Welcome to the MacNN Forums.

besson3c · Sep 3, 2009, 11:26 AM

I'm toying with getting a Macbook Pro right now, but if I do one option I'm unsure about is the integrated vs. standalone video card option. I'm wondering if you guys have any advice.

Specifically, I don't really need the video card right now. I don't do any intensive video work or gaming, and both cards are future proof as far as being OpenCL supported. However, I'm also a fan of the hardware I buy as a whole being future proof. What I want to know is whether I can put off this decision? Is there simply a free slot where I could plug in a video card of my choosing using a commodity video card built for a notebook, or are there motherboard alterations for this card?

Big Mac · Sep 3, 2009, 12:20 PM

No free slot, sorry besson. As a geek i'd assume you'll only be happy with a dedicated GPU model.

besson3c · Sep 3, 2009, 12:28 PM

Not surprising, but thanks for the confirmation. Care to offer a perspective other than mine on the pros and cons of the dedicated GPU? I want to make sure I've thought of everything, I'm kind of the opposite of the impulsive shopper in agonizing over details, and I especially want to consider whether or not my choice is future proof as best as I can future proof myself within reason right now.

turtle777 · Sep 3, 2009, 01:09 PM

Ok, so just to be clear, there are only two options:

a) integrated only
b) integrated + discrete

If if you get option b), you can never utilize BOTH GPUs at the same time.
You'd only want the two GPUs if there are apps that would need the optimum result out of the either GPUs.

Since you have to restart the Mac to switch the GPU (AFAIK), you would use a different GPU depending for different apps or games. This just doesn't sound like something you would need.

To be honest, you sound like someone who would be ok with just the integrated GPU.

-t

Big Mac · Sep 3, 2009, 01:33 PM

You can switch by logging in and out, AFAIK.

Simon · Sep 3, 2009, 03:12 PM

Originally Posted by turtle777

If if you get option b), you can never utilize BOTH GPUs at the same time.

That is partially correct. You, as the user, have to chose. But OpenGL doesn't. Even when you chose to run the OS X window server on the 9400M, OpenCL can still access the 9600M at any time and vice versa. OpenCL can actually use the CPU, the integrated GPU, and the dedicated GPU all at the same time.

So if you believe OpenGL is the next big thing and you're worried about future proofing, you should clearly opt for the 9600M model.

besson3c · Sep 3, 2009, 03:13 PM

So, after rebooting or logging back in using the faster GPU, why would you ever want to switch back to the slower, integrated one?

Simon · Sep 3, 2009, 03:15 PM

For starters because of an extra hour of battery life...

besson3c · Sep 3, 2009, 03:20 PM

Originally Posted by Simon

That is partially correct. You, as the user, have to chose. But OpenGL doesn't. Even when you chose to run the OS X window server on the 9400M, OpenCL can still access the 9600M at any time and vice versa. OpenCL can actually use the CPU, the integrated GPU, and the dedicated GPU all at the same time.

So if you believe OpenGL is the next big thing and you're worried about future proofing, you should clearly opt for the 9600M model.

This makes sense, otherwise it would have made more sense to just provide the other video card as a BTO option and in doing so reducing the price.

So, it really comes down to how many apps make use of OpenCL, huh, and whether I would be running an app that would leverage it? It's kind of hard justifying this right now especially since there are no apps that use OpenCL now, and most of the work I do is in development and not content creation or gaming... What do you think? Sound rationale?

besson3c · Sep 3, 2009, 03:25 PM

Integrated = better battery life?

Simon · Sep 3, 2009, 03:31 PM

Originally Posted by besson3c

So, it really comes down to how many apps make use of OpenCL, huh, and whether I would be running an app that would leverage it? It's kind of hard justifying this right now especially since there are no apps that use OpenCL now, and most of the work I do is in development and not content creation or gaming... What do you think? Sound rationale?

I'd say you have it about right.

