@Hammer

Quote:

In relation to video encoding, this is dependent on middleware/driver and support from GpGPU manufacture (e.g. PS3.0 from ATI R5xx GpGPU).

I don't think so. Fragment shaders are embedded in a 3D graphics pipeline. This restricts the operations you can do and the kind of operations required for compression do not fit well. You could possibly perform some operations such as preprocessing for motion analysis. However, the rest just won't work efficiently with fragment shaders. If the graphics card has dedicated compression hardware then that can be used. Otherwise, the CPU should be used to do the task. Some of the newer cards do in fact have encoding hardware built-in. These are in a separate processing unit, outside the main graphics pipeline.

PS3.0 is Pixel-Shader standard for DirectX. It's designed for 3D graphics effects, not general purpose computation. It won't help with video encoding.

I have spoken to someone who has worked on GPU assisted ray-tracers. He said that for some tasks, highly-optimised CPU code using SSE will beat GPU-based code. His colleagues and him have actually performed benchmarks and confirmed this.

Hans

_________________
http://hdrlab.org.nz/ - Amiga OS 4 projects, programming articles and more. Home of the RadeonHD driver for Amiga OS 4.x project.
https://keasigmadelta.com/ - More of my work.

@Hans

Quote:

That's an interesting idea. However, you're still stuck with the fragment shaders being part of a graphics pipeline.

This is true with the current ecosystem (not quite with DX10.x). Both ATI and NV plans to introduce unified shader model ecosystem for non-MS Vista operating system.

Refer to
http://www.ati.com/technology/crossfire/physics/Asymmetric_Physics_Processing_with_ATI_CrossFire.pdf
ATI's Data Parallel Processing Architecture without the "bloated" DX3D or OpenGL .

Quote:

If they change this so that the processing units are fully general purpose then the performance will go down. There's a reason why general purpose CPUs don't have the same processing power (in terms of vector operations/s) as GPUs. It's flexibility.

SIMD flexibility is not the issue i.e.
1. Freescale is competing against Intel CPU/NB/SB/IGP/WiFi stack (Intel Centrino).
2. Claims of crappy "Intel vector units" would be false in the light of Intel GMA 900 in one of Apple’s target market e.g. NLE "3D Edit" style application.

With point one, this is one the reasons for AMD/ATI rumored unification.

Last edited by Hammer on 23-Jul-2006 at 08:11 AM.
Last edited by Hammer on 23-Jul-2006 at 08:01 AM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

Quote:

Hans wrote: @Hammer Quote: In relation to video encoding, this is dependent on middleware/driver and support from GpGPU manufacture (e.g. PS3.0 from ATI R5xx GpGPU). I don't think so. Fragment shaders are embedded in a 3D graphics pipeline. This restricts the operations you can do and the kind of operations required for compression do not fit well. You could possibly perform some operations such as preprocessing for motion analysis. However, the rest just won't work efficiently with fragment shaders. If the graphics card has dedicated compression hardware then that can be used. Otherwise, the CPU should be used to do the task. Some of the newer cards do in fact have encoding hardware built-in. These are in a separate processing unit, outside the main graphics pipeline. PS3.0 is Pixel-Shader standard for DirectX. It's designed for 3D graphics effects, not general purpose computation. It won't help with video encoding. I have spoken to someone who has worked on GPU assisted ray-tracers. He said that for some tasks, highly-optimised CPU code using SSE will beat GPU-based code. His colleagues and him have actually performed benchmarks and confirmed this. Hans

Using ATI's slides, significant inefficiencies is with the current Gfx middleware
i.e. cite http://www.elitebastards.com/cms/index.php?option=com_content&task=view&id=69&Itemid=29&limit=1&limitstart=2

In http://www.elitebastards.com/cms/index.php?option=com_content&task=view&id=69&Itemid=29&limit=1&limitstart=4
With WDDM, it presents "ATI Driver for direct access to ATI hardware" module.


Quote:

I have spoken to someone who has worked on GPU assisted ray-tracers. He said that for some tasks, highly-optimised CPU code using SSE will beat GPU-based code. His colleagues and him have actually performed benchmarks and confirmed this.

