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

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

YES OR NO

  • YES

    Votes: 10 9.8%
  • NO

    Votes: 92 90.2%

  • Total voters
    102

Topweasel

Diamond Member
Oct 19, 2000
5,436
1,655
136
I think one of the limitations of slot/cartridge based CPU's is the cooling. These days if you look at how tall cooling is, it would not be trivial on high end CPU's to do the same when the CPU is orientated horizontally. Yes, they could use heat pipes, or AIO's or something else. But the design's would always have to work around the some "safe zone" so that it didn't intrude on memory or PCI slots. Basically, there is no benefit, to having the CPU package with all it's pins on the board like we have now.
*edit* Oh, and think about the weight of the current norm for coolers, how do you support 2+lbs of copper horizontally?

Well a Back brace would help AMD and Intel did this for their slot cards. But its still a good question. For the most part we would be better looking towards Video Cards for our inspiration for this. I mean maybe in a 9" length you could do a 100w cooler for most desktop chips that's single slot. But take overclocking or larger workstation and server chips in, you need either the monsters we work with on sockets or an adapted version of the coolers we use for GPU's and a plug in mechinism less like a free standing cartridge and more PCIe expansion card. Maybe something in between where you can fit a rail system in for support like PXI. But honestly for it to happen, I see a PCIe like setup being the answer as that keeps chassis design kind neutral.

But that leads to why this will never happen. You do something like and all off a sudden. Only mid tower ATX and larger become viable solutions and what wasn't the case back then, is the case now. Mobile. There wasn't a real demand for laptops because they were too expensive being behind in mobile processor solutions wasn't an issue. Intel isn't going to develop desktop solutions that aren't tarted up laptop cpu's SB - CFL-R and even Icelake are first and foremost laptop solutions.
 

NTMBK

Lifer
Nov 14, 2011
10,247
5,038
136
I think one of the limitations of slot/cartridge based CPU's is the cooling. These days if you look at how tall cooling is, it would not be trivial on high end CPU's to do the same when the CPU is orientated horizontally. Yes, they could use heat pipes, or AIO's or something else. But the design's would always have to work around the some "safe zone" so that it didn't intrude on memory or PCI slots. Basically, there is no benefit, to having the CPU package with all it's pins on the board like we have now.
*edit* Oh, and think about the weight of the current norm for coolers, how do you support 2+lbs of copper horizontally?

I mean, AMD managed to cool a 250W Hawaii GPU+DRAM with an air cooler in a PCIe slot.
 

joesiv

Member
Mar 21, 2019
75
24
41
Well a Back brace would help AMD and Intel did this for their slot cards. But its still a good question. For the most part we would be better looking towards Video Cards for our inspiration for this. I mean maybe in a 9" length you could do a 100w cooler for most desktop chips that's single slot. But take overclocking or larger workstation and server chips in, you need either the monsters we work with on sockets or an adapted version of the coolers we use for GPU's and a plug in mechinism less like a free standing cartridge and more PCIe expansion card. Maybe something in between where you can fit a rail system in for support like PXI. But honestly for it to happen, I see a PCIe like setup being the answer as that keeps chassis design kind neutral.
I thought about that, but things like the distance to system memory, and trace lengths become a problem (or troublesome), the PCI slot has pretty good retention capabilities as you suggest GPU's get heavy, but GPU's also have their memory embedded, so trace lengths can be evenly spread out on the GPU PCB. The closer the memory the better for processors, also even trace lengths makes a big difference for timing, currently, CPU's get memory on both sides which helps to even out the distances, throw it in a slot, and there is only one way to go, farther and farther away from the CPU slot.

I mean, AMD managed to cool a 250W Hawaii GPU+DRAM with an air cooler in a PCIe slot.

Yes, if they were to line up the CPU slot similar to the GPU at the back of the board, so that it could have the rear top mounting point, and enough space, cooling could probably be worked around.

If they go with a server style mentality, and have the cooling run the length of the chassis, the orientation of the CPU/GPU, vertical/horizontal doesn't matter... though to get that air flow, the fans need to move a lot of air, might be noisy.
 

Topweasel

Diamond Member
Oct 19, 2000
5,436
1,655
136
I thought about that, but things like the distance to system memory, and trace lengths become a problem (or troublesome), the PCI slot has pretty good retention capabilities as you suggest GPU's get heavy, but GPU's also have their memory embedded, so trace lengths can be evenly spread out on the GPU PCB. The closer the memory the better for processors, also even trace lengths makes a big difference for timing, currently, CPU's get memory on both sides which helps to even out the distances, throw it in a slot, and there is only one way to go, farther and farther away from the CPU slot.

That's going to be a problem for all "cart" based CPU solutions. They came from a time that the Chipset did all the dirty work and the CPU just got information and processed it. But back to this solution. It would be assumed that this is a compute module that would have its own memory and would use PCIe 5.0 or some other high speed interconnect to a main board that would basically act like the IO chip on AMD's system. It would limit the max memory limit of a system like this. But if you could get like 64GB on board and use some kind of Xpoint/Optane solution for increased memory size for systems that require it you might have a solution. But this is basically paper napkin theory writing for what is basically a junk idea that doesn't have foundation in a usable product and is pretty far from the idea of a real cartridge based CPU that was first mentioned. This isn't to much different then a Razor Shelf. We already have this.
 
  • Like
Reactions: Veradun and joesiv
Apr 20, 2008
10,161
984
126

I made the case back in 2017. It does make sense now to go back.
 
  • Like
Reactions: moonbogg

Cogman

Lifer
Sep 19, 2000
10,277
125
106
No. Here is why not IMO.

First off, most pins on a modern CPU aren't for data transmission, they are for power and clock. Why, you might ask? Because having a stable voltage for a CPU is critical for correct operation. Having a consistent clock across the chip is also critical. When you turn the CPU into a socket, both those things become a lot harder to deal with. If the socket isn't making a good connection, you can have voltage variances (pins are easier to make strong connections with). All the sudden, for consistent clock cycles, you need to start worrying about trace length to the CPU, YUCK! And further, to get the amount of power and clock lines into the CPU that you need, you end up needing fine traces or a big socket. In the case of fine traces, you now need to worry about clock cross talk (yuck). In the case of a bigger socket you now need to worry even more about trace length to the CPU.

You can alleviate the clock problem by putting the clock on the socket itself, but then you have to synchronize that clock with the rest of the system (Yuck!).

But the next problem is that your traces get longer. We are already at the point where the speed of light is messing with the speed of which we can get data out of ram. By adding extra long data traces, that will only increase latency between the CPU and all devices. You could move to push ram onto the socket into more of a SoC type solution.. but then I really wouldn't like that. Maybe you add ram onto the socket and have it act as another cache layer (L4 cache if you will) That would definitely increase cost and complexity.

Finally, the packaging would simply be more expensive for AMD and Intel. Nobody really wants that. The current packaging solution is just about as cheap as you can get. The only thing cheaper would be integrated CPUs (like with laptops).
 
  • Like
Reactions: misuspita

whm1974

Diamond Member
Jul 24, 2016
9,460
1,570
96

I made the case back in 2017. It does make sense now to go back.
Why? I don't see any advantages to going back to slot CPU cartridges at all. They are more expensive then socket processors, and have the mentioned HSF mounting problems. In addiction to the blocking to decreasing space on the motherboard for everything else.

Wither you like it or not, sockets are here to stay.