Concept SENTRY 3.0: Development and Suggestions

[SaperPL]

Hi there, I have a sentry 2.0. I am planning on building with 5080. Your manual says the highest length of the GPU is 305 mm. I saw one GPU with 306 mm and 50 mm thick. Can I cram that inside the Sentry 2.0? I couldn't find any info on the thickness allowance in the manual.

If the card isn't taller/wider than what's in the drawing, then you have 306mm minus powder coating thickness of 0.1~0.125 per side so 305.75, but that's assuming perfect bends, but these maj be slightly tighter for various reasons, so you should assume something like 305.5mm. Sentry 2.0 wasn't stamped but each bent was made separately so that's some leeway there, but you have the case so you should be able to measure this somehow. The question is how precise is that 306mm length of the GPU you want to buy. Also note the surface of the side bracket attached at the front that will affect this if the card is significantly taller/wider than the reference card.

As for thickness, we've seen 50mm cards fitting in Sentry, but we've also seen different ways of measuring the thickness because pci bracket is slightly offset from the card, and also there are some fans that can move outside of the card thickness on their bearings and potentially scratch themselves against the perforation. The case was designed for 2 slot cards, so we didn't show a specific thickness anywhere. 2 slots means ~41mm of thickness.

EDIT: I forgot another thing - how are you planning to squeeze in the 12 pin power cable?

 

[SaperPL]

Not saying you should do a Sentry 3.0 now, but, it sure looks as an improving landscape for it.

The angled connector is a step in the right direction, but we're still in the situation where the card measurements are not all that needs to be taken into account it's fitting. Actually Inno3D's 4070 stealth approach is the only approach so far that was within roughly reference sized cards outline, but seeing the list of the cards, there's no 5000 series cards like this from them listed.

Founders Edition card having that angled connector is still an inch taller than the reference sized card, which creates this situation where it's not really good for console cases unless they have really a lot of space for the GPU and at the same time it's not good for sandwich cases if it's going to push the hot air into the back of motherboard and PSU. And also most of us probably won't be able to get our hands on the FE models anyway as they will be scarce and scalped right away.

Curiosly enough back in june 2024, NVIDIA decided to launch the SFF-Ready program, in which they defined the category "SFF-Ready Enthusiast GeForce Cards"
with this size limitations:

• 151 mm maximum height including power cable bend radius
• 304 mm maximum length
• 50 mm or 2.5 slots maximum depth

Which no RTX 4090 complied and very few 4080s, and also noticed they have quickly updated the diagram picture showing the dimensions from a 40 series design to a 50 series design.
In hindsight seems like they were preparing themselves.
Again, seems too much effort for just one-off of reasonable sized GPUs

I wrote an editorial about this here on sff, the SFF-Ready program has a lot of shortcomings. Maybe a step in right direction, but still seems like a paper marketing to promote friendly brands of cases.

Now think about it like that - that 5090 is technically compliant, and there are SFF-ready sandwich cases where it'll try to blow that hot air right into the riser behind the motherboard and back wall of the PSU - is it really that properly planned thing?

 

[Shak]

If the card isn't taller/wider than what's in the drawing, then you have 306mm minus powder coating thickness of 0.1~0.125 per side so 305.75, but that's assuming perfect bends, but these maj be slightly tighter for various reasons, so you should assume something like 305.5mm. Sentry 2.0 wasn't stamped but each bent was made separately so that's some leeway there, but you have the case so you should be able to measure this somehow. The question is how precise is that 306mm length of the GPU you want to buy. Also note the surface of the side bracket attached at the front that will affect this if the card is significantly taller/wider than the reference card.

As for thickness, we've seen 50mm cards fitting in Sentry, but we've also seen different ways of measuring the thickness because pci bracket is slightly offset from the card, and also there are some fans that can move outside of the card thickness on their bearings and potentially scratch themselves against the perforation. The case was designed for 2 slot cards, so we didn't show a specific thickness anywhere. 2 slots means ~41mm of thickness.

EDIT: I forgot another thing - how are you planning to squeeze in the 12 pin power cable?

