I predict the return of cartridge based CPU's (poll inside)

[moonbogg]

Shrink limitations will occur and core count will need to increase. Can't increase the socket too much, so they will flip perpendicular and grow "up" off the board since they got nowhere else to go. PCB with nice water blocks designed to bolt right onto it and whole assembly slots right into the board. Will happen.

YES OR NO
YES OR NO
YES OR NO

 

[moinmoin]

No chance, cartridges offer way too few pins. Even PGA sockets are too limiting not to cause compromises (like AM4 offering only 24 PCIe lanes out of 32 possible ones). LGA and increasingly larger packages will be more common.

 

[UsandThem]

The last time we saw cartridge-style CPUs, I bet many of the newer users on here weren't even born yet.

I have to say I really liked how easy they were to install compared to carefully handing modern CPUs without damaging the pins on the motherboard (Intel) or the pins on AMD CPUs. I've done it so many times, and I have never damaged anything, but it almost feels like I am playing the game 'Operation'.

 

[VirtualLarry]

I'm with you, Bogg. I've suggested something similar in the past too. I think it makes a LOT of sense, at least to me.

And for those claiming that there aren't enough pins, what about two, or three stacked card-edge connectors? I think that the Xeon slotted CPUs utilized two rows, and Neo-Geo cartridges definitely did. Although, if CPU "carts" get that large, hmm, dunno. Maybe an AMD mega-APU? 300W with integrated water-cooling? GPU and 32-core CPU in one?

Edit: If CPUs were in the form-factor of a Neo-Geo cart (roughly VHS-sized, if you've never seen one), it might make it more likely to be upgraded during the lifetime of the PC. Or it could allow for just-in-time assembling of PCs, build them, and slot in the CPU just before shipping. Either way, you wouldn't need a tech, or to be a tech (like us), to swap CPUs.

Edit: With the larger-sized carts, maybe they could included HBM2 or 3DXpoint NVDIMM-P soldered to the CPU cart, much like the L2 cache was soldered on to the Pentium II Slot-1 carts.

 

[amd6502]

Wasn't there a rasperryPi-like compute stick on DDR2 socket interface sometime in the past 3 years?

I could see compute sticks as inline module.

For anything passively cooled (low wattage) it might not be a bad format. The water cooled idea is not bad but I still can't imagine this sort of arrangement in anything high powered.

I like the compute module soc idea though. Why aren't more servers built like this?

 

[Ajay]

No way. Moinmoin is correct.

 

[Gikaseixas]

Still remember my pentium III. Cooling would be a challenge IMO.

 

[VirtualLarry]

No way. Moinmoin is correct.

Not if the memory is all on-cartridge too. Then many/most of those pins, don't need to be connected externally on the package.

 

[moonbogg]

Claims of design limitations preventing carts from resurging is really describing an imagination limitation IMO. If it makes more sense to grow outward (like a GPU) in order to keep main boards a reasonable size, then the necessary accommodations will easily be made. Take the size of a common high-end GPU PCB for instance; how much space does such a board allow for CPU growth, not to mention all the other interesting things that could be added as suggested by VirtualLarry?

You can make the move to a PCB like that and have enough room for growth for many years to come. GPU's run hot as hell and have no issues with cooling. Cart CPU's can be cooled easily, either by air or water. They can be made to slot in next to each other, increasing the core density of a motherboard without resorting to a huge board with multiple sockets. Makes much snese. Much snese indeed.

 

[Vattila]

Nah. Let's have two-sided motherboards with 4 fully connected sockets on each side (perhaps a mix of socketed CPU, GPU and APU), with low-profile water-cooling, making it all fit in a 1U server.

 

[BSim500]

Don't CPU's have to be a certain physical distance from the RAM sockets for latency reasons? And also physically close enough to VRM's to avoid needing an additional array of capacitors to minimise voltage fluctuations? Of course, you could move the RAM sticks, VRM's, heatsink, fan and 8-pin power cable onto the cartridge. But then you've ended up with less of a "cartridge" and more an L-shaped motherboard...

 

[DigDog]

Im gonna go with NO but if the option had been there, i would have picked WUT? instead.

Never seen vertical cpus, and not convinced about this idea that a cpu could be too big to fit on a mobo.

 

[moonbogg]

Im gonna go with NO but if the option had been there, i would have picked WUT? instead.

Never seen vertical cpus, and not convinced about this idea that a cpu could be too big to fit on a mobo.

Pentium 2/3. Also, with chiplets being the new way to do things, they will just start piling them on like pepperoni on a pizza.

 

[DigDog]

I have never SEEN a vertical cpu. Ok, so they existed, but the argument is that the dies are getting bigger (not really, but we've recently seen an increase in core count, so that could continue) and this makes them too big for a future socket ...

Nah.

I would guess that vertical cpus were due to fabrication, not necessity. The socket itself is today larger than it needs to be so more of it could be filled, i have no problem imagining a 30-core cpu fitting into a AM2 sized socket. Also not buying into there being any practical limitations on future socket size. If the future demands it, we'll have 100 cores on fist-sized sockets; why would we have *any* issue with that ... pcbs are extremely easy to work with.

 

[Kenmitch]

I had the Celeron 300A back in the day.

I'm leaning towards undecided.

 

[chrisjames61]

Remember the "Slotket" adapter cards around the early 2000's time frame?

 

[Arachnotronic]

Never.

 

[maddie]

Moonbogg, you're getting bogged down by aiming too near. Don't be limited by your name and aim for the stars.

I say a fractal, slots on slots unfolding into a new computing paradigm. What say you?

 

[moonbogg]

Moonbogg, you're getting bogged down by aiming too near. Don't be limited by your name and aim for the stars.

I say a fractal, slots on slots unfolding into a new computing paradigm. What say you?

I say you're trolling. You asked.

 

[IEC]

My last slot CPU was a Celeron 300A OC'd to 450.

 

[sdifox]

You'll see "pinhead" like config before you see return to slots.

 

[gorobei]

a simple search suggests that slots were a result of ease of manufacturing (way to get l2 cache chips closer to cpu before they started putting it on die) and business (get off common socket to screw over amd/cyrix/via). https://forums.anandtech.com/threads/why-did-intel-go-from-socket-to-slot-to-socket.431403/

so slot based for more pins doesnt really get you anything, and just complicates cooling (you arent mounting a 5 pound noctua air tower on a slot cartridge. most of those slot cpus had a tiny 20mm fan) if pins/board area was really an issue they could just go with the mezzanine type interconnect that nvidia is using for the hpc stuff. even then they would never mount the die perpendicular to the board like a slot card.

 

[moonbogg]

When they are announced there will be no sweeter vindication. Think of your reasons and hold on to them real tight.

 

[coercitiv]

Not if the memory is all on-cartridge too. Then many/most of those pins, don't need to be connected externally on the package.

Yeah, let's put some I/O on that cartridge and call it a motherboard.

 

[Ichinisan]

A Pentium II / III size cartridge could fit a whole computer these days. The larger "motherboard" would really just be for interfacing with storage and I/O.