News 800 Gigabit Ethernet Launched to Double Bandwidth of Current Standard

bit_user

Splendid
Ambassador
The specification announced Monday will be used by enterprises way before you're able to use it with your best desktop PC.
Desktop PCs will never use it. I wouldn't even call it an enterprise standard. This is truly just for datacenters.

Just try to wrap your mind around 800 Gigabits/sec - that's literally 100 GB/sec. PCIe 3.0 x16 is only about 16 GB/sec, per direction. So, just too feed that link, you'd have to go to PCIe 5.0 x32 or PCIe 6.0 x16. And I'll go out on a limb and say that you're not going to find PCIe 6.0 slots in a desktop PC, ever.
 

hotaru251

Honorable
Oct 30, 2014
497
52
10,940
40
Desktop PCs will never use it. I wouldn't even call it an enterprise standard. This is truly just for datacenters.

Just try to wrap your mind around 800 Gigabits/sec - that's literally 100 GB/sec. PCIe 3.0 x16 is only about 16 GB/sec, per direction. So, just too feed that link, you'd have to go to PCIe 5.0 x32 or PCIe 6.0 x16. And I'll go out on a limb and say that you're not going to find PCIe 6.0 slots in a desktop PC, ever.
iirc they stated pcie 4 will be a short lived standard and 5 should come much faster than 4 did.
 

drtweak

Illustrious
Desktop PCs will never use it. I wouldn't even call it an enterprise standard. This is truly just for datacenters.

Just try to wrap your mind around 800 Gigabits/sec - that's literally 100 GB/sec. PCIe 3.0 x16 is only about 16 GB/sec, per direction. So, just too feed that link, you'd have to go to PCIe 5.0 x32 or PCIe 6.0 x16. And I'll go out on a limb and say that you're not going to find PCIe 6.0 slots in a desktop PC, ever.

That is what some guy said in a A+ book that i read back in the 2000s (Was made in 96) during the time of the 133Mhz Pentium, just only in the terms from the mid 90s. Things like not needing any faster CPU's or more than 32MB of ram, etc. and look at us now lol. I remember the time of people saying HT is all you need. Don't need multiple cores. That is data center stuff. Now look. You can get the more powerful CPU in your home Rig.

Well at least on the PCIe part. Yea I don't see 800Gbps connections coming to the home PC for many decades if it ever does. I might see it in my life time. Maybe. XD
 

mattkiss

Reputable
Sep 22, 2016
35
6
4,535
0
That is what some guy said in a A+ book that i read back in the 2000s (Was made in 96) during the time of the 133Mhz Pentium, just only in the terms from the mid 90s. Things like not needing any faster CPU's or more than 32MB of ram, etc. and look at us now lol. I remember the time of people saying HT is all you need. Don't need multiple cores. That is data center stuff. Now look. You can get the more powerful CPU in your home Rig.

Well at least on the PCIe part. Yea I don't see 800Gbps connections coming to the home PC for many decades if it ever does. I might see it in my life time. Maybe. XD
Just curious, which book?
 
Reactions: bit_user

kanewolf

Titan
Moderator
Desktop PCs will never use it. I wouldn't even call it an enterprise standard. This is truly just for datacenters.

Just try to wrap your mind around 800 Gigabits/sec - that's literally 100 GB/sec. PCIe 3.0 x16 is only about 16 GB/sec, per direction. So, just too feed that link, you'd have to go to PCIe 5.0 x32 or PCIe 6.0 x16. And I'll go out on a limb and say that you're not going to find PCIe 6.0 slots in a desktop PC, ever.
800gb will be for inter-switch links with 100gb nics in large hosts.
 
Reactions: bit_user
Apr 8, 2020
1
0
10
0
Desktop PCs will never use it. I wouldn't even call it an enterprise standard. This is truly just for datacenters.

Just try to wrap your mind around 800 Gigabits/sec - that's literally 100 GB/sec. PCIe 3.0 x16 is only about 16 GB/sec, per direction. So, just too feed that link, you'd have to go to PCIe 5.0 x32 or PCIe 6.0 x16. And I'll go out on a limb and say that you're not going to find PCIe 6.0 slots in a desktop PC, ever.
What a lack of imagination.

PCIe 6.0 will be here in a few years and I'm sure 7.0 after that or some other standard.