Thanks for the response.
I think with this generation of GPU we will have smaller cards. I just saw one 5080 with 41 mm width from Inno3d. I should wait for the GPU purchase, I am optimistic that we can see a blower-style GPU from one of the vendors this time. Also from the last generation Asus ProART 5080 was awesome. If I can fit everything within, I wish to play with the thermal management.
By the way, will be happy if we can see the Sentry 3.0 very soon. ***Great Design *** Sentry 2.0. A few mm of headroom for the v.3 will be very appreciated given the trends of the cards. Although you already have some space left if you consider the space behind the GPU.
Another thing is all of GPU now a days have flow through design. If you could consider the perforation on the back side of the GPU that would be a great addition to the Sentry 3.0.

 

[Shak]

Watch out this page

small-form-factor-sff-ready

small-form-factor-sff-ready

www.nvidia.com

 

[SaperPL]

A bit of an update on why we are stuck with the project, which can be clearly shown right now with the most recent GPU launches.

When we launched the first Sentry, the overall estimated compatibility rate of the graphics cards for the current at the time generations of the card, so GTX 10xx series and AMD 4xx series was roughly 60% and it wasn't just lower end cards - there were founders edition cards available up to the top tier, most of the vendors had at least one or two top tier cards in pci-e reference size and we had all PNY cards being compatible etc.

When we launched Sentry 2.0 this estimated compatibility dropped to roughly 56% in RTX20xx / AMD Vega / 500 series, but it wasn't that easy to find a fitting high end card anymore - most of the high end cards got big, but there were still a lot of sub-top-tier cards, so 2080s that were going to fit.

This is the reference in compatibility spreadsheet

When the RTX 30xx came it got obvious that things are going wrong way for us with how big cards are and how hard it will be to get a fitting card for the Sentry 2.0 compatibility spec, so we were slowly working on other issues in the case design and trying to wait and where it will all come after the pandemic.

Looking at the launches starting this year it became even worse. Out of all the cards released, I think there is just a single PowerColor Radeon RX 9070/XT Reaper design that will fit Sentry 2.0, so two cards out of few hundreds in total? And it's also not sold at MSRP anymore and it's hard to get, but we might see the more availability in the upcoming weeks.

Going up in size significantly doesn't make any sense from our point of view - if we go up, and everyone goes up in size for the sake of supporting bigger cards, the cards will just keep getting beefier because people will keep buying them for SFF builds.

There is a way to support bigger cards in Sentry/Steam Machine configuration but it requires specific type of motherboard that allows intrusion of pci bracket ends into the IO area. What this allows is connection of the graphics card with just single 90 degree PCB riser, no additional extension required, which probably should come handy when it comes to pci-e 4.0 and 5.0 support because even if separately two PCB risers are spec'd for 4.0 or 5.0, it doesn't necessarily mean putting two of them together will guarantee stable work in this version of pci-e.

Asrock has been doing their boards this way since AM4 B550 boards, so for 3 generations of boards right now, but it's not an official feature, it may be just outcome for optimising their layout. I've shown this on reddit here and here. I haven't seen other board vendors doing it this way apart from some server or low end boards that don't have a lot of connectors. The question is if we should be willing to risk Asrock optimising AM6 boards differently and ending support for such configurations.

Anyway with such boards and a single PCB riser, we gain at least 1 slot width / 2cm of the shortest PCB extension riser size. This shift potentially means even supporting 5090 founders edition in Sentry 3.0, but at the requirement of using board such as the current Asrock boards.

We have investigated options to support both configurations, but we don't want to go that way because it will cause one of the configuration work like an afterthought/mod and be sub optimal, while also creating and overhead of additional parts OR forcing the end user to 3D print adapters, and also forcing us to make room for those adapters making the structure weaker. While shifting the spine in back-to-back sandwich cases is easy to design, here we would need to shift the attachment point of the card which in Sentry is part of the case structure, and simply we cannot make this part fully modular without consequences.

This is kind of a hen and an egg problem where either we wait for cards to get back to reasonable dimensions OR we jump the gun and go either way - supporting only Asrock and alike boards blindly, OR going out with a case that you cannot get any recently released GPU apart from that PowerColor 9070 for it. But we don't really want to blindly go in with this right now because this time around we want to make mass production tooling for the cases for the production run, so we don't want to have surprise right when we're ready to manufacture that now again the cards will grow, or Asrock changes their design of itx boards and so on.

 

[Talyrius]

IIRC, Wendell of Level1Techs has had discussions with ASRock engineers in the past. It couldn't hurt to ask if he can still get in contact with them.

