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

121–130 of 144 posts

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

#121

Post by old-gregg »

hinkley wrote: post #119
Xeoncross wrote: post #118

Earlier quoted context omitted.

Solution: Keep all CPU's active. go build main.go & go build main.go & go build main.go & go build main.go & go build main.go & go build main.go & /s

I get that you're being funny, but there's a valid question in there. Does go not support parallel build jobs? If so then he should be setting the count to six instead.

The bigger question, at least for me, is either high-core machines improve your desktop experience if you're mostly running low-threaded workloads, for example a web browser.

One humble piece of evidence I gathered says that no, more cores make your low-threaded applications run slower because the Linux scheduler keeps placing them onto "cold" cores running on idle frequency, and once they "spin up" the thread gets moved to another cold core.

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

#123

Post by spatular »

old-gregg wrote: post #117

Before you get excited about unleashing the power of 16 cores on your desktop... I have recently upgraded from the 4-core to 6-core i7 with roughly the same frequency/turbo and noticed that compile times have actually increased. Disappointing, I thought, especially because in addition to extra two cores I now sport a quad-channel DDR RAM. Well, turns out that compiling my Go projects does not generate enough "hot" th…

You can also lower switching threshold, in my experience it helps:

echo 50 >/sys/devices/system/cpu/cpufreq/ondemand/up_threshold

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

#124

Post by mattkrause »

old-gregg wrote: post #117

Before you get excited about unleashing the power of 16 cores on your desktop... I have recently upgraded from the 4-core to 6-core i7 with roughly the same frequency/turbo and noticed that compile times have actually increased. Disappointing, I thought, especially because in addition to extra two cores I now sport a quad-channel DDR RAM. Well, turns out that compiling my Go projects does not generate enough "hot" th…

Perhaps you could try setting the process affinity manually with taskset or numactl?

Clearly it'd be nicer if the scheduler could figure this out automatically, but it might do in a pinch.

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

#125

Post by GameOfTrolls »

old-gregg wrote: post #117

Before you get excited about unleashing the power of 16 cores on your desktop... I have recently upgraded from the 4-core to 6-core i7 with roughly the same frequency/turbo and noticed that compile times have actually increased. Disappointing, I thought, especially because in addition to extra two cores I now sport a quad-channel DDR RAM. Well, turns out that compiling my Go projects does not generate enough "hot" th…

So you're saying that software needs to be optimized for multi-core usage?

Stop the presses. Have you submitted your work to Dr.Dobb's Journal yet? When can us mere plebs get in on a round of funding your startup?

kys.

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

#126

Post by mikhailt »

zyx321 wrote: post #94
BatFastard wrote: post #58

Earlier quoted context omitted.

Interesting, was watching the video on the new MS Surface Laptop at the end, I saw 4 gig of Ram and shut down the page...

It's designed to run UWP / Store apps exclusively. In the default configuration, it will not run win32 programs at all.

Just to be clear, Win32 apps submitted to the Windows Store will run.

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

#127

Post by mikhailt »

beagle3 wrote: post #106
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…

That's why an ARM64 Windows desktop is unlikely (unless microsofts x86 emulator is magical). But ARM64 chrome books or android books are practical; I have 5 year old ARM 32 chrome book and it is still useful for browsing and emailing.

It's already happening, Microsoft's using its WOW64 subsystem to emulate Win32 calls on top of native Windows 10 for ARM.

Source: https://channel9.msdn.com/events/Build/2017/P4171

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

#128

Post by floatboth »

shaklee3 wrote: post #45
floatboth wrote: post #12

Earlier quoted context omitted.

A GPU can use 16x 3.0 lanes, theoretically. In practice 8 lanes is enough, people have measured performance between 16x and 8x, it's almost the same. SSDs use 4x, if you do buy an NVMe drive that is. (In my country, these things are TWICE as expensive as SATA SSDs lol)

Maybe for games, but for cuda this is certainly not the case. Double the lanes is double the host to device bandwidth.

"CUDA" is not a particular application. You could very well be doing something with CUDA that does extremely little data transfer to/from the host, but a lot of computation on the GPU. Like mining :D

Also, host to device can be bottlenecked by memory speed on either side.

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

#129

Post by maksimum »

drewg123 wrote: post #108
hishnash wrote: post #69

Earlier quoted context omitted.

Gamers might the most vocal high-end PC users out there but they are not the people with the most money. The threadripper is not targeted at gamers. It is targeted at people who make money from using their machines, these people can, therefore, afford to upgrade. Consider a user who does video work. (most modern films and tv is filmed in 8k know) if this is in Raw format as it comes off the camera for a 90 min episod…

I do a lot of compiles of a codebase which is pretty big (Netflix version of FreeBSD). A full recompile takes me ~1 hr on my box (Xeon E5-2630 v3, 8 cores, 16 threads, 32GB DDR4). Most of this is a parallel make that is 100% CPU bound. So I'm probably going to update to thread ripper when it comes out. Doubling my core count should come close to halving my compile times. I probably will do storage via a ZFS array of…

> A full recompile takes me ~1 hr

Could you explain why full recompiles on a single machine are necessary? Are you testing different compiler settings?

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

#130

Post by beagle3 »

mikhailt wrote: post #127
beagle3 wrote: post #106

Earlier quoted context omitted.

That's why an ARM64 Windows desktop is unlikely (unless microsofts x86 emulator is magical). But ARM64 chrome books or android books are practical; I have 5 year old ARM 32 chrome book and it is still useful for browsing and emailing.

It's already happening, Microsoft's using its WOW64 subsystem to emulate Win32 calls on top of native Windows 10 for ARM. Source: https://channel9.msdn.com/events/Build/2017/P4171

When I wrote "magical" I meant "sufficiently performant". Which I will only believe when I see in action.
Post reply on HN