Sign in with
Sign up | Sign in
Your question

Multi-core processors may replace physics cards, says Intel

Last response: in CPUs
Share
September 26, 2006 2:19:01 PM

Quote:
"With quad-core you could have things like rabbits running around and realistically do what rabbits usually do"


So how many cores will we need to handle all the rabbits that's going to generate?
a b à CPUs
September 26, 2006 8:15:50 PM

IMHO that is very unlikely. If you have a quad core or multi-core and a game requires all of it to run then there is no way a cpu will have the time
to calculate physics.
September 26, 2006 8:26:32 PM

they (games) wont be optimized for some time yet, therefore you have extra idle cores ready to do some physics, at least until games become multi threaded :) 
September 26, 2006 8:32:24 PM

Quote:
"With quad-core you could have things like rabbits running around and realistically do what rabbits usually do"


So how many cores will we need to handle all the rabbits that's going to generate?

lol...interesting question; at IDF today, Intel announced the introduction of 80 cores in one die. They expect to have it ready -as for mainstream I'm really not sure- in the next 5 yrs. That might answer your question!
September 26, 2006 8:44:44 PM

Well it would be a good strategy for Intel to incorporate the physics calculations in the cpu. After all it's what it's made for, calculations. I'd rather have the physics processor in the cpu or gpu for it is faster that way, rather than to buy another card to do it. Take Ageia for example, with their Physx card didn't have a big impact (for now) and it is limited due to games not supporting it.
September 26, 2006 8:48:20 PM

80...i think you mean 8 core thats planned for 2009 on the Yorkfeild Core. The only mension of 80 was to do with
Quote:
The traditional outlook into the future included a 300 mm2 processor die that integrates 80 floating point processing cores
That's something entirely different from mainstream.
http://www.dailytech.com/article.aspx?newsid=4336

As far as physics go....gfx cards will always do that much better than any standard processor could. There was an article that showed how a processors calculations differ from a video cards calculations and also showed that GPU's are moving at a much faster pace than CPU's and Gpu's are much more capable at processing information such as physics. It said that Current GPU's are about 40times more powerfull than a modern processors. But for the life of me I am unable to find the article, it was posted sometime withing the last month.
September 26, 2006 8:55:49 PM

I think when they mean that games will be multi-threaded will be that there is a different thread for each individual aspect of gameplay, i.e. AI, physics, etc... Which would make a lot of sense. I want to see the AI in future RTS games with multi-threading...
September 26, 2006 9:35:32 PM

Thanks for the correction! I recall the article you're trying to make reference to, but -seriously this time- I think that if a given core can integrate 80 FP units, what is that telling you? It can actually perform as or outperform the best GPU solution out there as we speak. If I'm not mistaken on this -as I'm quoting straight from what I can remember from the article- was that it such core can do handle one trillion calculations. That's blazing fast...well, not that fast compared to other MP units out there. Thanks for clarifying the 80 FP cores thing!
September 26, 2006 11:00:20 PM

No problem, and as far as processors go, although the current trend is that gpus are advancing faster than cpu's, that doesn't mean that the trend cannot change. Intel obviously is/has introduced a lot of revolutionary new products/ideas into/onto the market within the last 2 months. Thats some serious progress made in such a short time. Maybe those huge layoffs over at intel did help them out quite a bit *hehe*.

Processors cannot generate an image without the help of a some sort of graphics accelerator, whether its an intel igp,via igp,sis igp, nvidia go, or radeon xpress, or a true video card such as a Radeon 1950xtx or a Nvidia 7950gx2 the processor cannot process images alone.

Just try to imagine playing Oblivion on a Core 2 duo e6700 with a ati radeon xpress 200 igp :!: ....it's damn near impossible because the cpu is made for calculations not visuals. But GPUs are so powerfull they can calculate stuff in real time faster than any cpu as well as generate all sorts of insanely complex physics,geometry, arcitecture,effects, and shaders at the same time. 8O

Now with the introduction to stand alone physics processors/ video cards its going to get even better.
September 26, 2006 11:22:16 PM

Quote:
No problem, and as far as processors go, although the current trend is that gpus are advancing faster than cpu's, that doesn't mean that the trend cannot change. Intel obviously is/has introduced a lot of revolutionary new products/ideas into/onto the market within the last 2 months. Thats some serious progress made in such a short time. Maybe those huge layoffs over at intel did help them out quite a bit *hehe*.

Processors cannot generate an image without the help of a some sort of graphics accelerator, whether its an intel igp,via igp,sis igp, nvidia go, or radeon xpress, or a true video card such as a Radeon 1950xtx or a Nvidia 7950gx2 the processor cannot process images alone.

Just try to imagine playing Oblivion on a Core 2 duo e6700 with a ati radeon xpress 200 igp :!: ....it's damn near impossible because the cpu is made for calculations not visuals. But GPUs are so powerfull they can calculate stuff in real time faster than any cpu as well as generate all sorts of insanely complex physics,geometry, arcitecture,effects, and shaders at the same time. 8O

Now with the introduction to stand alone physics processors/ video cards its going to get even better.