I'm sure by end the decade we will see 1tbps standard and then 2tbps 5tbps, 10, 20, 40, etc. Will never end.

Will continue to get faster and faster. And before long, it will be petabits a second, but even that won't be fast enough one day, I better sooner than most can even imagine. Space my friends opens a whole new realm of possibilities and demands. We will see private space travel soon. Can you imagine a "zoom" call to mars. hello! One day it will be a group video chat between planets. Sick around and see. For all you know, light speed has already been broken, don't bet against it.

Seems so many people lack the imagination. I was born in 1984, the start of the "Internet". Little before that, around Jan 1 1983 TCP/IP came online, but not much there. Not real Internet. Didn't have web till 1990. Crazy how far it's come already.
 

kaalus

Distinguished
Apr 23, 2008
8
2
18,515
0
"640kB should be enough for everyone"?

Desktop PCs will never use it. I wouldn't even call it an enterprise standard. This is truly just for datacenters.

Just try to wrap your mind around 800 Gigabits/sec - that's literally 100 GB/sec. PCIe 3.0 x16 is only about 16 GB/sec, per direction. So, just too feed that link, you'd have to go to PCIe 5.0 x32 or PCIe 6.0 x16. And I'll go out on a limb and say that you're not going to find PCIe 6.0 slots in a desktop PC, ever.
 

bit_user

Splendid
Ambassador
iirc they stated pcie 4 will be a short lived standard and 5 should come much faster than 4 did.
Exactly who stated this?

I do not expect to see PCIe 5.0 slots in desktop PCs. Not in the foreseeable future, if ever. It adds significant cost and increases power consumption, yet the need for it is not there.

The motivation behind PCIe 5.0 is for use in servers.
 

bit_user

Splendid
Ambassador
That is what some guy said in a A+ book that i read back in the 2000s (Was made in 96) during the time of the 133Mhz Pentium, just only in the terms from the mid 90s. Things like not needing any faster CPU's or more than 32MB of ram, etc. and look at us now lol. I remember the time of people saying HT is all you need. Don't need multiple cores. That is data center stuff. Now look. You can get the more powerful CPU in your home Rig.
The arguments are completely different. The argument you cite is that home users had no need for it. While I think that's also true of 800 Gigabit Ethernet, my point was that there would be technical hurdles in rolling it out for home users.

Back in the mid-90's, CPUs were made on a 600 nm process node. Now, we're at (arguably) 7 nm. Because that's a linear distance and CPUs are (roughly) 2D, the density improvement is a factor somewhere on the order of 7000.

The situation we're in, today, is that Semiconductor process nodes are already nearing their limits. Also, there's only so fast you can send a signal over a network cable-length piece of copper. So, that pushes networking > 10 Gbps exclusively into the optical domain, which further increases costs.

Put another way, every trend stops at some point. If you go back and extrapolate CPU clock speeds from the 1990's, you'd probably expect us to be in the realm of 100 GHz, by now. But that didn't happen... because physics.

So, you can't just blindly extrapolate trends into the indefinite future. Not if you care about being right, at least. You have to look at what's underlying those trends and figure out if it has room to continue. Even then, sometimes you're blind-sided by an effect you didn't appreciate, like how Intel got burned by leakage current, with the Netburst architecture of their Pentium 4.

Finally, addressing the "need" side of the equation, just look at 10 Gigabit Ethernet. It's been around for > 15 years, yet there's never been strong enough demand for faster networking among consumers to drive adoption and achieve mass-market pricing. Yet, at least. That's one development I think we could actually see. Eventually.
 

bit_user

Splendid
Ambassador
What a lack of imagination.
It's not lacking imagination, it's looking at what the technology involves and being realistic about the potential for it ever to trickle down.

PCIe 6.0 will be here in a few years
This is a little closer to the realm of the plausible, but I also think you won't ever see desktop boards with PCIe 6.0 slots.

I'm sure by end the decade we will see 1tbps standard and then 2tbps 5tbps, 10, 20, 40, etc. Will never end.
See my point about trends, in the post above. If you want to be truly successful in business or investing, you will have to learn that every trend ends at some point. The real skill isn't simply extrapolating trends like a trained monkey, but rather predicting when they'll break down.

