL is not like using FPU on the CPU. the FPU on the CPU allows opcodes to be mixed, with a graphic card, you need to upload something too it (the graphic card), so it's not a replacement. Is the CPU better then CPU in Sam460, yes.. But you need to remember this is a CPU with different instruction set, and so you need recompile programs to get most out of programs on this CPU, and AltiVec optimization won't work, as the SIMD instruction set is different, so that means a rewrite of optimized code in codecs and so on, who is going to do that exactly? UVD is not around the corner. SMP god how long we have waited for that.

The ability to seamless mix CPU/FPU and GpGPU FP instructions are dependent on programming language framework.

Has AmigaOS platform has OpenACC ported?

Intel has a different idea by expanding X86's SIMD to 256bits (AVXv2 includes GPU style gather instructions) and 512bits (AVX-512, Skylake Xeon, Canonlake) wide. It forces AMD to follow Intel's X86 SIMD evolution. AMD's dominance in current gen game console hardware market helps AMD to reach mass numbers for it's math processing model which then use to leverage other market segments.


Intel AVX-512 replaced Xeon Phi's 512bit AVX-512 like ISA which directly competes against GpGPUs in the HPC market. AVX-512 expands it's register set to 32.

Last edited by Hammer on 11-Jan-2016 at 12:45 AM.
Last edited by Hammer on 11-Jan-2016 at 12:45 AM.
Last edited by Hammer on 11-Jan-2016 at 12:39 AM.
Last edited by Hammer on 11-Jan-2016 at 12:25 AM.

_________________
Ryzen 9 7900X, DDR5-6000 32 GB RAM, GeForce RTX 4080
Amiga 1200 (rev 1D1, KS 3.2, TF1260, 68060 @ 63 Mhz, 128 MB)
Amiga 500 (rev 6A, KS 3.2, PiStorm/RPi3a/Emu68)

@Epsilon

Quote:

Epsilon wrote: @kyle Ask and sometimes you shall receive @all Today I have written a blog entry looking at the demo of the upcoming Wings Remastered Edition on AmigaOS 4.1 Final Edition on my X1000. Link is here: http://amigax1000.blogspot.com.au/2015/12/wings-remastered-demo-on-x1000.html Please enjoy! Catcha, Epsilon

Nice game. AmigaOS platform stakeholders should promote it together with OS/HW solution.

@CarasGhant

Quote:

CarasGhant wrote: @BSzili If A-EON had commissioned an OpenGL driver instead of faffing about with Warp3D... But the Tabor could be a big step up for WinUAE-OS4 users looking to get further involved at a reasonable price.

One wonders why PowerPC stakeholders mucks around with legacy killing/anti-software investment protection with incompatible instruction set.


The embedded low cost/small size AMD Jaguar CPU in game consoles are compatible (up to Intel AVXv1) with their larger desktop PC CPUs.

ARM Cortex A7 to A57 doesn't play this kind incompatible instruction set kit-bashing.

Last edited by Hammer on 11-Jan-2016 at 02:06 AM.
Last edited by Hammer on 11-Jan-2016 at 02:02 AM.

_________________
Ryzen 9 7900X, DDR5-6000 32 GB RAM, GeForce RTX 4080
Amiga 1200 (rev 1D1, KS 3.2, TF1260, 68060 @ 63 Mhz, 128 MB)
Amiga 500 (rev 6A, KS 3.2, PiStorm/RPi3a/Emu68)

delete post

Last edited by Hammer on 11-Jan-2016 at 02:03 AM.
Last edited by Hammer on 11-Jan-2016 at 02:02 AM.

_________________
Ryzen 9 7900X, DDR5-6000 32 GB RAM, GeForce RTX 4080
Amiga 1200 (rev 1D1, KS 3.2, TF1260, 68060 @ 63 Mhz, 128 MB)
Amiga 500 (rev 6A, KS 3.2, PiStorm/RPi3a/Emu68)

@kamelit0

Thanks for the link; that was interesting.

_________________
"Unix is supposed to fix that." -- Jay Miner

@kamelit0

Quote:

kamelit0 wrote: I suppose that culture is always good even if not 68k/PPC. https://www.youtube.com/watch?v=eDmv0sDB1Ak Kamelito

Very interesting. Thanks!


@Hammer

Quote:

Hammer wrote: The ability to seamless mix CPU/FPU and GpGPU FP instructions are dependent on programming language framework. Has AmigaOS platform has OpenACC ported? Intel has a different idea by expanding X86's SIMD to 256bits (AVXv2 includes GPU style gather instructions) and 512bits (AVX-512, Skylake Xeon, Canonlake) wide. It forces AMD to follow Intel's X86 SIMD evolution. AMD's dominance in current gen game console hardware market helps AMD to reach mass numbers for it's math processing model which then use to leverage other market segments. Intel AVX-512 replaced Xeon Phi's 512bit AVX-512 like ISA which directly competes against GpGPUs in the HPC market. AVX-512 expands it's register set to 32.

Exactly, but AVX/2 and especially AVX-512 (which is a splendid SIMD instruction set) serve the opposite: use the CPU for computing (mostly floating point, but integers have a very good coverage too nowadays) operations that historically were made better by a GPU, trying to compete in the GPGPU area.

That's the exact contrary of what was stated here: offloading all FP operations to the GPU, which doesn't make sense et all, since it may greatly slowdown the execution (imagine UAE that generates GPU offloading commands for each FP operation).

Thinking to use the GPU to solve Tabor's FPU weakness isn't feasable nor desirable: it's something to totally avoid.

@cdimauro

Quote:

imagine UAE that generates GPU offloading commands for each FP operation).

Although I agree with your general perception of GPU usage for FP operations, this example is pure nonesense. If you don´t want to play Quake under UAE, FPU emulation is unnecessary.

Quote:

Thinking to use the GPU to solve Tabor's FPU weakness isn't feasable nor desirable: it's something to totally avoid.

There are applications, where such use may make sense (eg. Blender).

@pavlor

sorry but


Quote:

here are applications, where such use may make sense (eg. Blender).

if i use blender with gpu accelerations the rendering become 1000 times more faster

same is Fs-uae ... if i dont use gl it become sloooooooow

Last edited by tlosm on 11-Jan-2016 at 09:40 AM.

_________________
I love Amiga and new hope by AmigaNG
A 500 + ; CDTV; CD32;
PowerMac G5 Quad 8GB,SSD,SSHD,7800gtx,Radeon R5 230 2GB;
MacBook Pro Retina I7 2.3ghz;
#nomorea-eoninmyhome