My comment's intent was not for 3D-raytracing since we should already know Cinebench’s CPU scores from the last barefeats benchmark postings. "Crappy Intel vector units" claims has been debunked in real world applications that run on MacOS X and Windows.

Also, there are performance differences between ATI and NV GPUs when a workload is outside their intended design e.g. http://www.pcper.com/article.php?aid=226&type=expert

Last edited by Hammer on 23-Jul-2006 at 07:18 AM.
Last edited by Hammer on 23-Jul-2006 at 06:57 AM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@T_Bone

Quote:

T_Bone wrote: @Bobsonsirjonny How does old debt get cancelled without a bankruptcy judge declaring the debt cancelled? (allegedly )

I dont know how it works in the US - but in the UK it used to be the case that you could legally do such alledged asset shifting in order to avoid paying your bills. However, this has been tightened up now - so they can still persue you.

Indeed the new system makes the differences between a sole trader and limited company mute.

@cHaOs667

Quote:

cHaOs667 wrote: @Turrican AFAIK you can get a licence only for a Hardware Plattform, not for a Software Plattform

You can get a license for moon cheese if you can talk them into it.

_________________
"If you want to tell people the truth, make them laugh, otherwise they'll kill you." - Oscar Wilde

@Hans

Quote:

Hans wrote: @Georg Quote: Georg wrote: @Frags [quote]He means that they can`t grant a license for it because it doesn`t exist. I think that he means that AInc can't grant a license because current AmigaOS == AmigaOS4 which partly "belongs" (rights) to Hyperion, 3rd party coders, etc.

That's what I thought he might mean. I remember that certain parts of OS3.5/OS3.9 could not be included with OS4 because their respective authors did not give permission (e.g. that fancy skinned CD player). If this were true then AInc could give someone a license for x86 without consulting with OS4 code owners, but then have to start from scratch with the OS3.x sources. No thanks.

Hans
[/quote]

Bernd was licensing components from AOS4 for Umilator2, IIRC.

_________________
"If you want to tell people the truth, make them laugh, otherwise they'll kill you." - Oscar Wilde

@Hammer

Hopefully this will be the last post regarding the GpGPU stuff. I get the feeling that we're talking about slightly different things.

Yes, the current graphics API's add overhead. Yes, replacing them with a better one will make things more efficient. But, no, the new API won't allow you to use the GPU efficiently for any algorithm. You are still tied into the graphics-pipeline. Looking at the ATI doc you referenced, their DPPAI is still using RGBA textures for input to the shaders, and RGBA images for output. The 3D raytracing thing we discussed is an example of this.

Having said all this, physics simulations are generally well suited to GPU implementation, particularly if it's all based on first-order dynamical systems. It's great that they're looking at creating a new API that officially support this.

Finally, the AMD + ATI merger is now official. It's happened.

Hans

Last edited by Hans on 24-Jul-2006 at 08:24 PM.

_________________
http://hdrlab.org.nz/ - Amiga OS 4 projects, programming articles and more. Home of the RadeonHD driver for Amiga OS 4.x project.
https://keasigmadelta.com/ - More of my work.

@Hans & @Hammer

I take it that you guys have probably heard about AMD buying ATI today right?

_________________
When swimming with sharks, make sure to bring lots of band-aids...

@elatour

I know that many of you will hate me for causing this thread to rear it's ugly head again, but I can't resist in wanting to offer even more food for thought on the subject.

In case none of you have seen the latest Consumer Reports, seeing as Mac laptops are now x86-based, they are now not relagated to a seperate class of computer or oddity in CR's evaluations and have been pitted side by side against Windows systems. Well, guess what? ...they got top marks in 3 out 4 classes of laptops. This is bound to add more momentum to the surge in the Mac's growth in the desktop market. This would appear to be more validation that the Mac move to x86 has been a huge success for Apple. But again, only time will tell whether this will last or not.

Last edited by elatour on 04-Aug-2006 at 04:33 AM.
Last edited by elatour on 04-Aug-2006 at 04:32 AM.

_________________
When swimming with sharks, make sure to bring lots of band-aids...

@elatour

My God, man! What are you doing? Are you trying to get the thread back onto the original topic??

