Christine: The World's Most Modular PC Design | Hacker News Reader

Christine: The World's Most Modular PC Design

razerzone.com

69 pointsbado59 comments

And this is what you get when you have a designer "imagine" what a PC would look like.

One of the nice things about the Mac Pro was that it was engineered to look good, as opposed to being "designed" to look good. (or not depending on your design tastes).

The modern PC has three things you have to manage when engineering a 'new' one, heat management from both the CPUs and the GPUs, signals management which can carry exceptionally high frequency signals across "busses" that act as transmission lines, and modularity.

This vision has the latter covered but introduces challenges that make keeping the other two in check probably impossible.

First (and perhaps foremost) a modern PC generates a lot of heat, and managing that heat is central to any design, like having a place to eat and a place to sleep are essential to any home design. The components will generate up to 1200W of heat (equivalent to a space heater or hair blow-dryer set on 'high') Which needs to be removed from the components fast enough to prevent them from reaching their critical temperature.

Next up, modern PCs use PCIe channels that clock at 5Ghz with low voltage differential signaling. These signals absolutely do not like going through connectors, so systems design need to minimize connector transitions to 1 or zero. The renders for this design seems to show components plugging into a back plane (minimum two connectors). Also at 5Ghz, the wavelength of these signals is 6 cm. That means that bits going through a wire that is longer than 6cm long start taking a "long time" (relative to the signal) to arrive. Now PCI has some mechanisms in the protocol to support this but it starts resulting in additional bus transactions and signals. That increases latency in a way that can turn your awesome GPU into something decidedly unawesome in terms of its performance. It isn't the GPUs fault, its getting the data too and from the CPU which is the problem. There is a reason the 'video' slot on PCs is the one right next to the processor :-). This problem is made even more manifest in memory systems. If you get a chance to look at the circuit board traces for memory DIMMs some time notice how some are really squiggly and some are straight. The trick is that they are all exactly the same length so that data that starts coming out of the chip arrives at the same time at the CPU and vice versa. Fully Buffered DIMMs (FBDIMMs) were a shot at 'fixing' this and few people bought them. The chips ran hotter and performance was worse, lose lose to most consumers.

All that said, I love the concept of modularity in a PC but at the moment I can't see how to get that without sacrificing a lot of performance. That is one of the reasons I was so impressed with Apple's effort. It gets a lot of things right.

Thank you for the insight on PC signaling.

You're right but ignoring cost I'd rather have something that looks like this than a Mac but there's no accounting for taste.

Regarding the PCIe comment: people are already using PCIe over cable as an expansion technology. We use lots of these, for example: http://www.onestopsystems.com/pcie_over_cable_3810.php

Works in a pinch although it's of course not going to have the same performance as a connectorless solution.

The point is that GPUs need to be physically near CPUs and RAM. You can do all kinds of external PCI expansion, including Thunderbolt cables several meters long, but you'd never attach a GPU to that other end.

People are running GPU's through thunderbolt on Macbook Airs. TB Actually has provisions for doing this, but they are severely underdeveloped.

That's not going to work well. Thunderbolt 2 is 20 Gbit/sec per connection, vs 128Gbit/sec for a 16x PCIe 3.0 slot.

As Wise points out the challenge is the bandwidth. You can run video cards over USB (see DisplayLink) but the update rates suffer from being constrained to 480 megabits/sec. So if you are paying a lot of money for something, you probably want commensurate performance.

ChuckMcM never mentioned cost at all... ChuckMcM explained very concisely why a PC design like this one COULD NOT EXIST

>You're right but ignoring cost I would prefer this over a Mac ... are you retarded?

Minor nitpicking. The wavelength of 5GHz signal on typical pcb is more like 3-4 cm. 6cm is assuming vacuum.

Would superconductors like Graphene or Carbon nanotubes allow for low latency signalling over longer distances?

No, its a speed of light issue. Signals can only propagate at the speed of light. The other issue with signals integrity is when a signal goes from one wire to another (like it does through a connector) it behaves a bit like light going through one lens into another. Most of the signal goes through, some of it gets reflected. The result is that instead of the signal being there or not being there (ideal) you end up with what folks think of as any 'eye' pattern (it looks a bit like an eye) which is the point where the signal is reliably all there, or all not there (in digital terms a 1 or 0). More reflections, more noise, the smaller the eye and the harder to reliably read the signal.

