Is Intel too far ahead of its time? - Page 2

Last message on previous page:

18th August 2008, hardly an ancient article by any means. Maybe one or two revisions have come out since then, but I doubt they would have anywhere near the customisation of a CPU based encoder in this short period of time. What does it have to do with Adobe? Uhh, nothing, it has everything to do with ENCODING. As far as Photoshop goes, as I mentioned earlier GPU acceleration is only available on certain functions (canvas rotating and zooming): http://www.tgdaily.com/content/view/39433/140/1/1/

So maybe having a CPU do the few GPU accelerated functions are foolish (not that anyone will override GPU acceleration in any case), but what about everything else? I'm sure theres a WHOLE LOT MORE to Photoshop than just image rotation and zoom, but whatever, using a CPU is 'foolish' apparently... LOL!

I swear some of you guys see a few marketing slides on CUDA and think the CPU is now redundant, well yeah if you wanna encode in baseline quality which looks like absolute **** or ONLY use zoom and rotation in Photoshop, then go ahead, but like I said GPU based acceleration of common desktop apps has a LONG way to go, so in the meantime I'd say the gains Nehalem brings are VERY relevant.

JD, this is just an observation of mine, but it appears everytime you start a thread, you've already made up your mind (just thinking back to the 'is AMD as fast as Intel in gaming' thread) and the rest of your replies thereafter involve trying to 'convert' people into your frame of thinking, rather than discussion of the actual topic.

Mathos,

On RISC, in PC, you answered the example yourself with the AMDs. RISC could not be succesfully implemented on a commercial basis, and was not. It was hybridized, in such a manner that what was released was not RISC

On multithreaded softeware, I dont know how you arrived at the conclusion that the reason why software developers have not been writing code optimised for quadcore is because "they have been waitng for Intel to go native quad", but the evidence says otherwise.
1) In terms of execution, there is no functional difference between MCM and native, and as such there no reason to wait for a native multi anything, unless you beleive in AMDs "multicore for dummies", which AMD itself resoundingly proved fallacious
2) Need. There is software which can benifit from multithreading, and software which doesnt. Most software doesnt. Word processors, web browsers, most localized commercial databases, hardware SDKs etc, dont. Transcoding, rendering and higher function theoretical math/physics can. Virtualization does.

And what the evidense says is that current single thread programming techniques/methods do not apply well to writing multithread. New methods provide faster results, but are not optimal in terms of maximizing efficiency. And the rest is just reverse logic, assuming that because there are quadcores, there must be quadcore optimized software.


http://www.tomshardware.com/forum/ [...] -threading
I suggest you visit this thread, and read the article that started it, and the posts from actual code writers

http://money.cnn.com/2008/08/13/te [...] /index.htm

Heres a few quotes

Heres a nice one that points out new methodology to ease the writing of multithread, unfortunately, as Mrak points out there are some other potential problems. What Mark didnt mention is the "overhead" the cost of using threads themselves as commuincation conduits...which consumes resourses (cores) to latterally translate data, with no forward progress.

Nark follows up with this, but presnts no proof. Furthermore, while there is incentive to utilize the extra cores in certain genres of games for particle physics, that is a long way off, as the contiuning trend in development of the higher complexity games has been divergence in developement, with one company creating the engine and selling it to many 'game houses'. This trend will continue as the smaller companies can afford niether the time nor cash to dedicate to developing new engines. Regardless, one only need look at development time to see the trend is to longer times, meaning it will still be years before any significant number of engines optimized for quadcore are available.

And heres another good one that demonstrates the real 'benefits'

That was a good thread, you should read it.

