Acert93 said:
Definately, definately not 5 or 7 times as much as a bloody P4.
A bit of info on Cell's physic's abilities.
- Thread starter scatteh316
- Start date
Definately, definately not 5 or 7 times as much as a bloody P4.
I think it's safe to say that the SPEs are running SP code, SPE DP performance is 1/7th of SP performance, so there's no way a SPE would beat a P4 if it were DP.
A reason why the PPE VMX version is so low could be that they just run an plain C version, using doubles and no vector intrinsics.
Cheers
Gubbi
A reason why the PPE VMX version is so low could be that they just run an plain C version, using doubles and no vector intrinsics.
Cheers
Gubbi
It has to do with offline graphics because that's what this cloth simulation is for. It's to calculate animation frames for an offline renderer. Alias aren't setting out to create a realtime cloth simulation but a super-accurate lifelike system (presumably, seeing as cloth passing through characters is more acceptable in a game then a CGI render). AFAIK all modelling/rendering software like Max, Maya, Lightwave and Cinema use models constructed in double precision because they need the accuracy. Not sure. Laa-Yosh is the person to talk on this matter.Jawed said:
However, until we know for sure whether this article is talking about single or double precision, we can't make any observations of Cell's performance, especially realtime in games. It would be more representative to compare something like Novodex or Havok cloth simulation on Cell and P4 which are optimized for peak performance.
Kind of a serious question. Working with network facillities and procuring servers for clients taught me one thing: Prices are not linear to performance, and that rarer models are always exponentially more expensive. e.g. I can walk out with a 3GHz chip for $150, or a 3.6GHz for $400london-boy said:
Not that it cost the manufacturer any more to make one over the other. And moving up to chips that sell in lower volums the issue expounds. That said, the server market really is not that bad these days--pretty dirt cheap if you ask me (relative to what it does). Very tight market where support and many of the other things I noted above are just as important. I am sure CELL has a place in there somewhere, but having worked with the market to a small degree I wont be jumping to any conclusions. If I was running the above program and If, if, you can get a quad-proc P4ish system for the same price as CELL workstation, I would be hard pressed to switch platforms for the many reasons mentioned above (and not considering other processing factors). Without know the cost of a CELL workstation it is really hard to say much. Further we would have to consider other market forces (like other products from Intel/AMD and other companies, like Clearspeed) that will be in effect at the time of CELL Workstation availability or soon there after.I guess my head is more in the workstation/server market with this information. I am sure someone running Alias (Laa-Yosh??) could give us his thoughts--and they would be more definitive than anything any one of us could say. I am just basing what I am saying based on the very little and incomplete data we have and my meager experience with servers and workstations.
Link anyone for the retail cost of a CELL processor or CELL blade? Are they even available yet?
Like I mentioned earlier I am not sure this information tells us anything about game performance in that games are not designed to be accurate simulations but take a lot of short cuts with the goal of speed+believability. Running Novodex and Havok, as you say, would be very interesting. Obviously we would want one specialized for each environment (i.e. one specialized for a Pentium D, X2, CELL, etc). On the other hand since they are not in competing market places I am not sure that comparison tells us much.Shifty Geezer said:
It would be akin to running Windows on CELL and comparing it to a Pentium D. What does this really tell us? Probably not much in regards to the intended use and platform of the respective chips.
I totoally disagree. Alias is simply making a realistic/accurate cloth simulation program. it runs in real time or offline. Either way it's the same calculations required. I don't understand why you're trying to de-legitimize the usefulness of this test. They are maxing out the computational abilities of Cell, and as MANy people have already said, peak performance numbers simply aren't realistic.
Acert93 said:Hey, you brought up cost
It costs Intel $40 to make a chip. Further, there is very little additional cost for Intel to add a second core (see: multicore chips are not 2x as expensive, not even close). Further, with over 80% of the desktop market (a 200M unit/year market) Intel has an insane advantage in cost.
And this does translate into street price. Dual core Xeons, Opterons, P4s, X2s, etc are on the market. The MBs and Memory are readily available and CHEAP.
This can be purchased today--have it delivered tomorrow if you want. And it will work with your suite of tools and plug right into your network. Even more you can hire thousands of blokes who can effeciently work with it in a productive environment using the software that is out right now.
How much is a CELL workstation again? How fast of a processor? (You keep assuming there are faster CELL workstations readily available in the market right now. No point comparing a P4 of today with a CELL of tomorrow). How much memory? Availability? What about all the other stuff? Can I get overnight mail and on my door TOMORROW?
Well, personally, it'd cost be over $2000 to get 4 dual-core 3.2Ghz P4s on my door tomorrow
I don't know if that changes when you're in the "server market"..i'm guessing you have figures?I wasn't talking just about dollars when I talked about cost, though. I meant also the cost in terms of silicon, the cost in terms of space and room in a box to house 8 P4s, or 4 dual-cores. I'm approaching this a little, admittedly, from the perspective of what it'd take to get an alternative with similar performance into a box like PS3. That's a seperate market to that of the "server market", with different requirements, but it is also the one that should be of more interest on these boards, I'd have thought.