What Apple gets right, however, does not include modularity. Even the two GPU's are not identical twins. It's engineered as consumer electronics kit and to merchandise well under retail lights.

Reminds me of the BMW/Thermaltake collaboration from 2009. [0]

About 10 years ago I thought I wanted something like this.

Over the years I realized that my meticulously researched, highly expandable PC builds were rarely being expanded or upgraded. By the time I really wanted, much less needed something new there was little if anything worth salvaging.

This creation looks like a great way to spend even more on building/upgrading something that will be relatively big, slow, power hungry / hot in 3 years all the same.

0: http://dl.maximumpc.com/galleries/level10/l10beauty_lw.jpg

It is still alive and is pretty affordable now: http://thermaltakeusa.com/store/Product.aspx?C=1014&SC=1017&...

I will stick with my 900D though.

I know very few people that constantly upgrade their PC builds. The ones I do, upgrade for the sake of upgrading, not because the hardware is no longer able to keep up. For example, a new minor version of a video card comes out, and they drop $1000 (AU prices.. sigh) on upgrading the card. They might do that 2 times per year.

Most of my friends that build new PCs will do so every 3-5 years but, like you said, very little is salvageable. Non-OS storage being the main reusable component.

Well. I don't think sata is going anywhere at the very least. A chassis that supports expandable and swappable drives wouldn't go amiss.

Easily handled. SATA backplanes with modular drive slots are available and they fit into two or three or four 5.25" slots. Many enthusiast and server chassis already have the option for SATA backplanes as well.

As new PC technology evolves, Project Christine can evolve with it.

I read that and was taken back to the early 90's and Zeos PC's with CPU daughter cards that would protect my investment according to the glossy ads in Computer Shopper. Then there was the 487 socket for upgrading 486SX CPU's...by disabling the FPUless chip completely for the fully functional chip that should have been bought to begin with. Not that upgrades were ever done in significant numbers.

Of all the forms of premature optimization that occur in computing, non-standard upgradable hardware is among the most persistent.

This could be future of personal computing, however I am afraid that prohibitive cost would kill this. I would expect markup of $100-$150 per module, unless it is a very simple module that needs no cooling.

I am also not sure about cooling system. Little isolated compartment with nearly no holes for cooling beefy GPU or CPU with overclocking sounds a bit farfetched. They mention that each individual module would have it's own water cooling system. I think it is madness, much more practical and economical to transfer heat to the central tower and have large radiator with quiet fans there.

I built few custom rigs with water cooling, and whole point of water cooling is to transfer heat to the large radiator.

Hard to tell from the snazzy-but-vague renderings, but it looks like the intent is that every module has "in" and "out" bushings for water-cooling - the larger circles on either side of the electrical connections - with the large radiator in the bottom.

Would make "hot-plugging" more like "wet-plugging", one suspects...

They specifically mentioned self contained. While there are quick connect/disconnect water cooling connectors [1] those do leak a bit of liquid each use.

[1] http://www.xtremerigs.net/2013/07/02/2013-quick-disconnect-r...

I don't think it's actually self contained in the sense that the pump, fans, and radiator are also duplicated in every module. The would be a terrible waste of resources and the modules don't look like they have any room for the radiator (which would need to be exposed to the outside air to work).

I could see a pump being included in each module. In a modular design such as this the flow rate through each waterblock would need to be guaranteed with many different possible restrictions.

Additionally, if the pumps were run serially, it could add some redundancy to the cooling loop (especially benefitial since each unit's pump might not have the same MTBF)

The problem is that you have another hand in the middle. Intel makes the CPU and then razr wraps it up and sells it to you at a markup. Nvidia makes a gpu and then someone throws it on a board for razr and then razr wraps it up and sells it to you at a markup. Why wouldn't I just buy it on a pci-e board for less because there will be competitors. Anyone here can build a PC - it's not hard. But that is the problem that this is trying to solve - that it is hard.

"For more than 30 years, only the most hardcore enthusiasts were able to take advantage of PC customizability. Convoluted hardware made it insane for the average person: knowing what does what, what works with what, and how to connect the pieces."