@turpit

Thats just the thing, I'm mainly going by what I've seen as new processors have came out, compared to when games and programs that utilize their abilities come out. Most of the time, for example with games that were able to run 2 threads, everyone has waited for Intel's processors to come out, with a few stragglers just going ahead and not waiting. Going by memory I didn't all that many two threaded games/apps at all till after C2d came out, even though Athlon X2 had been out for almost 2 years prior, as well as the Pentium D's. Going by that it would appear that they all waited for Intel's native dual core processor. My other guess is it has to do with the way the Cache systems work on the processors. Intel's MCM has two large L2's which reduces memory hits, and the need to go to FSB. When they are switching to Nehalem things will change, due to the smaller L2's per core, but unified L3.

There are games out now that are multitread capable. Anything based of Unreal Engine 3 is, which seems to be a lot of the fps genre games from smaller publishers lately. Mass effect for example. The one thing I've noticed though. Even at present most programs and games aren't going well over 2.7-3 threads. Which is why in some benchmarks the Phenom X'3 perform near the same as the X4s. I believe once Nehalem comes out, that things are going to change and improve for anyone running a multicore processor regardless of brand, especially after about 1 year.

As far as programming goes, yeah, I'd assume it would be difficult with current tools. But like the one guy said if you're using the proper tools to do the coding it won't be a problem at all. People don't seem to remember that the server space has had parallel or multithreaded apps for a long time now.

Was Pentium Pro ahead of its time? Pentium Pro/i7 - feels like the same thing. Some with deep pockets or server type applications will jump in right away, the rest probably wait for the Pentium III mass-market iteration (at least from what I'm seeing).

GPGPUs... we're talking about their infancy. Right now, Nvidia runs CUDA, AMD runs CAL, S3 runs their as-yet-unreleased API - it's like coding for Itanium but you have 3 incompatible Itanium vendors. No wonder there's such limited serious GPGPU coding.

CUDA could very well go the way of Itanic; minority share APIs almost certainly will. It'll come to how much can be salvaged when DirectX 11 or some GPGPU standard forms.

I've been running some CUDA on Vista - it feels unpolished. You can forget about playing games and "multitasking" the CUDA app, despite every CUDA app being inherently full of parallel code. You can probably say goodbye to Aero too - very choppy with CUDA active. This stuff is certainly not ready for the mass market; it's only surviving because it piggybacks off GPUs mainly sold for other purposes.

Oh come on Mark, you didnt say core, you said overhead, which can easily translate to cache or transfer. If you really meant core(s), I can buy that. You havent demonstrated any tendencies (that Ive seen) to lie, backtrack or ignore like some of our other members, so if you tell me thats what you meant, Im good with it, but as I read it, it was both vague and broad, and thus not particluarly meaningfull and certainly not specific to consumption of a core.

And what what executes the code that controls the queue? Its not the cache. You can either dedicate a core to that task, or execute it sequentially within a thread. Either way, it takes time and resourses. If its executed flawlessly in either method, then the net sum difference between methods should be zero, theoretically.

Regardless, with vast amounts of data/messages, a dedicated core is literally demanded, especially in fine threading if you are going to run hundreds of loops,which is what will be demanded with high realism particle physics. If you were to try and run that amount of trffic within a thread, it would bring the code execution to a screeching halt as it made all those calls and waited for the replies.

I agree, but there in lies the problem itself. Even with newer methods, the complexity still increases, and with the increase in complexity comes the attendant increased risk of conflicts. You can slice the processes/loops any way you want. You can code the data calls/exchanges anyway you want. It doesnt matter. As you improve realism the data volume and interdependance are still increasing (unless you stop development and use the same engine for every new title), so you are increasing the opportunites for conflicts. Anyway you do it, you cant get rid of the base data conditioning wth out moving in the wrong direction....reducing the level of realism. The graphics side is nothing, but the physics side is a differnet story, and once you start cutting that, I dont care how photo realistic your graphics are, you are cutting content.

Well crikies, I dont remember now. What was the quote?

Mark, I agree with you, but simply because you and I agree that this is the way it "should" be, doesnt mean thats the way it can be. Going "all the way" would be easy if all they were doing was programing extra ghosts in 'Pac Man'. Thats not where the industry is going. And thats not where the industry is going because the consumer demand within the market segement we're talking about, that segment that actually has some use for multicore, is for increased realism. Which means only one thing: the increased demand is for particle physics.

