A Bunch of New SFF Cases for Pascal/HBM2 GPUs?

[Dedaciai]

Hi Everyone,

What are your thoughts on us seeing a slew of smaller or at least more creatively small SFF cases after the release of Pascal GPUs? Since HBM2 is going to shrink the size of Nvidia's top of the line cards (let's hope) to about the size of the Fury X (7.5 in. or 190mm) or Fury Nano (6in. or 152mm) do you think case manufacturers will follow suit and offer new cases to suit the smaller GPUs? The reason I bring this up is because it seems like all of the really cool SFF cases out there are always bigger than they need to be for the sole reason that some of the top of line GPUs exceed 11in (280mm). I am for one looking forward to more SFF options.

Dedaciai

 

[jeshikat]

The problem I see is that most builders have been trained that bigger coolers are better. We've seen with the Fury (non X) that the AIBs have no problem taking something with HBM, and making it just as large, if not larger, than previous gen cards.

I hope to see more small cards and cases to take advantage of them but I expect that large cards will still be the norm.

 

[PlayfulPhoenix]

The other thing, too, is that arbitrarily large cards can support larger coolers, which means lower temperatures and noise when all else is held steady. Though cooler chips provide manufacturers with the option to make "SFF" cards, the default improvements will probably be in the categories of performance, temperatures, and acoustics, rather than size exclusively.

 

[|||]

Nvidia has also mentioned there will be two form factors for the upcoming Pascal cards. MXM or at least something similar to MXM) and add-in-cards. The reference AIC may shrink in size, but you can be sure that when the vendors start pumping out their own interpretation, cards like the EVGA Classified, MSI Lightning, and others are going to grow in size to at least what the current reference cards are when they start adding 16+3+1 phase power delivery circuitry, multiple PWM controllers, voltage read-out headers, etc.

 

[EdZ]

Those coolers are often necessarily large. The R9 Nano is a good demosntration that even with cherry-picked chips a cooler that small will struggle to maintain temperatures low enough to keep core clocks up.
We'll probably see something similar to the GTX670 where the PCB gets smaller but the coolers remain the same. OEMs won't be producing PCBs larger than they need to be except for over-the-top 'halo' products, as it's cheaper to produce smaller boards.

 

[Dedaciai]

The problem I see is that most builders have been trained that bigger coolers are better. We've seen with the Fury (non X) that the AIBs have no problem taking something with HBM, and making it just as large, if not larger, than previous gen cards.

I hope to see more small cards and cases to take advantage of them but I expect that large cards will still be the norm.

True. At the very least I hope we see more cards in the Nano/970 mini size and capabilities than just the Nano and Asus' and GB's mini 970.

 

[PlayfulPhoenix]

I think we will, too - they just won't be the "standard" cards or flagships coming out of nVidia. Which is fine, insofar that any ITX or SFF-oriented graphics until now were done by manufacturers in spite of the fact that nothing architecturally made it easy. Having direction from above, and parts specifically made for that purpose, will provide tremendous incentives for that niche to be filled this time around.

 

[iFreilicht]

I agree with the overall sentiment. Pascal/HBM2 are good steps towards smaller cards, and we will probably see a few more of them because of that, but the main market will still be the large coolers, which will probably become even quieter and/or cooler.

It could be a good time to get cases design for ITX or otherwise shorter GPUs out into the market and make yourself a name in the industry for that, but I don't see large manufacturers really going all-in on that. ITX cases will probably stay too large for quite some time to come.

 

[Phuncz]

While most are hoping Nvidia's possible future cards will have an SFF option, let's not forget what AMD has already done for SFF in 2015, which in all likeliness is going to continue in 2016. I'm expecting more from AMD that will benefit us than Nvidia for 2016.

 

[CC Ricers]

To be more SFF friendly NVidia could also make some of the lower end reference designs single-slot friendly. All the spiffy single slot coolers were last seen on the GTX 750 Ti and they were also not exactly easy to buy from if you lived in North America (don't know about Europe, though).

 

[|||]

It is getting to the point where they should just put a USB extension that goes to the motherboard and have it pass through to a USB Type-C connector on the back of the card that has USB + DisplayPort (dual-mode supporting HDMI and DVI, with a proper adapting cable) connectivity.

 

[iFreilicht]

So you're talking about USB connectivity on the GPU I/O? Why would that be helpful?

 

[|||]

Small reversible connector that is useful for a lot of different I/O. Hopefully it doesn't take much longer for most of the I/O on the rear I/O panel of motherboards to move to USB Type-C, as well.