So this is for the guy that wants to build his PC but not actually learn how to build it. It's like radio shack xmods - you can put a new engine in your car to make you feel like you're working on a car but you're not working on a car. You're working on a toy that looks like a car.

I wouldn't be surprised if many of the Christine modules were adapters for standard interfaces. The graphics card module could just be a mounting bracket with a PCIE slot and would fit most consumer GPU's. Similarly the S/Hdd modules would have power and sata attachment points. Those cases could be sold for less than 100$, and if Razor wouldn't make them I bet a third party would.

It's water cooled.

Wow, so much extra plastic. And are those compression connectors? Last I've seen those was in SGI computers, and they were notorious for failing to contact properly after being exposed for a while (rubbing with a soft cloth and alcohol seemed to do the trick though).

http://techpubs.sgi.com/library/dynaweb_docs/hdwr/SGI_EndUse...

Prior discussion:

https://news.ycombinator.com/item?id=7020090

Wow that's a pretty cool idea. The only issue I see going forward is whether or not the target market can afford the machine. The biggest pro of customizing a PC is the price for performance. I'm able to save quite a bit of $$$ by building my own machine. If the people who don't know how to customize their own machine are price insensitive they might as well go for something that requires no effort from them (e.g. Apple products, high end Dell/HP). If they are in fact price sensitive, I imagine a modular machine such as this would probably be out of their price point given that every part is specifically for this one type of PC.

The greatest advantage would be partially insulating the owner from platform changes, though that depends on how Razer implements the generic backplane in the tower.

For example, lately Intel has been changing CPU sockets and interface chipsets like I change my underwear. Upgrading the processor requires upgrading the motherboard, which may trigger a cascade upgrade of other parts. With this system, it may be possible to upgrade the CPU module (with all bridge chipsets included) without having to touch any other parts.

Also, you are no longer limited by what a motherboard vendor chooses to put on their board. No worrying about number of RAM slots, number of USB headers, how many of each type of PCIe slot they decide to include.

How well it works, and whether it justifies the price, remains to be seen.

>> Oculus: "[...] whether or not the target market can afford the machine."

> vonmoltke: "[...] though that depends on how Razer implements the generic backplane in the tower." [...] "Upgrading the processor requires upgrading the motherboard, which may trigger a cascade upgrade of other parts."

-----

The above are the two big items in my mind. Another potentially minor (or major) issue is how the seals for liquid cooling are made, their durability and reliability. I am assuming that the pump is located in the backplane and a quick connect/disconnect type connection is used for liquid connections.

It sure looks cool. It looks like Razer's designers and/or engineers really liked the look of Thermaltake's Level 10 and wanted to take it in a different direction. It could also be that these are OEM'ed from the same manufacturer.

It might be a better strategy to keep the module prices as low as possible and depend on frequent upgrades. I wouldn't mind committing to buying a new CPU/RAM module every year for a few years if the price was right.

This looks ridiculous. There is no way this design could be practical. You would be paying so much money just to encapsulate the same exact hardware components you have now inside of a plastic enclosure with a custom connection (and a controller for these connections, I'm assuming) and maybe unnecessary lights, colors, displays, etc. It is a proprietary trap.

Is it cool? Sure. But definitely not practical, efficient in any way, and it offers nothing more than aesthetics

I'm not sure how powerful it is which cannot tell from the looking. If it's the same powerful as the existing PC, I found All-in-one PC now is less than $500. So why do we need to have this nice-looking but still heavy machine? If it's super powerful, then a lot of engineering issues need to be taken care before how it looks like.

EDIT: I'm sorry, it's fully expandable. But sometimes the OS may not support the hardware expansion that much or become very low performance.

But anyway, I agree that we should make PC more customizable and more pluggable if it's doable.

Aren't most PCs modular? I can swap out the drives, expansion cards, memory, CPU, PSU, etc with standardized alternatives. There must be a better word for what this is.

No. A standardized PC is componebtized, but not really modular. You can't install RAM in a hard drive bay. This concept I see as more in line with traditional stereo systems....you can have many modules, and they all talk to each other via RCA cables.

Wow.. so in the video they plugin a CPU module. Doesn't the CPU need equal length parallel lines to the memory? Or is there a serial interface to memory nowadays?

Crazy idea, I hope it works. Would definitely freshen things up.