This pretty much boils down to a gamble on future proofing. If OpenCL becomes popular and you keep this MBP for 2+ years you'll want to have the 9600M. If OpenCL doesn't get adopted or you plan on swapping the MBP sooner, you could try saving some money.

Simon · Sep 3, 2009, 03:34 PM

Originally Posted by besson3c

Integrated = better battery life?

Yeah. Even Apple advertised the difference of about an hour. I've seen that number confirmed by actual tests. Depending on what you do it's a bit more or less, but roughly you're looking at an hour more with the 9400M.

turtle777 · Sep 3, 2009, 03:58 PM

Originally Posted by besson3c

Integrated = better battery life?

Apparently, but only if you don't task the discrete GPU with OpenCL...

-t

mduell · Sep 3, 2009, 05:06 PM

The performance difference between the two GPUs for games is considerable, but may not be so significant for other types of apps even when they are eventually using OpenCL due to Amdahl's law.

chabig · Sep 3, 2009, 05:32 PM

The integrated graphics is what you'd get if the government built video chips. But due to the wonders of capitalism, you can choose the discrete chip if you want to.

Simon · Sep 4, 2009, 02:03 AM

Originally Posted by chabig

The integrated graphics is what you'd get if the government built video chips. But due to the wonders of capitalism, you can choose the discrete chip if you want to.

Can we keep the politics nonsense out of a technical thread, please?

Simon · Sep 4, 2009, 02:24 AM

As a teaser to show off potential, here are some numbers from OpenCL Benchmark. Note how well even the 9400M does.

GeForce 9600M GT: 2.805 seconds
GeForce 9400M: 3.081 seconds
Intel Core 2 Duo @ 2.40GHz: 15.459 seconds

besson3c · Sep 4, 2009, 07:28 AM

Thanks for that, Simon. It looks like mduell's theory is correct, the additional card does not provide a world of difference.

Simon · Sep 4, 2009, 08:12 AM

Originally Posted by besson3c

Thanks for that, Simon. It looks like mduell's theory is correct, the additional card does not provide a world of difference.

Hmm, I think you might have misunderstood. Those numbers are for each of the processors alone. They do not reflect how much you gain from combining the CPU and both GPUs to solve a problem. Each has been tested by itself. The numbers reflect how much better the 9400/9600 are at solving a specific problem compared to the C2D.

Amdahl tells you that your performance increase from parallelization will eventually saturate (weakest link in the chain). But Amdahl doesn't tell you that you shouldn't make the effort to distribute load. At this stage of OpenCL I don't think Amdahl is an issue because we don't have a large number of processing nodes with highly different parallel performance. We have two GPUs that perform similarly (in this example) and a dual-core or quad-core CPU that doesn't. Ideally OpenCL will then uses those two GPUs to solve a subset of problems that the GPUs are good at while it frees up the CPU cores for other things. Note also, that OpenCL is one benchmark that tests something the GPUs are very good at. You could easily find other examples where the performance difference between them is negligible or where the CPU is better than the GPU.

turtle777 · Sep 4, 2009, 08:19 AM

So, Simon, are you saying that right now, OpenCL wouldn't be able to efficiently use BOTH GPUs concurrently, but that in the future, that may be possible ?

-t

Simon · 08:42 AM

Not really. I was just pointing out that's not the way that benchmark worked.

OpenCL can definitely use both GPUs right now. But the question is just how efficient it does that. And that depends very much on the specific task you're looking at. On a problem like the one above where both GPUs perform similarly, the performance increase could be significant.

Right now it's IMHO simply too early to judge OpenCL's impact. But it definitely has the potential to accelerate certain operations by up to an order of magnitude. OTOH this does not mean a user application will see the same speed boost.

The example above I think shows three things:
- there are certain operations that can be handled much better by the GPU
- if your OpenCL code can offload such operations to the GPU you should see significant performance increases
- if your OpenCL code can offload to both GPUs (on systems that have two) and balance well, the increase will be even bigger
Those last ifs are the problem. Right now we simply don't know what we'll get.

