MXM versus BGA GPUs

Hey everyone, I spent a while thinking about MXM and BGA graphics set-ups, and I have some pointers...

First, MXM is an add-on type card, with its own pin-out and port, and the port is usually used on notebooks and other embedded systems (e.g. Drive PX 2). The GPU itself, VRM, VRAM, chokes, etc. are all on the board, and it's self contained. However, isn't the GPU still soldered onto the MXM add-on board? So, in other words, it's still 'BGA', but now on an extra add-on board.

On the other hand, traditional BGA laptops (most of what's released today) have the GPU and accompanying components directly placed on the motherboard.

So, theoretically, the 'traditional' BGA set-up ought to have better performance compared to MXM because of sheer proximity to the CPU , barring all the other advantages of MXM over BGA, like upgradeability, ease of repair, etc. However, power users here (you know who you are ) claim the opposite...

I'd like to understand your reasoning behind MXM > BGA in terms of performance. How is an add-on card faster than a built-in GPU?

Note that I don't support BGA in any way, before you fire a volley at me...

 

The latency due to longer 'traces' and connects with a MXM card compared to a BGA type setup is still negligible. So it doesn't gain any advantages compared to MXM only disadvantages (can't be upgraded unless you replace the whole mainboard, repair is more difficult and so on).

 

In recent times finding a chip in order to replace your burned one is more and more difficult. Not to mention that there are parts of the world where people don't repair (and repair shops are scarce), but just throw away and get the new model, hence why BGAs are called disposable. So getting an MXM module instead of a whole motherboard (if you manage to find one) makes more sense. You can always try to upgrade as well.

 

I think you really hit the nail on the head here. Not only what you said is true, but you also take the fact that MXM GPUs often have more VRAM than BGA GPUs, and thus are more futureproofed even if you don't want to upgrade your GPU - you can still take advantage of the extra VRAM in the future.

BGA is terrible.

 

Death to BGA!

 

NBR likes to derail and/or misunderstand posts... I've already mentioned MXM's advantages over BGA in the OP.

I was asking how an MXM GPU could be faster than a BGA GPU when there's nothing that inherently improves performance of an MXM GPU.

We're not talking about upgradeability or repair or whatever here.

 

And we told you, BGA is NOT faster than MXM GPU, and in the future - more VRAM will = more performance, so that is how MXM is faster than BGA - because of futureproofing. It's not derailing just because you don't like the answers you are getting.

BGA is trash. Deal with it.

 

If the GPU die is so close to the CPU, it is likely the GPU will run hotter than on a separate board. You also have the issue of VRM overheating and possibly less VRMs than separate MXM boards. A MXM GPU would likely, if all else is equal, hold boost clock a lot better and overclock a lot better. The minuscule latency is irrelevant for more or less everything. If at most, the GPU is 1 meter away from the CPU, you have a worst case latency of 4*10^-6 ms of delay. When your panel have delays in single/two digit ms? It simply doesnt matter.

There is no real argument for BGA besides saving valuable space and weight. For certain laptops, it does make sense to run BGA. Even then, when my friend told me his core M laptop had soldered RAM, I just laughed at how silly that is. In the case of a power user's machine, it is silly. The manufacturer pays more for RMA and the consumer pays more for lack of upgrades.

 

Calm down. You really didn't read the first post at all, did you? Check the last paragraph, please. Furthermore check my post history where I mentioned that I wasted $600+ on two motherboard replacements because the BGA GPU in my laptop failed. Stop hollering that 'hurr durr BGA sucks deal with it yo'. I know.

I'm not saying anything has a performance difference over the other (forget the VRAM difference - it plays no part in my question). I'm asking if they do.

Let's make it simple, and scientific. We test only one variable, and all else is kept constant.

Assume laptop A has an rPGA 947 i7-4900MQ CPU, 8 GB 1600 MHz DDR3 RAM, an M.2 Samsung Evo 840 and a BGA GTX 970M (3 GB VRAM).

Laptop B, has the exact same specs and cooling set-up as laptop A, except for the fact that it has an MXM GTX 970M (also with 3GB VRAM - I know it doesn't exist, just hold on, please...)

Now, let's say we put both these laptops in separate temperature-controlled environments, both at equal pressure and air temperature.

Then, we run a computation program that measures FLOPS only (forget Fire Strike, relies on too many variables) over several hundreds to thousands of times to account for variations in the scores, performing the same number of tests for both laptops, with a 5 mins cooling period between tests. After all this, we do some tallying, and get two normal distributions of scores; one for each laptop.

My question is this: will there be a significant difference between the two means of the distributions just because one laptop has an MXM GPU, and the other, a BGA one?

@tgpier's response just above appears to be on the right track. Thanks.

 

I think MSI has an MXM GTX 970 3 GB, so it's real. And BGA GTX 970m 6 GB and GTX 980m 8 GB exist too, so there's no ruling those out if we wanted to do tests with them and their MXM form factors.
But regardless, as long as the VRAM on both laptops are the same, the experiment should be fine, yes?

 

I still stand by my post 100%. Currently, it probably does not matter too much either way - MXM (6GB) and BGA (3GB) performance should be about the same. However, in the future, once more graphic-intensive and VRAM-intensive games come out, the MXM will be the better performer. How much in the future, I don't know, but I know eventually this will be true - kind of like how 1GB VRAM isn't really enough these days.

