The Next MacPro

BobTheDog

http://www.apple.com/macpro/design.html

Uses Xeon Nehalem processors which have all cores on a single die making cache access faster. Also the two processors are directly via a pipeline. Also the memory controller is built in to the processor.

Up to 1.9x the speed of the previous 8 core MacPro if you believe apple.

Pretty fast machine this one.

BobTheDog · Posted: Tue Mar 24, 2009 10:11 am Post subject:

Well the fastest machine that Geekbench have ever tested with a score of 17665.

http://browse.geekbench.ca/geekbench2/view/115605

To put it in perspective my 4 core MacPro 3Ghz scores 5508!

v-un-v · Posted: Tue Mar 24, 2009 10:26 am Post subject:

Are these speeds of any computer- including those made by other companies, or just Mac's?

cbm · Joined: Oct 25, 2005 Posts: 381 Location: San Francisco

BobTheDog · Posted: Wed Mar 25, 2009 1:41 am Post subject:

Sorry I should have said workstation not computer!.

BobTheDog · Posted: Wed Mar 25, 2009 1:42 am Post subject:

Inventor · Posted: Wed Mar 25, 2009 4:44 am Post subject:

Honestly I don't understand this need for speed. I've been saying for at least 10 years that computers have gotten fast enough for what we do with them and what we really need is a better compromise between speed and power with emphasis on novel user interfaces and custom coprocessors.

That's why I like my little netbook so much. It may only be around 1/3 the speed of an overheated bloatware battery hog notebook, but that's plenty of compute power for even CPU-intensive tasks like ChucK programs. And it only cost $300 US, cheap enough even for my budget.

Instead of making super-duper-stupor processors faster, us consumers should be demanding innovative coprocessors like music chips, neural net chips, solid-state disks, graphics chips, and stuff like that. How about a chip that's hard-wired to run a specific programming language or an operating system. If they would make a hardware operating system, then even Microsoft's bloatware would run properly.

Not to mention the lack of touchscreens, laser projectors, voice interfaces, and any intelligent use of the camera as a communication device. Let's get with it people: stop the CPU wars and let's do something innovative for a change!

Oh, sorry, that turned into a rant! Well, it's been irritating me for a long time now...

seraph · Posted: Wed Mar 25, 2009 5:31 am Post subject:

Inventor · Posted: Wed Mar 25, 2009 5:35 am Post subject:

Yes, I remember my first computer, a TRS-80 Model 3 had a hardware operating system. With Flash these days, the hardware OS processor would be upgradeable too.

BobTheDog · Posted: Wed Mar 25, 2009 5:50 am Post subject:

Well it depends what sort of software you are running.

For instance if you are doing lots of rendering or calculations, multi user servers etc you need as much power as you can get hold of and some more.

If you are just running most end user apps now then you are correct you don't need this sort of power.

My MacPro hits 100% cpu on all processors a fair bit but it doesn't really bother me that much, when the warranty runs out I will get one of these new ones, probably the lowest clock speed one.

Inventor · Posted: Wed Mar 25, 2009 6:51 am Post subject:

BobTheDog · Posted: Wed Mar 25, 2009 7:12 am Post subject:

We are not hitting a wall on processing speed at all, multi cores and processors give us increased performance, the more cores or processors we have the better. Even single core performance is still improving by leaps and bounds.

Specialised processors, say a rendering processor the idea still stands:

Lets say we have a decent coprocessor for rendering, something like this: http://www.nvidia.com/object/product_tesla_s1070_us.html

Now lets say to render a HD frame it takes 5 seconds (very low just for an example).

Now I have a 5 minute clip at 25 frames per second so I have 7500 frames so this render is going to take 450 minutes.

Now if I expand my system to have 512 of these coprocessors it take less than a minute!

v-un-v · Posted: Wed Mar 25, 2009 7:25 am Post subject:

Inventor · Posted: Wed Mar 25, 2009 8:08 am Post subject:

Inventor · Posted: Wed Mar 25, 2009 8:18 am Post subject:

BobTheDog · Posted: Wed Mar 25, 2009 8:22 am Post subject:

But we are not there yet. Also this old viewpoint certainly ignores multi core/processor systems.

BobTheDog · Posted: Wed Mar 25, 2009 8:26 am Post subject:

Have a look at http://www.nas.nasa.gov/Resources/Systems/columbia.html this was a SGI machine built to provide simulations of large systems (like space shuttles) for NASA.

Note the clock speeds are low but the power is high.

Inventor · Posted: Wed Mar 25, 2009 9:00 am Post subject:

Well, yes, for tasks that parallelize well, supercomputing clusters and multicores and such are great. But not all software parallelizes well. For example, code that requires information from a previous computation must be run serially.

I was reminded of this when I created a program in POV-Ray which had an AI algorithm that searched out the space of possible chainmaille weaves. The other chainmaillers and I wanted to use the multiprocessor version of POV-Ray to speed it up, but we soon found that only the renders were multiprocessor-capable. The actual AI code which took up most of the CPU time had to be run on a single processor. Now, that code happens to be parallelizable but the application that supported it was not set up that way.

So going parallel is great if your application supports it and you take the time to write your code in parallel form, but a lot of software doesn't work that way. Maybe that's largely just a lag in progress, I dunno.

BobTheDog · Posted: Wed Mar 25, 2009 9:11 am Post subject:

Most processes can be run in a parallel way, its just that most pc/mac developers haven't done it much in the past.

Also just because some parts of code may be serial it doesn't mean the whole system will be.

Also algorithms that are seen as serial (a good example os the Sieve of Eratosthenes for finding prime numbers) can be made parallel with a bit of thought.

Also with serial processes as long as you have decent core to core data transmission each core can be operating on the result of the previous cores calculation while the previous core is working on the next calculation.

The Tesla processor above works like this, 960 simple cores and each can be used to calculate part of the algorithm to pass onto the next core.

I think you are correct with your lag in progress idea.

Antimon · Posted: Wed Mar 25, 2009 9:13 am Post subject:

BobTheDog · Posted: Wed Mar 25, 2009 9:18 am Post subject:

Visual Studio 2008 is like this as well, this doesn't run well on any PC!

Inventor · Posted: Wed Mar 25, 2009 9:39 am Post subject:

Yes, I suppose that as developers write parallelized code in more applications the multicore solution will help us continue to keep up with Moore's Law. Still, there are many applications where this has it's limits.

As long as battery technology continues to evolve slowly notebooks will only be able to have so much computing power. Also heat management gets to be a problem in desktop systems with lots of cores. I mean, we're certainly not expecting tomorrow's desktops to have 256 cores, it would heat up your house!

So what will happen when each core hits it's limit and the power density gets to be too much for even liquid cooling to handle in a desktop? We'll hit the wall. It will take either new tech or specialized processors or who knows what to keep progressing.

BobTheDog · Posted: Wed Mar 25, 2009 9:42 am Post subject:

nobody · Joined: Mar 09, 2008 Posts: 1687 Location: Not here

I'm 100% with you on this one. I don't care how fast it is. Can it work for me?

nobody · Joined: Mar 09, 2008 Posts: 1687 Location: Not here