IRT particle physics, Its not even possible to 'go all the way' right now. Current and upcoming hardware couldnt even come close to handling high volume realistic particle physics. Even attempting to minimze the problem by moving to larger objects doesnt come close to elimintaing a fraction of the variables. Mass, density, composition, energy state, velocity, reactivity.....the list goes on. None of these properties can be ignored to go 'all the way', and have to be modeled mathmatically/in code at the molecular level.

One company or a dozen, it doesnt matter. Any company that developes an engine must deal with the problems of multicore. ID took at least 2 years to deveople Tech 5, and that barely scratches the surface of particle physics. Cryteks last engine for crysis, touted as a major jump in physics failed to deliver the level of physics demonstrated in their trailers, meaing either that they had to give up something to speed release (and it was late as it was) or they simple had problems they couldnt overcome and had to cutback. As for their next engine, it is now 2008 and their next engine in developement now. It wont be out until 2012. 4 years.

http://www.joystiq.com/2008/08/19/ [...] -due-2012/

You know, Its great that so many people seem to think that there are going to be quad optimizing games flooding the market, but here we are, just about 2 years after C2Q, and you can still count the number or quad optimized titles on your fingers. This miraculous flood of 'quad' everything software should have hit the market by now, or at least be showing up as I type this, but it hasent. Why? 3 possible reasons come immediatly to mind: it isnt in demand, it isnt needed to achieve accepatble results, or its difficult to do. Pick one or any combination. Eitherway, if a complex game began developement in november 2006, when the QX6700 was introduced, it should be here anytime.....unless you suscribe to mathos theory that the developers were waiting for nehelem. So I guess we'll wait and see how many ouad titles pop up this xmas. Who knows, the market may flood and prove you right, but I wouldnt put any money on that, at any odds.

With CUDA its not that its easy which C is pretty easy. Its that the software companies look at it this way. 1. tey will have to rewrite everything to CUDA. And 2. that will cost money. Do they want to spend money? Not really. Unless it becomes a more dominant market they wont because it wont profit them.

Now once it does become a bigger more dominant market you bett you ass every last company will write to support both CPUs and GPGPUs. That way they can still support both and profit. Thats all they care about unfortunately.

But they are in the crap bigtime for failing video card chipsets...

Taking a look at NVISION this year, it wasnt about nVidia , but about the gpu and its place in the overall market. CUDA is one thing, gpus another. nVidia is flexable, and having CUDA fail would be a setback, but having gpus doing gpgpu type things is still a win for them, and thats what they promoted during NVISION. Its obvious theyre in it for the long haul, regardless of how it plays out.
I agree totally with Turpit here. One way or another, itll be several years before we see most games coming out using a LIMITED amount of MT. UT3 is one such game, and it shows very little in performance anyways, which hopefully is more of a fluke than an eventuality.

I did state ages ago that the head of Nvidia should shut up and get on doing what they did best...

Oh no,, he slagged of Intel saying CPU's were the thing of the past, then all crap flew loose when the video cards started packing up in laptops and desktops.. Now hes in the sh!t

Now I did say that nowadays you have to work with Intel, not against them..
AMD tried it and they are suffering bigtime... You can just tell Intel wanted ATI as there was talks of a take over of Nvidia, but that fell through..

Wether they like it or not thats the way it is..
Intel are just to big and write the rule book...
Love or hate Intel, they have created PCI, PCI Express, AGP, USB 1 & 2 and soon 3...

Even if you dont like Microsoft you make the more money selling Windows against the linux based systems... Fact..
Plus every one wants Windows with Intel inside...

If I was AMD I would sell of ATI, but keep the chipset division and work on what it does best make processors...
But AMD is self imploading by greedy directors removing cash (allegedly)

They just did, gave him to the Arabs.