> Doesn't the CPU need equal length parallel lines to the memory?

Yes, but you could do that within the tower. Maintaining signal integrity through the connectors and/or propagation delay due to trace length would be the limiters.

> Or is there a serial interface to memory nowadays?

Not for PC level memory speeds.

> Wow.. so in the video they plugin a CPU module.

All the pictures show "CPU / RAM" in the same module, which makes a lot more sense to me as an EE.

> Yes, but you could do that within the tower. Maintaining signal integrity through the connectors and/or propagation delay due to trace length would be the limiters.

They can, and probably will, also use signal path equalizers.

That would be problematic at best given the clock speed and timing requirements of DDR3.

Maxim claims to have the technology for gigabit signal path equalizers: http://pdfserv.maximintegrated.com/en/an/AN996.pdf

I'm not seeing much connection between the IC being talked about in that App note and what would be needed to support DDR3.

Those are for a single serial signal - as the note indicates, mostly to replace very-short-run fiber optics. The challenges with routing DDR4 signals through connectors, etc are mostly about skew: if any signals are delayed by anything approaching 1/2 a clock period (~1cm on FR4 board) relative to each other, everything is fucked.

In addition, some digging on the MAX3800 referenced there indicates that it has a power consumption of 200 mW, and a "budgetary" price of $6.90 each. Since a typical DDR interface is 100+ signals, you're looking at spending 20+ W and the price of a midrange PC just on signal integrity, and that's assuming those chips even do what you want.

There are systems that work like Christine - but they are things like IBM mainframes that don't suffer from cost constraints.

Gigabit?

http://en.wikipedia.org/wiki/DDR3_SDRAM#JEDEC_standard_modul...

Try 17GB with little latency tolerance.

The inevitable consequence is that you will end up with monsters such as this http://www.d1.dion.ne.jp/~r_high/memorial/rocketship.html RiscPC with pods called "slices". I don't think that the board components were able to be sliced out, only peripherals.

What problem is this solving?

Lack of modularization?

Computers are extremely modular already. Plus, you don't need to buy a proprietary interconnect and case that only one vendor provides.

Modular, but not necessarily modular done right. I'd much rather plug and play than take the time to screw in each component and run wires around and slice my hands on heat sinks made out of razer blades. If that means we have to go with a proprietary mb... someones gotta take the first step. I don't see why this design paradigm couldn't use existing components, like the Level 10 should have been designed.

For quick and dirty modifications they may not be 'done right', but from an engineering standpoint there are MANY limitations preventing exotic designs like this from even being viable, not to mention the cost of proprietary modules.

Computers are increasingly unmodular and integrated; e.g. the proliferation and continued advancement of system-on-chip designs. Its only a matter of time until this starts hitting the high end, where we will see fewer and fewer parts that are not put together in a fab.

Keeping the design interns busy?

Making "desktop" computers not look like ass.

Razer has been trying to focus on making stylish PCs for the gaming community. This seems like a natural extension to their product line. Just look at the Blade laptop; it's an incredibly thin (and cool) gaming PC meant to compete head-to-head against the MacBook Pro.

I think it looks great, but my concern would be that of others who have already commented. Modularity is great, however in an increasingly un-modular world I'd be worried about sinking money into the platform only to have Razer release a new, incompatible version in the not-too-distant future.

Viewing that required moving my browser window to my 2560x1600 monitor and maximizing it. How utterly stupid. I'm coming to hate HTML5. Apparently rescaling is beyond it's capabilities or beyond those of the designer.

I'm guessing this is not an engineer designed product. I can see there will be a lot of technical issues to solve. Also the kind of connectors displayed can easily cost hundreds of dollars if they are custom made.

Gosh that's pretty.

This is the kind of ambition that I feel like is missing from PC's. So much of the PC business revolves around just cutting cost and not making something cutting edge. Somehow Razer has managed to build interesting hardware that pushes the limits in a world where everyone is stuffing off the shelf components into mostly generic cases.

Assuming that the 'tower' is the motherboard, it better cost the same as a module, or else this is a non-starter.

So concepts are now a thing here?

That one sucks btw; imagine this one: a small white aluminum cube with no visible openings that does everything you will ever need to do in your life, now that's cool.

Reply on news.ycombinator.com