 

The main advantage the BGA machine has is it's thinner, the lack of board over board reduces the Z-height of the setup allowing for thinner devices.

 

There is zero performance difference, it simply changes upgrade-ability and the amount VRAM.

Any advantages to BGA's proximity to the CPU are so small they can't be measured.

 

I don't know about the others, but I speak for myself, my post was related to this:

Yes it's still BGA, but not exactly and I've answered why.

With this cleared, no there are no advantages one over another given everything else is the same - same silicon chip, same vRAM (not only amount, but speed and manufacturer too), same power circuity, same vBIOS. BUT the thing is that a lot more corner cuttings occur on BGA machine than MXM one, especially on vBIOS level, and you might no be able to fix that. So you are more likely to get better performance if you go MXM.

Do note that I don't mention cooling...

 

It also makes it easier to diagnose a hardware problem like the one I am facing at the moment and then replace the faulty part economically both on the wallet and quite importantly the environment.

Likely my motherboard is dying but I won't need to throw away the gpu or cpu and the motherboard won't cost 1000 dollars (ie new system) it costs 100 dollars. 1/10 the cost.

It is kind of ironic that during a time of trying to reduce "carbon footprints" and recycle etc computer companies now universally want to create more waste and proactively encourage a throwaway society.

For those of us with children it is actually pretty sickening really! I shudder to wonder what world we will hand down to the next generation devoid of natural resources and clean air to breathe.

 

So all in all, ceteris paribus, BGA has no performance benefit over MXM, and vice versa?

That clears it up, thanks, everyone.

Now we also know MXM >> BGA for the various other reasons mentioned.

We need a standardised card layout instead of a pin/port layout. Three sizes that fit three different classes of notebooks. I have never yet seen a small form-factor notebook (13/14") with MXM in it, and there's certainly a market for such a machine. Maybe Clevo will do it.

 

Unlikely! The bottom line is money not quality.

Once things are no longer niche quality goes out the window.

A mxm based system doesn't have to be more than 3mm thicker than it's BGA counterpart. Particularly if GPU manufacturers stopped putting vram chips on both sides of the board to cut down heat.

As for 13 - 14" machines. It already is possible. With the lack of ODD's in new laptops all they have to do is utilise MXM 3.0A modules. Smaller and more compact yet still able to pack enough punch for gaming. Particularly if the next generation is as efficient as is being touted (aside from rumours of Pascal heat)...

 

Kind of a weird topic, since assuming you're comparing the "same" card (say a BGA 3GB 970M vs MXM 3GB 970M) the actual performance will likely be almost identical, assuming the OEM didn't screw anything up in the implementation.

Current performance, generally speaking, is not what makes MXM superior to BGA. However, a MXM card has the potential to perform better, as you're far more likely to get more VRAM on the MXM version, and the VBIOS seems to be a bit more flexible/easy to mod on MXM cards, allowing for better overclocks. Though I could be wrong on that last part.

Also, cooling on MXM laptops tends to be a bit beefier, so you often have a bit more OC room too.

In other words, its not that the MXM version is more powerful by default, but all the other aspects of the MXM format that allow it to often be a bit more powerful.

 

So BGA and MXM should perform identically, if nothing else is different. My question has been answered.

I see, so although there's nothing that inherently improves an MXM GPU's performance over a BGA GPU, the former ought to perform better just because of other factors? That sums it up nicely. How is it that MXM GPUs have better VBIOS implementations than their BGA counterparts?

Succintly put.

 

Essentially MXM cards are more robust in every way.

 

Nope, you most certainly are right about that part (and the other things you said, too).

 

There's also the added benefit for OC nuts of being able to shop around for a MXM card with better silicon, especially so with CPU's.

 

Did you mean this? http://forum.notebookreview.com/threads/high-quality-videocard-vs-low-quality-videocards.785085/ Forget the opportunity to choose quality hardware if you are one of those Kool-Aid drinking kids who enjoy/love BGA hardware in a slim Apple similar box Freely translated by @Mr. Fox very good description of oblivious gum kids.
For those who do not know what Drinking the Kool-Aid means https://en.wikipedia.org/wiki/Drinking_the_Kool-Aid
On time @woodzstack also visit this thread...

 

Well, simply put, it's because they're disposable. If something dies in a year or two and warranty is done, might as well buy a new one. Cheaper to use less vRAM.

Most MXM used higher vRAM counts because they were harder to replace, and when vRAM became a problem for older cards the laptop cards wouldn't really care (like my 280M for example; if that was a 512GB card instead of a 1GB card, it would have become worthless LONG before the 2013 death of my laptop). It was a trend that simply never went away. Of course BGA has high vRAM now and MXM has low vRAM too, so it's just a matter of taste. MSI uses both 3GB and 6GB 970Ms on their GT60/GT70/GT72 models, as well as 4GB and 8GB 980Ms.

Alienware chose 3GB/4GB 970M/980M, but Gigabyte chose 6GB/8GB 970M/980M. Clevo now offers 3GB/6GB 970Ms and 4GB/8GB 980Ms, at least via Eurocom, on theP6xxRx models.