Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

All AMD ThreadRipper CPU's will have 64 PCIe Lanes

anandtech.com

Post reply on HN
Print view View original

31–40 of 144 posts

Re: All AMD ThreadRipper CPU's will have 64 PCIe Lanes

#31

Post by adrianN »

jrimbault wrote: post #23
adrianN wrote: post #18

Earlier quoted context omitted.

My Raspberry Pi does almost all things that I would like to do with a computer. I could definitely imagine buying an ARM based computer.

Because we have the sources for most of the software, it can be recompiled. Not true for Windows, especially old closed source software.

Apple switched processors several times. They provided an emulator for old programs, so they could run without the user noticing anything (except worse performance). You could probably do the same for ARM.

Re: All AMD ThreadRipper CPU's will have 64 PCIe Lanes

#32

Post by redtuesday »

elorant wrote: post #17
nicktelford wrote: post #5

The huge number of PCIe lanes is pretty exciting to me. For consumers, the limiting factor in PCs these days tends to be I/O. With this, you could have 4 PCIe M2 SSDs (that's a bit under 16Gbps bandwidth, assuming RAID0), and still have 44 lanes (the number in the top-end Intel CPU) to spare for GPUs and other peripherals.

Realistically speaking, what's the use case for something like that? And secondly, how much a mobo like that would cost?

Intel X99 mainboards (socket 2011-3) ranged from 200-600 us dollars. AMDs offerings typically go for less than Intels, but Threadripper has an gigantic 4094 pin LGA socket, so I guess this will negate the typical AMD price advantage regarding motherboards and we will see prices in the 200-600 us dollar range as well.

EDIT - and since Threadripper will have 20 more PCIe lanes compared to Intels i9 series, motherboard makers will hopefully take advantage of this to offer more IO, which probably also increases the price.

Re: All AMD ThreadRipper CPU's will have 64 PCIe Lanes

#34

Post by masklinn »

vbezhenar wrote: post #8
nicktelford wrote: post #5

The huge number of PCIe lanes is pretty exciting to me. For consumers, the limiting factor in PCs these days tends to be I/O. With this, you could have 4 PCIe M2 SSDs (that's a bit under 16Gbps bandwidth, assuming RAID0), and still have 44 lanes (the number in the top-end Intel CPU) to spare for GPUs and other peripherals.

I don't fully understand. So 1 SSD need 1 lane. How many lanes need something like NVisia 1080Ti? What use of the remaining lanes? If I want to build something like 1 CPU/1 GPU/1 SSD configuration without additional peripherals, 62 lanes will be effectively wasted?

> I don't fully understand. So 1 SSD need 1 lane.

M.2 SSDs supports up to PCIe 4x, though there are also PCIe card formats which support 8x and 16x: http://blog.seagate.com/intelligent/seagate-demonstrates-fas...

> How many lanes need something like NVisia 1080Ti?

16 per (technically they need 4 per — that's what TB3 gives — or possibly less, but they'll use 16 if they can)

> If I want to build something like 1 CPU/1 GPU/1 SSD configuration without additional peripherals, 62 lanes will be effectively wasted?

You're at 24 lanes for the GPU (16), SSD (4) and southbridge (4). And unused lanes are no more wasted than unused USB slots, seriously.

But remember that Threadripper is an enthusiast's platform, people who might have 2~4 SSDs (8~16 lanes) or SLI GPUs (2x16).

Re: All AMD ThreadRipper CPU's will have 64 PCIe Lanes

#35

Post by Quequau »

wiredfool wrote: post #3

64 lanes would be good for something like 4-6 GPUs @ 8 lanes per, and some high speed storage/networking? Anxiously awaiting Intel's "I am not left handed either" response.

Intel already supports a significant number of PCIe 3.0 lanes with their Xeon E5 chips. If I recall supermicro sells a dual socket Xeon board with 10, 8xPCIe 3.0 sockets.