 

[SaperPL]

IIRC, Wendell of Level1Techs has had discussions with ASRock engineers in the past. It couldn't hurt to ask if he can still get in contact with them.

There was a rep from Asrock here on the forums - asrock-system - I talked with him directly for some time before the launch of Sentry 2.0, but for the questions about this specific feature of the boards when they came out I never got any response. I would assume that we're either too small for them to care about us at this point, or I asked something that they are not allowed to talk about on the outside at all.

Putting on a tinfoil hat I would say that IF they actually cooperated with Valve to make those 300 steam machine prototypes because their boards where in those, then maybe it's not a coincidence that this kind of board appears around the time steam deck is in development - maybe they have something waiting for valve to make the move. Or maybe it's just something different that optimising the board this way saves them a lot of money or allows for whatever else, and they won't talk about it.

Either way, the problem is that this feature is not a standard, and I think it actually should be, but it's not going to happen without a proof of strength in numbers that people would want this, and effectively this is either about showing that we or someone else would start selling something using this feature in significant volume, or community comes up with a way to ask for improvements in ITX form factor spec with proof of numbers like using one of the petition platforms and gaining significant number of signatures for a list of requested spec improvements. The problem is that getting there needs some level of professional approach and getting some youtubers/influencers on board to have reasonable amount of people willing to sign something, because most of the people don't understand such complex problems and assume you're a crazy person wanting to throw your ideas at professionals (Homer Simpson car design case) until that's a won battle and they want to be a part of it.

 

[MarcParis]

I agree with the status on the current market (present & even near future).
I would even add that the SFF world above/equal of 10L is becoming a bit boring with tons of uninspired copy/paste setups, with unoptimized airflow, unpractical layout, oversized & uneffective cooling solution...

As you mentionned, innovation could come from the MB, here are the leads I can suggest :

• Minisforum with mobile cpu (BD790i for example) is offering interesting exhaust path through rear i/o shield and 2xm.2 on the recto of the MB. Issue of the location of the heatsink is that is can interfere with the case (got the issue with the SM550, not with the formD T1)
• Arrival of new SOC like AMD Ryzen™ AI Max 300 Series with powerful integrate GPU (approx performance of a RTX 4060/4070). However except some OEM, like Framework (mini ITX), I did not see any other more standard OEM going to this market

On the GPU side, maybe you could check compatibility with deshroudded GPU, with a specific can fixation on the side panel to make easy process.

 

[SaperPL]

Arrival of new SOC like AMD Ryzen™ AI Max 300 Series with powerful integrate GPU (approx performance of a RTX 4060/4070). However except some OEM, like Framework (mini ITX), I did not see any other more standard OEM going to this market

I don't think those APUs will be actually approx performance of desktop RTX 4070 - maybe when compared to laptop ones, where you have similar chip, even when it's not cut down number of compute units for laptop version, but the TDP is reduced in comparison to desktop. With that said, 4060 laptop is really close to 4060 desktop so if that's comparable to those new APUs, then it is interesting.

On the GPU side, maybe you could check compatibility with deshroudded GPU, with a specific can fixation on the side panel to make easy process.

If you mean removing the PCI bracket and figuring out how to match it with some custom PCI bracket - I did look into this, but it is again a mess that within same generation and same nvidia chip one vendor may have different positioning of the display ports and placement of the screws used to attach the bracket to the card, so making something universal is close to impossible.

Ideally if each vendor would stick to same layout across all of their cards, or at least would stick to the same way of attaching the bracket with screws in the same place that is not connected to the display outputs, we could consider adding multiple brackets to be swapped in those cards. The problem is that some of the cards are made in a way you need to detach the backplate, and sometimes the backplate is connected with something on the side, and sometimes removing the backplate may damage thermopads on the memory chips, and I've seen that some vendors like inno3D are still adding "warranty void if removed" stickers to scare away consumers from disassembling their cards.

Also I think that at some point the connectors should be removed from the cards - if you have 2 slots at the back of the card, it is ideal place to exhaust all the heat the way you actually want it when designing the card. Imagine having there 2 thick 40mm fans pulling out the heat, helping the blower ad the other end, and both sides could be lower RPM in this push-pull configuration. But this requires handling and bandwidth for routing back the display output to the board or some additional display pass-through cable connected to the card on its side instead on the bracket.