Yes, Alias is aiming at the market that wants farms of servers and soforth, so Cell should be put up against alternatives in that market (so at this point I'd invite you to provide a little more solid information on the costs of x number of P4s or dual-cores for that purpose, if you have different figures to that which I've provided). But on the other, we can from this see the value of Cell for a box like a games console when it comes to this kind of simulation and types like it - and the cost for alternatives to provide similar performance would be huge, if it was even physically possible to place enough of an alternative in the box.
On the other hand the CELL processor did a pretty good job in this simulation.Qroach said:
I think as availability, cost, and more benchmarks (on an entire array of tasks) come out compared to systems in the same market/price range become available we will be able to temper our expectations within a real framework.
Until then some will be dissappointed, some excited, some in the middle. But until we know more (a lot more) and given a context of its market these tests are only vague indications.
My hope is Intel sees that desktop chips needs a lot better performance in certain areas. But I don't think they work that way...
Acert93 said:
Running Maya you mean? Alias|Wavefront is the company that makes Maya, the famous 3D package.
Anyway, one Cell chip (and they're talking about the same chip that will be in PS3 but running slower, which strengthens my point), will never cost 5 times as much as a P4 3.2Ghz. That's just crazy.
Still, this demo proves very little. Cloth simulation (and whatever else is being performed by Alias' demo) is just one thing that a processor needs to do. The fact that Cell happens to be 5 times faster than a P4 at this particular program doesn't mean much to us.
I'm not trying to de-legitamize it! Just saying there's an unknown quotient here that people are filling in with their own guesswork. Everyone is assuming it's SP, with their reasons, but there's no official word so no-one actually knows. And as I've said, AFAIK offline rendering solutions work in DP. I'm not, nor have I ever said, that this is a DP solution. Only that it might be. Which no-one can disprove, unless they've got word from Alias, in which case please post it so we can better understand these results!Qroach said:
Any facts, figures, or statistics are only as meaningfu as the explanation behind them.
blakjedi said:
?
I'm going to guess that the reason you have a client-server link and do the rendering on client-side, in this instance, is because not everyone can have a cell-blade, or more realistically, a farm of blades on their desk. You'd connect to it as a client and do what you want.
Jawed
Legend
Except that at best the P4 in this comparison could be 10GFLOPs (see the graph I posted earlier).Heinrich4 said:
So 15x or more supposed theoretical.
5x actual simply demonstrates, to me, what a poor architecture Cell must be. Running at 1/3 efficiency? Laughable. Particularly for something that is so strongly suited to it. Supposedly.
Jawed
Easy enough to get. The variable is whether you are doing it yourself (pretty rare), going with a smaller company, or going with DELL and looking for a complete package of components+support. Needless to say if you are going for a 4 dual core setup (dual core Opterons Xeons are just hitting the market now), gonna stick 16GB of memory into on, get the proper MB, rack, power supply, redundant everything, etc... it is gonna cost you multiples more than $2,000. But there is more than one way to skin a cat (numerous individual white boxes with or without dual core processors, etc).Titanio said:Well, personally, it'd cost be over $2000 to get 4 dual-core 3.2Ghz P4s on my door tomorrow
But without knowing what a comparable CELL Blade costs it is hard to say. Are we comparing a server with 512MB of memory or 16GB? Is this a task where we need 4 dual cores, or will 2 dual cores (or 4 single cores) work, or even daisy chaining?
Limiting our discussion to this one benchmark (i.e. you plan on running this benchmark every day for the next 6 months) it would be hard to say what kind of configuration you would need without knowing more about the cost of the CELL server. Cost being inclusive of the things I mentioned and the ones you mention below. But needless to say you are going to be spending a bit more (hahaha) than $2,000 on an Intel/AMD configuration to compete with this. But without knowing how an Opteron or Xeon performs or the price point of CELL it would be hard to compare.
And of course in the real world you are not going to be running this one demo over and over.
And those are questions every IT guy has to ask. This is why CELL will find itself a place in the market somwhere (who knows how big, we don't know enough). This is very similar to the "power" question in the Itagaki thread--the answer is not the same for every company. There are many variables on "cost" and there is always the ultimate question of what are you doing, what does each platform offer in this regards, how big is your budget and short your time frame?
I would not even venture there because this benchmark doesn't tell us anything about gaming.
Besides the CELL chip being 2x as much silicone, the fact is the P4 IP is not available to the console market (i.e. look at the Xbox1 problem) and they are not really competing. Intel is much happier charging an arm and a leg in the PC Desktop market than trying to make the cheapest part possible for a console market. One of the reasons MS ditched Intel was due to FP performance to (not to mention IP issues and cost).
Really, this test is definately oriented toward content creation in the workstation/server market. It would be really hard to extrapolate that to the console market--outside of the fact this is the kind of task CELL is well suited for and it performs better than a single P4 (or even a Pentium D).
What that would tell us in the balance of a game (where a lot of different code needs to run) is unknown. Obviously it will chew through this type of code faster than a P4/Pentirum D.
It is quite odd, but the new consoles have gone maintream with the GPUs, they have gone alternative routes for the CPUs. I think this is because the overlap in GPU features/tasks is very similar while console processing and desktop processing have really forked.