News Optical PCIe 7.0 connection hits a blazing 128 GT/s

[Admin]

Cadence teases proprietary optical connectivity solution for PCIe 7.0 at 128 GT/s.

Optical PCIe 7.0 connection hits a blazing 128 GT/s : Read more

 

[bit_user]

Cool, so now PCs won't pose only an electrical hazard, but they could also blind you?
: O

Seriously, given the power & wavelengths they're using, what's the potential for eye injury from this class of fiber optics?

 

[toffty]

Cool, so now PCs won't pose only an electrical hazard, but they could also blind you?
: O

Seriously, given the power & wavelengths they're using, what's the potential for eye injury from this class of fiber optics?

I'm going to assume there is a negotiation between host and client cards before lines are saturated, in which case 0 chance anyone being able to harm their eyes if they stair at the ends of these cabled and one side is plugged in.

This probably also doesn't use optical light. Fiber optics are usually at 850, 1300, and 1550 nm wavelengths which are above optical light.

 

[bit_user]

I'm going to assume there is a negotiation between host and client cards before lines are saturated, in which case 0 chance anyone being able to harm their eyes if they stare at the ends of these cabled and one side is plugged in.

I hope you're right, or else PC repair techs better make sure they max out their personal injury insurance & employers better watch out for workmans comp claims!

This probably also doesn't use optical light. Fiber optics are usually at 850, 1300, and 1550 nm wavelengths which are above optical light.

That might reduce the potential for damage, but I'm sure even non-visible light can cause tissue damage and scarring, given sufficiently high intensity.

To me, it almost seems like a downside that you wouldn't be able to catch a glimpse of a light pattern on the ceiling, before you put your face in the beam path. Maybe they should add a low-intensity visible light component, not to carry information, but just as a visual warning, not unlike the way that "rotten egg" smell is added to natural gas.


Can anyone answer another question I have about this stuff? What's the potential for routing these optics through a PCB, rather than relying on cables? It would seem to be moving backwards, if we suddenly end up with cases full of fiber optic cables going to all the peripheral devices. Not only that, but you must also take care not to bend them too sharply. That would make for very unfriendly tech, IMO.

 

[kanewolf]

Cool, so now PCs won't pose only an electrical hazard, but they could also blind you?
: O

Seriously, given the power & wavelengths they're using, what's the potential for eye injury from this class of fiber optics?

Probably no different than any other single mode fiber laser.

 

[Dementoss]

Cool, so now PCs won't pose only an electrical hazard, but they could also blind you?
: O

Seriously, given the power & wavelengths they're using, what's the potential for eye injury from this class of fiber optics?

If you're stupid enough to point one into your eye, while it is carrying data, then you probably deserve what's coming to you...

 

[LabRat 891]

I can only imagine the crazy PHY interface, if backwards compatibility is to be retained.

Not really sure how this ever could make its way into the consumer market.
Copper or Optical (or both), a speck off dust, etc. is going to drastically effect the link.

 

[derekullo]

I'm going to assume there is a negotiation between host and client cards before lines are saturated, in which case 0 chance anyone being able to harm their eyes if they stair at the ends of these cabled and one side is plugged in.

This probably also doesn't use optical light. Fiber optics are usually at 850, 1300, and 1550 nm wavelengths which are above optical light.

To be fair it isn't just optical light that can blind you/be bad for your eyes.
At 850-1550 nm wavelengths that would be infrared light which is literally what we feel as heat.
The real question is what is the amplitude of said infrared beams.

 

[thestryker]

With PCIe 7.0 I could really see it launching for enterprise and not propagating down for a while if at all depending on the signaling issues. The large scale manufacturing costs associated with optical have got to be very high and I can't imagine client side ever seeing that and potentially not even workstation.

 

[LostFate]

Probably no different than any other single mode fiber laser.

I wouldn't expect them to use single mode for this, no one is going to negotiate a PCIe link over multiple KM. This will be a low power multimode setup with a power level low enough that you can look directly into the fiber without (much) risk of complication.

 

[kanewolf]

I wouldn't expect them to use single mode for this, no one is going to negotiate a PCIe link over multiple KM. This will be a low power multimode setup with a power level low enough that you can look directly into the fiber without (much) risk of complication.

The graphic in the article shows "SMF" which I interpreted as single mode fiber.

 

[DS426]

...

Can anyone answer another question I have about this stuff? What's the potential for routing these optics through a PCB, rather than relying on cables? It would seem to be moving backwards, if we suddenly end up with cases full of fiber optic cables going to all the peripheral devices. Not only that, but you must also take care not to bend them too sharply. That would make for very unfriendly tech, IMO.