One deals with a static flow of information very well GPU, the other deals with dynamic flow of information very well CPU. To say one is more powerful than the other based on individual code bases is very narrow minded and misleading for that matter.
September 26, 2006 11:28:18 PM

Well, calling me Narrow minded for basically repeating what i read on anandtech somewhere is pretty harsh.

Yes those both deal with seperate calculations.

Also if you whould have read the article you would have seen that the guy modified his gpu using c++ or something rather to get some godly performance gain using his gpu to assist calculations on his cpu. Like 40x as fast. God I wish i knew where i saw that article.
September 26, 2006 11:35:15 PM

Quote:
Well, calling me Narrow minded for basically repeating what i read on anandtech somewhere is pretty harsh.

Yes those both deal with seperate calculations.

Also if you whould have read the article you would have seen that the guy modified his gpu using c++ or something rather to get some godly performance gain using his gpu to assist calculations on his cpu. Like 40x as fast. God I wish i knew where i saw that article.


I never said you were narrow minded your thought on the matter though is.
September 26, 2006 11:36:32 PM

It's not my thought...Like i said, I'm basically repeating what I read from Anandtech.
September 26, 2006 11:48:38 PM

I think the implementation of multi-core games will arrive quicker than you think, although it won't be done elegently at first.

This is because mre and more developers are forced to develop for multi-platform, i.e. consoles and PCs, and consoles are already at the multi-core stage. Crude implementation may well happen soon therefore, as it results in significantly cheaper development.

This may also mean that whilst today's graphics cards can run physics faster, physics will first be implemented on the CPU, and this won't change until the next 'next-gen' consoles are out in 2 years or so.

More and more, publishers cannot afford to allow a developer to code for one platform only, and then re-code for another.
September 26, 2006 11:56:37 PM

Quote:
It's not my thought...Like i said, I'm basically repeating what I read from Anandtech.


Fair enough Anandtech doesn't sit very high on my list of "perfered sites" anyways. Sorry for the misunderstanding I should be more clear in regards to my posts, I have been slipping over the last fews days.
September 27, 2006 3:06:55 AM

Quote:
IMHO that is very unlikely. If you have a quad core or multi-core and a game requires all of it to run then there is no way a cpu will have the time
to calculate physics.


Wait a second, games with good physics exist without physics cards, just what do you think is calculating all those collisions?

Excellent point.

Of note in support of that point is that the Physics card manufacturer, Ageia, was "cheating" when they released the dazzling Cell Factor demo. To play the game, you had to have a physics processing card. This was, of course, to handle the massive amount physics processing required by the level of interactivity within the games environment. In reality the game was simply soft locked and within a short time some clever person found a work around alowing the game to be played without a Physics Processing Unit (PPU)

http://www.tomshardware.com/2006/07/19/is_ageias_physx_...

As shown in the link above, there was a work around enabling users to bypass the lock and play without the PPU card, but what is more important was game performance with and without the PPU

The comparative results, located on page five of the article show no signifcant improvement using the PPU vs not using it. In short, its a waste of money

http://www.tomshardware.com/2006/07/19/is_ageias_physx_...

The article is a very good read for those who may have heard the future is PPUs. The stand alone or add on PPU seems to be an idea whose time came and went before anyone knew what it was. If we were still at the athlon thunderbird/PIII CPU level, PPUs would be a must, but now that were about to have CPUS with umptyfratz cores, its seems kind of pointless, except from a marketeering perspective.
September 27, 2006 3:16:16 AM

Quote:
Wait a second, games with good physics exist without physics cards, just what do you think is calculating all those collisions?


Combination of cpu and gpu, or just gpu may me :?:
September 27, 2006 3:44:50 AM

Quote:
I think the implementation of multi-core games will arrive quicker than you think, although it won't be done elegently at first.


I read that Crysis's recommended specs call for a dual core CPU; it looks dual-core support has finally arrived (mind you it looks like Crysis is going to be a PC slayer).
September 27, 2006 4:18:53 AM

Quote:
IMHO that is very unlikely. If you have a quad core or multi-core and a game requires all of it to run then there is no way a cpu will have the time
to calculate physics.


Wait a second, games with good physics exist without physics cards, just what do you think is calculating all those collisions?

Excellent point.

Of note in support of that point is that the Physics card manufacturer, Ageia, was "cheating" when they released the dazzling Cell Factor demo. To play the game, you had to have a physics processing card. This was, of course, to handle the massive amount physics processing required by the level of interactivity within the games environment. In reality the game was simply soft locked and within a short time some clever person found a work around alowing the game to be played without a Physics Processing Unit (PPU)

http://www.tomshardware.com/2006/07/19/is_ageias_physx_...

