13" 320M: Some questions, clarification

[Komodo Rogue]

What's up guys. I know there's already been some discussion about the new GPU in the 13" model but I'm still a little confused about some things, the shared memory in particular. My knowledge of hardware is pretty limited so any insight would be appreciated.

From some tech site:

The NVIDIA GeForce 320M is an integrated chipset graphics card for Core 2 Duo based laptops and successor of the GeForce 9400M. It does not feature dedicated graphics memory but uses the systems main memory instead (shared memory). Therefore, the performance is not as good as similar cards with dedicated graphics RAM.

Like most of you I was bummed when I read that the 13" used shared memory...

But to be honest, I'm not sure where exactly the bottleneck in performance comes from: is it slower for the GPU to access the system's RAM then to have it's own memory right there (if so, why? I mean isn't the DDR3 RAM pretty fast?), or is it because the main memory already has stuff loaded on it like the OS etc so shoving the GPU's needs onto an already taxed memory just slows things down (but this confuses me too because it challenges my understand of memory... I thought it just kinda held whatever data and the GPU or CPU accessed and processed that data as needed. Or is that the problem right there, that the GPU and CPU are accessing the same memory bank but there's a bottleneck because the GPU and the CPU need to take turns when accessing the memory?)? Or is it something else entirely?

Another thing that I don't understand is why the GPU would be limited to 256mb if it's using the system's memory. Would it be possible to just manually allocate more memory to the GPU, or does the GPU's architecture limit it from being able to use more than 256?

Sorry for all the questions. I've been browsing around trying to get a better understanding of this, but most tech sites are way over my head.

 

[kny3twalker]

DDR3 is slow in the GPU world, and accessing system memory is not as fast as accessing the dedicated memory on a dedicated GPU. And the minimum allocated is 256 MB, it can allocate more.

But really the 320 is not a high end graphics card, so the shared memory is a decent solution, considering the benefits

I would not be disappointed by this; the 9400M in mine is as fast as the dedicated graphics in the thinkpad T400 I was considering at the time, so I would think the 320 is comparable to other low end dedicated graphics.

 

[therealseebs]

The big problem:

The total bandwidth to memory available has to be split between the video card and the CPU. Also, dedicated video card memory is usually faster.

But basically, imagine that you have a computer under heavy load. With discrete graphics using its own memory, you have:

15GB/sec from CPU to and from main memory
20GB/sec from GPU to and from main memory

These occur simultaneously, and neither slows the other down.

With shared memory, you have:

15GB/sec from CPU and GPU put together to and from main memory.

You can see how this would be slower.

 

[Pikkoz]

I'm curious if it would help in gaming performance and how much , to change the regular ddr3 1066mhz to a higher clocked ddr3 1333mhz (maybe with a lower cas).
Is there anyone tech savvy enough clear this question ?

 

[Cali3350]

I do wish there was a proper review of the 320M out there. I am curious how it performs in real world with the 8600M GT and a 4650M.

 

[Cali3350]

I'm curious if it would help in gaming performance and how much , to change the regular ddr3 1066mhz to a higher clocked ddr3 1333mhz (maybe with a lower cas).
Is there anyone tech savvy enough clear this question ?

Not enough to matter. Bandwidth helps, and all bandwidth is good, but its still low enough that memory bottlenecks will impact performance to such a high point that the small difference is too minor to matter.

 

[broncopde]

DDR3 is slow in the GPU world, and accessing system memory is not as fast as accessing the dedicated memory on a dedicated GPU. And the minimum allocated is 256 MB, it can allocate more.

Can someone else (or a few someones) verify this bold statement? I have always thought that the 256 MB (in this card's case) is the max that the integrated graphics could allocate, but I'd love to be wrong about that.

 

[mikeo007]

Can someone else (or a few someones) verify this bold statement? I have always thought that the 256 MB (in this card's case) is the max that the integrated graphics could allocate, but I'd love to be wrong about that.

I know windows will allocate more RAM if necessary. It will even augment the VRAM of dedicated graphics cards if it's needed. I would assume mac OSX would also do this in some way.

 

[ipsg]

What's up guys. I know there's already been some discussion about the new GPU in the 13" model but I'm still a little confused about some things, the shared memory in particular. My knowledge of hardware is pretty limited so any insight would be appreciated.

From some tech site:


Like most of you I was bummed when I read that the 13" used shared memory...

But to be honest, I'm not sure where exactly the bottleneck in performance comes from: is it slower for the GPU to access the system's RAM then to have it's own memory right there (if so, why? I mean isn't the DDR3 RAM pretty fast?), or is it because the main memory already has stuff loaded on it like the OS etc so shoving the GPU's needs onto an already taxed memory just slows things down (but this confuses me too because it challenges my understand of memory... I thought it just kinda held whatever data and the GPU or CPU accessed and processed that data as needed. Or is that the problem right there, that the GPU and CPU are accessing the same memory bank but there's a bottleneck because the GPU and the CPU need to take turns when accessing the memory?)? Or is it something else entirely?

Another thing that I don't understand is why the GPU would be limited to 256mb if it's using the system's memory. Would it be possible to just manually allocate more memory to the GPU, or does the GPU's architecture limit it from being able to use more than 256?

Sorry for all the questions. I've been browsing around trying to get a better understanding of this, but most tech sites are way over my head.

why in the world would you care about the amount of video memory?
256M is more than enough even if you want to hook up a 30'