Re: All AMD ThreadRipper CPU's will have 64 PCIe Lanes

#36

Post by DaiPlusPlus »

nicktelford wrote: post #14
xbmcuser wrote: post #6

I was of the opinion that Intel will only face real competition when arm reaches parity with desktop class cpu. Nice to see AMD rise back up again. Hope Intel is not able to strong arm the oems like last time when AMD had design wins but Intel using anti competitive practice stalled its growth till it was able to catch-up and later go ahead.

I doubt we'll be seeing desktop ARM systems any time soon, especially high end workstation or enthusiast systems. The problem is that the ARM64 architecture is completely different from x86-64, which almost every desktop application is compiled for exclusively . Remember when Microsoft launched an ARM Surface? It had a special version of Windows, and almost nothing would run on it, as very few applications are compil…

> It had a special version of Windows, and almost nothing would run on it, as very few applications are compiled for Windows on ARM.

That wasn't the reason. It was because Windows RT was artificially locked-down to only allow sandboxed "apps" from the Windows App Store to run. There's not much of a use-case for only glorified web-service clients on a machine with a desktop UI regardless of the ISA they're compiled for.

Had MS allowed arbitrary ARM code to run on the Surface RT, free of sandbox restrictions, I think the platform would still be alive today - probably as a successor to Windows CE for kiosk applications.

(Though there is now finally a non-locked-down version of Windows 10 for ARM... so let's see...)

Re: All AMD ThreadRipper CPU's will have 64 PCIe Lanes

#37

Post by latch »

thedjinn wrote: post #28
nicktelford wrote: post #14

Earlier quoted context omitted.

I doubt we'll be seeing desktop ARM systems any time soon, especially high end workstation or enthusiast systems. The problem is that the ARM64 architecture is completely different from x86-64, which almost every desktop application is compiled for exclusively . Remember when Microsoft launched an ARM Surface? It had a special version of Windows, and almost nothing would run on it, as very few applications are compil…

I wonder if it's possible to use some kind of translator, like Apple did with Rosetta to run PowerPC code on x86. This is of course not interesting for the Surface but it might be a nice solution for a desktop transition to ARM.

Like: https://en.wikipedia.org/wiki/Transmeta_Crusoe ?

Re: All AMD ThreadRipper CPU's will have 64 PCIe Lanes

#38

Post by swsieber »

adrianN wrote: post #31
jrimbault wrote: post #23

Earlier quoted context omitted.

Because we have the sources for most of the software, it can be recompiled. Not true for Windows, especially old closed source software.

Apple switched processors several times. They provided an emulator for old programs, so they could run without the user noticing anything (except worse performance). You could probably do the same for ARM.

And Windows has done the same thing for ARM. I think it was posted here in the last month or so.

Edit: found it,

https://news.ycombinator.com/item?id=14318406

Re: All AMD ThreadRipper CPU's will have 64 PCIe Lanes

#39

Post by wlesieutre »

nicktelford wrote: post #14
xbmcuser wrote: post #6

I was of the opinion that Intel will only face real competition when arm reaches parity with desktop class cpu. Nice to see AMD rise back up again. Hope Intel is not able to strong arm the oems like last time when AMD had design wins but Intel using anti competitive practice stalled its growth till it was able to catch-up and later go ahead.

I doubt we'll be seeing desktop ARM systems any time soon, especially high end workstation or enthusiast systems. The problem is that the ARM64 architecture is completely different from x86-64, which almost every desktop application is compiled for exclusively . Remember when Microsoft launched an ARM Surface? It had a special version of Windows, and almost nothing would run on it, as very few applications are compil…

Funny you say that, MS just announced they're taking another stab at it.

https://www.theverge.com/2017/5/31/15711334/microsoft-window...

Re: All AMD ThreadRipper CPU's will have 64 PCIe Lanes

#40

Post by nspattak »

altcognito wrote: post #19

Do the extra PCIe lanes help the GPUs for deep learning applications, or is the data typically pretty localized on the GPU board?

Extra PCIe lanes mean more peripherals might be available on a system, eg more GPUs and/or hard disks and/or other IO devices.
Post reply on HN