Cell Processor vs Dual-Processor Apple Power Mac G5

aldo

Newcomer
Spectrum Online - We're flying at about Mach 1.5 around Mount Saint Helens, in Washington state. IBM Corp. senior programmer Barry L. Minor is at the controls, rocketing us over the crater and then down to the lake at its base to skim over the tree trunks that have been floating there since the volcano exploded over 25 years ago. The flight is exhilarating, even though it's just a simulation projected on a widescreen monitor in a cluttered testing lab.

Then, at the flick of a switch, Minor turns the simulation over from his new Cell processor to a dual-processor Apple Power Mac G5, and the scenery freezes. The G5 almost audibly groans under the burden, though it's no slouch. In fact, it's currently the top of the line for PCs. But Cell is something different entirely.
Nothing too special, but thought some might find this interesting.

-aldo
 
aaaaa00 said:
Old. Sounds like the same terrain demo they've been telling us about for months.
That's what I'm thinking too. I just didn't know if there's very much revealed in the simulation aggressively out-performing the Mac G5 though?

-aldo
 
Sony showed it at the E3 conference - so Gamespot or IGN should have video (and I do remember Gamespot, at least, having an individual video for it.
 
cool, thanks for the vid

I think this maybe be usable in games but to a some extent, anyone know how much ram they used in the system to render this
 
1 Cell & 2 Cell

Dr Evil said:
If it is the same landscape demo that was seen at E3, it used 2 Cell's, still beating G5 by a healty margin though.

With 1 CELL it is 720P @ 30fps with 2 CELL it is 1080P @ 30fps but I do not know what is performance of PS3 CELL with 7 SPE. I think terrain demo CELL has 8 SPE.
 
Dr Evil said:
If it is the same landscape demo that was seen at E3, it used 2 Cell's, still beating G5 by a healty margin though.

I totally forgot about the details of that demo, was it using 2 Cells and a GPU or just 2 Cells?
 
Old news and yes, very (very very very) impressive. Still the fact they showed it off is at least good for credibility.
 
Synergy34 said:
I totally forgot about the details of that demo, was it using 2 Cells and a GPU or just 2 Cells?
2x 2.4GHz Cell DDE? (I guess DDE2) no GPU.
 
Synergy34 said:
I totally forgot about the details of that demo, was it using 2 Cells and a GPU or just 2 Cells?

http://www.research.ibm.com/cell/whitepapers/cell_online_game.pdf#search='terrain%20demo%20CELL%20SPE'

ihamoitc2005 said:
http://www.research.ibm.com/cell/whitepapers/cell_online_game.pdf#search='terrain%20demo%20CELL%20SPE'

Barry Minor of IBM answers some questions:

http://gametomorrow.com/blog/index.php/2005/07/26/beyond-polygons/

Yes, I believe that with the right algorithms and tuning the current Cell processor will be able to ray-trace complex scenes at real-time speeds. Philipp Slusallek has shown examples of this on Cell hardware at Siggraph 05. We (IBM) have shown ray-casting of multi-sampled 720P HDTV images at over 30 frames per second with only one Cell processor.

Terrain Rendering Engine

Also see this:

http://www.graphicshardware.org/presentations/damora-cell4graphicsandviz-gh05.pdf

CELL mentioned in presentation is 2.8Ghz with 8 SPE

30+ frames per second with only one Cell processor
– No graphics adapter assist
– 1280x720 (HD 720P) resolution
HD 1080P at 30+ frames per second via 2 way SMP
Advanced SPE shader function
– Ray/Terrain intersection computation
– Texture Filtering
– Normal computation
– Bump map computation
– Diffuse + Ambient lighting model
– Perlin Noise based clouds
– Atmosphere computation (haze, sun, halo)
– Dynamic multi-sampling (4 – 16 samples per
– Image based input (16 bit height + 16 bit texture)
– 29 KB of SPE object code
– 224 KB of SPE local store data
M-JPEG compression via SPE
Performance scales linearly with number of available
Written completely in C with intrinsics
Currently up and running on bring up systems!
 
ihamoitc2005 said:
With 1 CELL it is 720P @ 30fps with 2 CELL it is 1080P @ 30fps but I do not know what is performance of PS3 CELL with 7 SPE. I think terrain demo CELL has 8 SPE.