Question for Those Skilled In The Art: Are the MacTel machines fundamentally different from Wintel machines so that it is impossible or very difficult to dual-boot them?

Presumably Apple have made it difficult to dual-boot Windows?

_________________
cheers
tony

Hyperion Support Forum: http://forum.hyperion-entertainment.biz/index.php

@tonyw

Quote:

Presumably Apple have made it difficult to dual-boot Windows?

Quite the opposite. Apple is actually advertising "Macs do Windows, too."


They offer a free download right from there sight that enables you to dual boot MacOS X and Windows. It's called boot camp.

http://www.apple.com/macosx/bootcamp/

I think there are quite a few people that will buy Mac laptops just to run Windows. Many Windows users have always wanted one of those sweet Apple laptops, but didn't care to use MacOS. Now they can have the best of both worlds.


AmigaHeretic

_________________
A3000D (16mhz, 2MB Chip, 4MB Fast, SCSI (300+MB), SuperGen Genlock, Kick 3.1)
---------------------------------------------------------------------------
Back in my day, we didn't have water. We only had Oxygen & Hydrogen, & we'd just shove 'em together

Quote:

I wonder what is so hard to get about that.

What's so hard to get is why you haven't got a licence for it...

You said several times that thousand of money had been invested into the development of OS4. You also said that what was keeping OS4 from being released was (mainly ?) the lack of hardware.

Now the Pegasos2 is here, and has been around for a long while...

Let's see the story this way: you said the port to (Pegasos) could take about 3 months. If you decided to go the Pegasos way at this time, OS4+Pegasos could be bought by anyone since... well.. 3-4 months ? Let's say only 1 month if we include some testing, papers stuff,...

So: what are you waiting for ?

Maybe (surely) we don't know everything...

You said ExecSG belonged neither to Hyperion, nor to Amiga. So, who does it belong to ? I guess some other components belong to other people: are all these people so open-mindedless that they prevent OS4 from being licenced/ported to other hardware ?

Leo.

Last edited by Leo on 04-Aug-2006 at 10:53 AM.
Last edited by Leo on 04-Aug-2006 at 10:16 AM.
Last edited by Leo on 04-Aug-2006 at 10:15 AM.
Last edited by Leo on 04-Aug-2006 at 10:09 AM.

_________________
http://www.warpdesign.fr/

@Leo

Quote:

Leo wrote: You said ExecSG belonged neither to Hyperion, nor to Amiga. So, who does it belong to ? I guess some other components belong to other people: are all these people so open-mindedless that they prevent OS4 from being licenced/ported to other hardware ? Leo.

From what I hear, the others involved with OS4 are just as pissed as everyone else at the standoff. They worked on an OS that they'd like to see being USED by people.

_________________
"If you want to tell people the truth, make them laugh, otherwise they'll kill you." - Oscar Wilde

@Hammer

Quote:

Pixel shaders are just vector units and Intel builds these vector units in IGPs

And I seriously doubt that intel parts are comparable in speed to the latests ATI/nVidia cards

Quote:

Refer to why Intel markets the “Centrino” platform

I'm not interested in intel marketing. Pentium III accelerates internet and you don't need to buy graphic cards with lot of ram (because intel invention, AGP is marvelous, a good way to make the user think he needs a new cpu uh).

Tell the guys of simdtech that intel vector units are wonderful and later tell me their replies.

Quote:

Didn't you notice power consumption between IBM CPC925/CPC945 Northbridges vs ATI and Intel Northbridges

IBM can redesign stuff on demand, that's the reason Xenon, CELL, Gecko and other cpus exist.

Quote:

Mitac 8355 laptop with Athlon 64 mobile and ATI M10P has total system power of ~50 Watts while running a DX game. What is wattage of STI’s Cell again? Hint: ~80 Watts.

I doubt Athlon 64 mobile provides good performance comapred to desktop dual G5 or MPC8641D. And that (crappy) Athlon 64 requires Rosseta to run PPC binaries while MPC8641D would run them natively MUCH faster.

have you noticed that Cell is not finished and final versions will require less power?

_________________
The only spanish amiga news web page/club: CUAZ

@Hammer

Quote:

Personal issues is irrelevant in this case.