Space my friends opens a whole new realm of possibilities and demands. We will see private space travel soon. Can you imagine a "zoom" call to mars. hello! One day it will be a group video chat between planets. Sick around and see. For all you know, light speed has already been broken, don't bet against it.
And exactly does space travel have to do with networking speeds? You can't just break the laws of physics, because there's a market demand.


Seems so many people lack the imagination.
Oh, believe me, I've watched and read more than enough "sci fi" created by people whose imagination exceeded their grasp of physics. Anymore, it just seems infantile.

around Jan 1 1983 TCP/IP came online, but not much there. Not real Internet. Didn't have web till 1990. Crazy how far it's come already.
If you're going to give history lessons, at least try to be accurate.


There's actually an interesting segue, here, which I think could be informative. If you go back and look at some of the 1950's era sci fi, a lot of what they predict is based on the idea of limitless, low-cost energy. Things like flying cars and such. Also, a lot about space stations, bases on the Moon and Mars. If you think about the reasons why, it's that the technological trends they were living through were nuclear power driving down electricity costs and space suddenly becoming accessible.

However, those trends didn't continue. Or, at least, not at the pace they predicted. And yet, they barely conceived of things like smart phones or video games. The history of technology is one where a new technology is created and goes through a period of rapid improvement, until it plateaus. Sometimes, other technologies can enable new improvements, like how cheap microcontrollers enabled more efficient gasoline engines. But any trend in technological development doesn't and can't go on forever. And it's a really tricky business to look decades into the future and predict what will be the new technologies and trends of that era.
 

bit_user

Splendid
Ambassador
"640kB should be enough for everyone"?
It's funny how often this get repeated, but if you actually know anything about the context in which it was said, the CPU he was talking about could only address a total of about 1 MB. So, he was talking about 640 kB out of about 1024 kB, back when populating all of that would've been prohibitively expensive for most PC users.

I think he didn't imagine still being restricted to 640 kB on future CPUs, like the 80286, which could address 16 MB. Or the 80386, which could address 4 GB. At the time, Bill Gates probably never expected DOS still to be the main OS used on such machines.
 

hotaru251

Honorable
Oct 30, 2014
497
52
10,940
40
Exactly who stated this?
posted a year ago.
https://www.tomshardware.com/news/what-we-know-about-pcie4,39063.html
PCIe 4.0 May Be Short-Lived. PCIe 5.0 Is Almost Here
Because the PCIe 4.0 specification was finalized a few years late, we ended up with a situation where the PCIe 5.0 specification was finalized only a couple of years later.
The PCI-SIG group recently published the 0.9 version of the PCIe 5.0 specification, and we should see support for it in computing products as soon as next year. As usual, the PCIe 5.0 generation will double the bandwidth of PCIe 4.0 to 32GT/s or ~8GB/s per lane (bi-directional), for a total of 128GB/s for 16 lanes.
so becasue 4.0 was late it wont last as long as normally would.
 

bit_user

Splendid
Ambassador
Okay, fair enough. I didn't remember Paul taking such a strong line on it. But he didn't do his homework, in this case, and look at what PCIe 5.0 actually involves.

so becasue 4.0 was late it wont last as long as normally would.
The problem here is that PCIe 5.0 doubled the frequency of PCIe 4.0. When you get into such a high frequency range, the costs & power consumption really start to add up. It's one thing to use it for servers, which are already hot and expensive. It's another proposition to use it in desktop PCs.
So, coupled with the fact that PCIe 3.0 is really fast enough, for today's desktop PCs, it's really difficult to see the industry moving to 5.0. I'll grant that PCIe 4.0 has shown slight benefits, at the margins.

BTW, this is not a natural position for me to take. As far back as 2013, I've been arguing that we should have PCIe 4.0. My next PC will have it. So, I am a believer in faster connectivity, but when you really get into the details of what it involves and look at it realistically, I think you'd agree that there's no business case that justifies moving consumers to PCIe 5.0, in the foreseeable future.


Weirdly, 6.0 is probably a smaller step than 5.0, because it's built on signalling changes that don't increase the frequencies required. It still requires better signal integrity and more sophisticated signalling components, though. It would be interesting if we saw something like 6.0 signalling at 4.0 frequencies emerge as a new standard, in the next 5 to 10 years. But, that's the kind of timeframe we're probably looking at.
 
Last edited:

ASK THE COMMUNITY

TRENDING THREADS