I'm guessing OpenCL will be used by most developers to offload certain types of operations to the GPU or to both GPUs. But I doubt it will used to run the same stuff on both the GPU and the CPU at the same time.

We're looking at a very interesting time ahead as we'll observe how developers will attempt to make use of OpenCL. While the idea itself is all but new, we now have a common framework for this type of optimization rather than having to rely on specific hardware (CUDA for example) being present. This should help adoption and make it a much more interesting option for developers. It will take some time, but we should be seeing some pretty interesting developments in this area.

OreoCookie · Sep 4, 2009, 08:36 AM

Originally Posted by Simon

Yeah. Even Apple advertised the difference of about an hour. I've seen that number confirmed by actual tests. Depending on what you do it's a bit more or less, but roughly you're looking at an hour more with the 9400M.

Ditto, my office mate has just gotten a 13" ProBook recently and I'm really jealous he gets easily 5+ hours of battery life (6~7 hours depending on the use and screen brightness). That'd be very appealing, I could work for the majority of a business trip rather than only 2 hours.

Simon · Sep 4, 2009, 08:45 AM

Originally Posted by OreoCookie

Ditto, my office mate has just gotten a 13" ProBook recently and I'm really jealous he gets easily 5+ hours of battery life (6~7 hours depending on the use and screen brightness). That'd be very appealing, I could work for the majority of a business trip rather than only 2 hours.

I hear you!

To me that's the most important reason to upgrade to a new unibody. My current 15" MBP is still doing fine and performance-wise/feature-wise the new models don't offer a whole lot of improvement. But the prospect of going from 3.5h to 7h of battery life is huge. That alone makes it worth it IMHO.

besson3c · Sep 4, 2009, 09:54 AM

Simon: it will indeed be interesting to see its impact over time. I'm a little unclear as to what sorts of real world tasks the GPU processing will help with. I mean, we know that GPUs are good at shading, handling vectors, curves, stuff like that, but how exactly would this impact potentially, say Adobe Photoshop or even OS X itself? Any ideas/speculation?

turtle777 · Sep 4, 2009, 10:20 AM

Originally Posted by besson3c

I'm a little unclear as to what sorts of real world tasks the GPU processing will help with. I mean, we know that GPUs are good at shading, handling vectors, curves, stuff like that, but how exactly would this impact potentially, say Adobe Photoshop or even OS X itself? Any ideas/speculation?

Uhm you might wanna read up on OpenCL.

http://arstechnica.com/apple/reviews...-x-10-6.ars/14

What exactly is your question ?

-t

Simon · Sep 4, 2009, 10:39 AM

Originally Posted by besson3c

Simon: it will indeed be interesting to see its impact over time. I'm a little unclear as to what sorts of real world tasks the GPU processing will help with. I mean, we know that GPUs are good at shading, handling vectors, curves, stuff like that, but how exactly would this impact potentially, say Adobe Photoshop or even OS X itself? Any ideas/speculation?

I don't know if Adobe has made any use of SSE so far, but if they did, they can do many of the same things (like adding and multiplying vectors) much easier and faster with OpenCL. Easier because of the beautiful syntax and faster because not only can they use SSE on the CPU through OpenCL, but they can also do it natively on the GPU possibly in parallel without extra coding efforts.

Simon · Sep 16, 2009, 04:43 AM

As an update, here's another benchmark to show off GC/OpenCL performance.

The Application MovieGate was used on a MP (quad 2.66 GHz, GeForce 8800GT) for MPEG2 encoding/decoding. Using the exact same hardware GC/OpenCL on SL show 50% better performance.

Snow Leopard
150 frame/s for encoding in MPEG-2
70% CPU load for decoding
130% CPU load for MPEG-2 encoding (ffmpeg)

Leopard
104 frame/s for encoding in MPEG-2
165% CPU load for decoding
100% CPU load for MPEG-2 encoding (ffmpeg)

Hardmac.com : Le "Macbidouille" in English - Positive Effects of Grand Central and Open CL