I guess that means you are bored at home usually and spend your time trying to memorize buzz words and benchmarks.

_________________
The only spanish amiga news web page/club: CUAZ

@Hammer

Quote:

Altivec's superiority over SSE is dependant on 1. implementation of Altivec hardware, 2. implementation of SSEx hardware, 3. applications being run, 4. complier’s quality, 5. system platform quality i.e. core logic performance and availability. 6. Vector instruction availability e.g. _a. Altivec is only works in SIMD 32bit FP while SSE2 works in SIMD 64bit FP. A problem for Freescale’s G4 but not with G5’s dual 64bit FPU. _b. RC5's style instruction usage and gain doesn't reflect real world desktop applications. Should one rehash yet another barefeats benchmark posting? _c. Not everything can be vectorise. i.e. general ALU performance is important.

In the end Altivec is faster.

as you love so much benchmarks, check out this

_________________
The only spanish amiga news web page/club: CUAZ

@Hans

Quote:

I have spoken to someone who has worked on GPU assisted ray-tracers. He said that for some tasks, highly-optimised CPU code using SSE will beat GPU-based code. His colleagues and him have actually performed benchmarks and confirmed this.

Writing highly optimised CPU code over crappy GPU code the CPU will typically win so yes.

@T_Bone

Quote:

From what I hear, the others involved with OS4 are just as pissed as everyone else at the standoff. They worked on an OS that they'd like to see being USED by people.

This licence rubbish is destroying the whole Amiga revival!

Gateway - Why buy the Amiga patents if this is the outcome? Was it worth it just for the 2-button mouse Commodore patents that no-one honoured anyway?

Amiga Inc. - What a confused mess of a company! Who or what do they stand for? The first & greatest multimedia computer, ignored and instead used to promote Mobile Gaming?!!@#!

Hypernion - All the best turning a profit in this sea of bureaucracy!

Apple - Fair play! I guess years of slick advertising has paid dividends! Given enough time even Apple is capable of delivering useful and value for money computers (well MacMini and basic MacBook are a start). A far cry from its projects in the Commodore (1980's-early 1990's) era!

_________________
"Art challenges technology. Technology inspires the art."
John Lasseter, Co-Founder of Pixar Animation Studios

Quote:

Crumb wrote: In the end Altivec is faster.

However Intel and AMD hardware has been beating PPC hardware fairly substantially on Lightwave, Maya, Modo and Mirage and now the X86 macs beat there PPC counterparts rather handily across the whole gamut of Apple apps and with next weeks Big Box Intel announcement, we'll probably see even bigger deltas for the mac. On the other hand, I'd still like to see OS4 on a G5 mutlicore, whether thats one of those from Apple or one of those from Genesi, it would be nice. I'd probably even be willing to port some of the ImageFX multithreaded stuff back to it.
-Tig

_________________
We played the first thing that came to our heads, it just happened to be the best song in the world.

@Crumb

Not quite in same level as “Edit 3D Gold” in regards to GpGPU usage as SIMD/MIMD co-processor for RT effects....

Refer to
http://www.barefeats.com/mvdcpc.html
(Not factoring 2-way sockets)
http://techreport.com/reviews/2006q3/core2/index.x?pg=13

In Cinebench 9.5

1-way Dual Core IBM PowerPC 970 @2.7Ghz = 701 CB
1-way Dual Core AMD Athlon 64 X2 4400+ @2.2Ghz = 757 CB
1-way Dual core AMD Athlon 64 FX-62 @2.8Ghz = 879 CB
1-way Dual core Intel Core 2 Extreme @2.9Ghz = 943 CB

Against Intel Core Duo.
1-way Dual Core IBM PowerPC 970 @ 2.0Ghz = 521CB (MacOS X PPC)
1-way Dual core Intel Core Duo @2.0Ghz = 605 CB (MacOS X IA-32)

In Photoshop CS2

1-way Dual Core IBM PowerPC 970 @2.7Ghz = 24 sec.
1-way Dual Core AMD Athlon 64 X2 4400+ @2.2Ghz = 21 sec.

Note that AMD is now involved in this topic through ATI’s SIMD/MIMD GpGPU technology.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)