Mobo vendors have already released models that use direct-attach hardware for PCI-E power cable routing and such, so I imagine optics will also have plenty of through-PCB and rigid structure options -- especially because, as you pointed out, folks bending fiber optic cables too tightly will quickly prove to be a problem.

PCIe 7.0 will probably just be copper media for internal hardware in PC's; optical PCIe would mostly only live in the datacenter world, namely where distance is the challenge between systems and hungry PCIe 7.0 devices.

 

[coromonadalix]

normally single fiber optic cable dont cost a fortune, and i work in fiber optic related stuff 10-12cm of diameter roll is the limit before we see big deterioration

and i myself would love to see fiber connector on gpu mobo, mobo vs storage etc ... about time

dust is no problems if you dont play connect disconnect a lot, cleaners exists

already for years you have servers with fiber linked storages .... they go easily in the GB of bandwith

you have fiber to the home, they too can run at 1gb no stress

 

[jp7189]

If you're stupid enough to point one into your eye, while it is carrying data, then you probably deserve what's coming to you...

<sheepishly raises hand> that's how I make sure I've got the transmit and recieve lines plugged in to the right sides. You wouldn't believe how often fiber optic network issues come down to a pair getting crossed - esp on long runs. Maybe I'm wrong, but I've never been overly concerned about it.. that said my trick only works on 850 SR, I can't see 1120- 1350nm that's used in many LR deployments.

 

[bit_user]

If you're stupid enough to point one into your eye, while it is carrying data, then you probably deserve what's coming to you...

That might be fine for data centers and telcos, but not consumer products. A consumer product with serviceable components needs to be safe enough for children not to easily blind themselves.

 

[bit_user]

PCIe 7.0 will probably just be copper media for internal hardware in PC's; optical PCIe would mostly only live in the datacenter world, namely where distance is the challenge between systems and hungry PCIe 7.0 devices.

Not sure about that. Power is a rapidly growing problem, in the datacenter. Driving signals at PCIe 7.0 frequencies is sure to consume a lot of it. I think efficiency demands are going to push servers to integrate optical communication for internal I/O. At least, that's what companies have been saying.

 

[kanewolf]

That might be fine for data centers and telcos, but not consumer products. A consumer product with serviceable components needs to be safe enough for children not to easily blind themselves.

Single mode fiber is used in homes every day. It also has a warning sticker that says that eye damage can occur from invisible laser light. Most ISPs that provide fiber to home use single mode fiber. With that said, I don't believe that the PCIe 7.0 optical standard will be used anywhere beyond data centers. Costs will be too high for consumer use.

 

[Eximo]

On the safety side, if they build the optics into a connector, I don't see why it wouldn't work. Simply have self opening hatches (a nice wedge system to actuate a simple door would work), plus having some sense pin not connected until fully seated that prevents activation of the optics. Any unplugged cable or empty port would always be 'off'.

Desktop computers are going to get more and more niche as time goes on. I don't see why they wouldn't pass high costs onto enthusiastic consumers.

But they may just adopt a single board solution before that happens. CPU/GPU/Memory etc on a board with only external expansion necessary.

 

[jp7189]

That might be fine for data centers and telcos, but not consumer products. A consumer product with serviceable components needs to be safe enough for children not to easily blind themselves.

How bright does it really need to be to have an acceptable signal over a couple of inches?

For some reason I'm reminded of the star trek episode where they blind spock to remove a stingray thing from his back, and later decide they didn't need to use so bright a light after all.

 

[JayNor]

People were saying pcie5 would never make it to consumer chips, but it's already here.

Intel demoed a 4Tb optical chiplet this year... 64 channels of pcie5. 5 pJ/bit

 

[bit_user]

People were saying pcie5 would never make it to consumer chips, but it's already here.

I said it wouldn't happen for a long time (but I didn't say "never"). We got PCIe 4.0 in 2019 and I figured it'd be at least another 5 or 6 years before PCIe 5.0. It turns out I was sort of right. The biggest bandwidth consumers have traditionally been GPUs, and it looks like we won't have a consumer-oriented PCIe 5.0 GPU until at least 2025.

The one class of devices that has made the leap to PCIe 5.0 is SSDs. Ironcially, they aren't even properly-supported on Intel's first socket to feature PCIe 5.0 support. Sure, it can be made to work, but only by eating into the GPU's lane allocation.

Anyway, I feel somewhat vindicated. I also maintain that consumers don't really need PCIe 5.0 SSDs. Furthermore, the unwieldy heatsinks needed to keep some of them from throttling serve to underscore that the technology still isn't suited for the mainstream desktop platform.

Intel demoed a 4Tb optical chiplet this year... 64 channels of pcie5. 5 pJ/bit

That's cool. Do you want to make any predictions about if/when that might trickle down to consumers?