One of the SPUs was being used for jpeg compression of the framebuffer, to send it over the network to a client. 7 of the SPUs were being used for the actual rendering. If you were just interested in rendering the scene to a "local" screen, connected directly to the system, PS3 Cell's performance would be the same (well, difference in clockspeed aside, depending on whether you're looking at 2.4Ghz Cell numbers, or 2.8 or 3.2).

edit - Actually, to correct myself, from the article:

Because the compression takes less time than rendering the graphics, the compressing processor automatically switches gears when it's finished and runs the rendering algorithm on a portion of data until it's needed for compression again. With each frame, the process starts over.

So that SPU that's used for compression is also actually doing some rendering work. So even in the above scenario I described, there'd probably be a slight drop in performance.

Didn't realise initially that there were 3 pages to the article linked to!

Also, another interesting detail that was new to me at least:

When Cell runs Minor's volcano simulator, it waits for data to arrive from memory for only 1 percent of the time; the G5, in contrast, stands idle for about 40 percent of the time.
 
Last edited by a moderator:
Ah, I'd be more interested in seeing a dual core G5 against the triple core Xenon. Are IBM's inorder cores superior to their out of order cores?
 
variable SPE use

Titanio said:
So that SPU that's used for compression is also actually doing some rendering work. So even in the above scenario I described, there'd probably be a slight drop in performance.

Yes. This makes sense to me because SPE can take load as what is required. So maybe actual performance is same for PS3 CELL because since SPE can take load as what is required, 7 SPE @ 3.2Ghz of PS3 is same performance as 8 SPE at 2.8Ghz which is used for terrain demo.

I am curious how missing SPE has effect on jumps for EIB data transfer and EIB bandwidth efficiency.
 
It seems that the Earth Terrain Demo is made for a CPU with huge vector processing and it runs in Vector Processing Units only instead os Scalar Units.

The G5 isn´t underpowered, clocked at 2.5Ghz it could do 25 GFLOPS (FPU) and others 40 GFLOPS (VMX), is a very good Floating Point performance for a General Purpose CPU that only has one core.
 
I find this demo very interesting, and can see large height maps being used for many types of games. Now the resolution of this one was 10 meters between points, and would be suitable for a flight game of some sort, giving an area of 80 km by 80 km.

The matrix size was roughly 8000x8000 using 16-bit values, and that translates into 128 MB of data. That's a monsterous size matrix, and CELL had no problems dealing with it!

That matrix might be a bit big for a lot of games interms of memory storage, and 4000x4000 using 16-bit values might be more suitable, taking up 32 MB of memory. If it's a character on the ground type of game, you would need the data points closer togeather, like on the order of 1 meter, so 4000x4000 would give an area of 4 Km by 4 Km. Might even use some pixel shading over points to help hide polygonal edges. A matrix of this size would leave a lot more memory for other game assets, compared to a 8000x8000 matrix.

It would be interesting also if a game engine could create curved geometry between points, to help hide any point to point angularity. It could do this, only a certain distance from the user to keep such processing overhead low.

With CELL's massive bandwidth and floating point power, I see games using height maps much more with this upcoming generation than the last generation. I would be hard pressed to even think how many PS2 games used a height map.

Another thing to consider is that depending on your games visiblity distance (weather effects limiting visiblity distance) and the perspective in the game world (low to the ground or high in the air makes a difference), you could even stream in different height maps depending on your location in the game world. I think World of Warcraft uses this technique to good effect, as moving from one region to another is seamless, as it streams in the next area as you are moving towards it.
 
Last edited by a moderator:
Back
Top