 

[GuilleAcoustic]

Not much informations on Pascal, but from what I've read most of them will use GDDR5X. HBM2 is for top of the line cards and aren't expected before the end of the year.

Sadly ... coolers tend to be protruding more and more, like the Asus Strix for example. GPU almost doubled in height

 

[iFreilicht]

Small reversible connector that is useful for a lot of different I/O. Hopefully it doesn't take much longer for most of the I/O on the rear I/O panel of motherboards to move to USB Type-C, as well.

That's fine, but instead of running a cable from the GPU to the board, that data could be passed via the PCIe bus just as well, even though I see little point in it. If all other components used C-Type connectors, you'd have such a load of connections on the board alone there'd be little use for even more of them. I do believe that the ability of full-size DP cables to lock sturdily and securely is very useful though, so it seems like the better choice for video output, if size of the connector isn't absolutely crucial.

Not much informations on Pascal, but from what I've read most of them will use GDDR5X. HBM2 is for top of the line cards and aren't expected before the end of the year.

Sadly ... coolers tend to be protruding more and more, like the Asus Strix for example. GPU almost doubled in height

That is just crazy. If they used this space for nice, large fans, I could at least somewhat understand it.

 

[Dedaciai]

Not much informations on Pascal, but from what I've read most of them will use GDDR5X. HBM2 is for top of the line cards and aren't expected before the end of the year.

Sadly ... coolers tend to be protruding more and more, like the Asus Strix for example. GPU almost doubled in height

Yeah, the super tall/wide cards sort of annoy me as it can be very hard to find a normal sized SFF case to squeeze them into without having to take a side panel off, lol.

 

[|||]

That's fine, but instead of running a cable from the GPU to the board, that data could be passed via the PCIe bus just as well, even though I see little point in it. If all other components used C-Type connectors, you'd have such a load of connections on the board alone there'd be little use for even more of them. I do believe that the ability of full-size DP cables to lock sturdily and securely is very useful though, so it seems like the better choice for video output, if size of the connector isn't absolutely crucial.

In order to pass it through the PCI-e connection, they'd have to either:

1. Bifurcate the link and have one go to a third-party PCIe-to-USB controller (not efficient use of the bandwidth)
2. Have a PLX switch on the GPU and split off to a third-party PCIe-to-USB controller (not worth the cost or latency)
3. Integrate USB controller logic into the GPU silicon (requiring direct support from Nvidia and AMD)

The benefits of using USB Type-C connections is that you not only get the display connection, you get USB to support keyboards, mouses, and other USB devices through the monitor, and potentially even power it all with USB Power Delivery v2.0 (up to 100W) all through a single cable.

 

[iFreilicht]

In order to pass it through the PCI-e connection, they'd have to either:

1. Bifurcate the link and have one go to a third-party PCIe-to-USB controller (not efficient use of the bandwidth)
2. Have a PLX switch on the GPU and split off to a third-party PCIe-to-USB controller (not worth the cost or latency)
3. Integrate USB controller logic into the GPU silicon (requiring direct support from Nvidia and AMD)

The benefits of using USB Type-C connections is that you not only get the display connection, you get USB to support keyboards, mouses, and other USB devices through the monitor, and potentially even power it all with USB Power Delivery v2.0 (up to 100W) all through a single cable.

I see, didn't think about the power delivery. There could be problems regarding the power budget of the GPU and the load on the PSU, especially with smaller systems, but I'm sure those can be solved.

It would require direct support from the GPU manufacturer anyway, wouldn't it? Getting sound through HDMI or DP is a similar problem and that seems to be solved, do you know how that's implemented?

 

[Josh | NFC]

Those coolers are often necessarily large. The R9 Nano is a good demosntration that even with cherry-picked chips a cooler that small will struggle to maintain temperatures low enough to keep core clocks up.
We'll probably see something similar to the GTX670 where the PCB gets smaller but the coolers remain the same. OEMs won't be producing PCBs larger than they need to be except for over-the-top 'halo' products, as it's cheaper to produce smaller boards.

My Nano has a really nice cooler, but I wonder if the 285ITX and 380ITX coolers would be more effective. They LOOK and FEEL far improved over the Nano's. A vapor chambor + those sexy 6mm heat pipes and fin/fan design seem like it would help the Nano's cause--but I don't have a way to test this.

I will say that my Nano run nice and cool and doesn't throttle in my test bench, it is just when you stick it in a crammed USFF case that it gets hot to where you need to choose between noise and clocks. With Bitfenix, Corsair, and even Lian Li to an extent pushing massive "mini itx chassis" I bet these cards perform well in the mainstream chassis.