As shown in the link above, there was a work around enabling users to bypass the lock and play without the PPU card, but what is more important was game performance with and without the PPU

The comparative results, located on page five of the article show no signifcant improvement using the PPU vs not using it. In short, its a waste of money

http://www.tomshardware.com/2006/07/19/is_ageias_physx_...

The article is a very good read for those who may have heard the future is PPUs. The stand alone or add on PPU seems to be an idea whose time came and went before anyone knew what it was. If we were still at the athlon thunderbird/PIII CPU level, PPUs would be a must, but now that were about to have CPUS with umptyfratz cores, its seems kind of pointless, except from a marketeering perspective.

I downloaded and installed Cell Factor but couldnt figure out for my life how do do the second part where you had to go into the hard drive and find the file and change it from true to false so if anyone wants to help me out so i can see how amazing the physics are, i would greatly appriciate it. Cus i really wanted to try out everything and see how it worked but i couldnt get into the hard drive to change it and the little pathway that they give you didnt work, and yes i changed the drive letter to C: which is what my drive is so it wasnt that.
September 27, 2006 4:19:29 AM

This is the quote from our press release..

"To demonstrate how Moore’s Law will continue well into the future, Otellini showed a new research prototype processor that has 80 floating point cores on a single die. The tiny silicon die on this experimental chip, just 300mm², is capable of achieving a Teraflop of performance, or one trillion floating point operations per second. "
September 27, 2006 10:55:36 AM

Quote:
This is the quote from our press release..

"To demonstrate how Moore’s Law will continue well into the future, Otellini showed a new research prototype processor that has 80 floating point cores on a single die. The tiny silicon die on this experimental chip, just 300mm², is capable of achieving a Teraflop of performance, or one trillion floating point operations per second. "


I would say within 5 years there would be way more then enough hardware power for photorealistic games but will the game developers and programmers be able to take advantage of the powerful hardware...I doubt it.
September 27, 2006 11:09:11 AM

Quote:
This is the quote from our press release..

"To demonstrate how Moore’s Law will continue well into the future, Otellini showed a new research prototype processor that has 80 floating point cores on a single die. The tiny silicon die on this experimental chip, just 300mm², is capable of achieving a Teraflop of performance, or one trillion floating point operations per second. "


I would say within 5 years there would be way more then enough hardware power for photorealistic games but will the game developers and programmers be able to take advantage of the powerful hardware...I doubt it.

http://mfile.akamai.com/28603/wmv/intelstudio.download....

That is a link to the IDF conference. At about 16 minutes or so, they have a presentation of Alan Wake being run on a Kentsfield. It is a good demonstration, being presented by the game developer. They do state that they are developing it for multicore processors. The game does look really good IMHO. I don't even want to think about how much that computer cost though. Falcon NW running a 3.7ghz Kentsfield.

Good video though, check it out.
September 27, 2006 2:37:02 PM

a x86 cpu can t match a dedicated proc. in its own aps,even if the latter is at a much lower frequency.....for those who are familiar with the subject, a celery 333 with a 100 mhz rage pro videochip with just 4 megabytes can decode a divx as well as a p3 1 giga with a geforce.....
a b à CPUs
September 27, 2006 3:04:49 PM

Quote:
a x86 cpu can t match a dedicated proc. in its own aps,even if the latter is at a much lower frequency.....for those who are familiar with the subject, a celery 333 with a 100 mhz rage pro videochip with just 4 megabytes can decode a divx as well as a p3 1 giga with a geforce.....


Uh?

You mean DVD right? Not DiVx. ATi Rage Pro has hardware assisted DVD decoding.. not DiVx.

Now, a dedicated processor for physics can replace a dedicated physics processor but with some minor limitations. The reason why a dedicated DVD decoding chip is able to alleviate some of the processing burden off a Celeron 333MHz is due to the fact that A. 333MHz Celeron's are not fast processors to begin with. B. It's a single core processor having to process multiple threads already on top of the DVD decoding you're asking it to do. Multiple Cores means many cores will be dedicated to physics processing. In other words they won't be doing anything else but Physics processing. Totally worth it IMHO.
September 28, 2006 6:50:43 PM

In the Alan Wake Trailer in the Intel Conference, The Founder of Remedy explains about the game, during the hurricane scene, a whole cpu core is used just for the physics simulation alone. So this option already exists and working (although the game is not released yet). In that case, dual core and quad core are more important for gamers, than single core.

Some points:

-Do you think more games will enable a dedicated core just for physics in the future?

-Will Quad Core work better than Dual Core for Alan Wake? In Dual Core, one Core is sometimes used for the physics alone, and one core for the rest, so will Quad Core be even better than Dual Core for this game? Where are the other 2 cores used in Quad Core?
I just wonder about this game, and Dual Core vs Quad Core, if theres that much difference.

-Does it matter if the Quad Core chip used is a 2 'glued together'
dual-core processors, or one native single-die quad-core chip?

Will it affect anything if the Quad Core chip used for gaming/applications is the 'glued together' version and not the single die Quad Core chip?

Thanks
!