Building The Linux Kernel In 60 Seconds

[phoronix]

Phoronix: Building The Linux Kernel In 60 Seconds

In less than one minute, it's now possible to build the Linux kernel from source on a desktop...

http://www.phoronix.com/vr.php?view=MTAyNjU

[bnolsen]

agreed!

This past week I built a 3930X matx system with 32GB ram for a demo for work. We do high performance number crunching (double precision and 64bit integer).

This new system runs circles around our dual quad i7 xeons. On cpu intensive parallel tasks this is almost 2x faster per core ... btween 40% and 50% faster overall. For mixed loads (combined cpu and IO) it was on about 25% faster overall, although it may well be some of that could be IO bottlenecking.

The past few generational jumps by intel has been amazing. The jump from core2 quad to nehalem was just as dramatic.

Btw I really don't like these new radiators. They make installation that much harder. When installed it blocks access to one stick of ram, and worse yet all boards I've seen put the 8 pin power connector so it sits below the radiator and the power cable gets pinched between the radiator and that ram stick. (looks like that board doesn't have the power plug problem *but* it looks like it sacrifices mosfet cooling)

[olbi]

Amazing. But price for it isn't so cool :P I know that ppl who need power buy this CPU, but other should be grateful for Core i5 or Core i7

[chuckula]

Halt Di Klappe

in my point of view this is only marketing bullshit.

Every time you post here you make it obvious that English is not a language that you use with great facility. Consequently, I'll use some German for you: HALT DIE KLAPPE.

You contort yourself into a bretzen to defend the disaster that is Bulldozer, where AMD spent almost four years screaming MOAR COARS only to release a completely disappointing CPU. The server versions are still often slower than 2010 era Westmeres using less than half the cores! Even if you slap in 32 "cores" you'll only beat a 12 core Intel system that is almost 2 years old by about 20% in some benchmarks if you are lucky.. oh and those faster AMD systems cost more than twice the equivalent Intel setup (see here: http://arstechnica.com/business/news...atastrophe.ars)

Here's an English word for you: hypocrisy. Go look it up and do some soul searching about how it applies to your posting history for AMD followed by you turning around and accusing Intel of "marketing bullshit"

Further, you just come out with a blank assertion than the solution you put forth will somehow be faster than an Intel based setup. Let's ignore the fact that a 3930K would be much less expensive and still have nearly the same performance, meaning that an Intel based system could be much cheaper than your inflated price. Show us an actual benchmark run using your supposedly superior system. Slapping an extra 8 cores on and cutting the clockspeed by 50% from the desktop CPUs will not magically turn that chip into a winner.

[crazycheese]

in my point of view this is only marketing bullshit.

you can do the same in a singlesocket solution and much cheaper with AMD

Supermicro Motherboard H8SGL-F(230?)+AMD Opteron 6272 (470?)

this solution is much cheaper and you can put more RAM into it and it beats the intel solution in performance.

the intel solution is only better in single threated tasks.

Someone having similar Opteron system to what you mentioned should answer and confirm that.

It does not beat in energy efficiency for sure though.

[bnolsen]

in my point of view this is only marketing bullshit.

you can do the same in a singlesocket solution and much cheaper with AMD

Supermicro Motherboard H8SGL-F(230€)+AMD Opteron 6272 (470€)

this solution is much cheaper and you can put more RAM into it and it beats the intel solution in performance.

the intel solution is only better in single threated tasks.

Note: I'm working with a WAY cheaper 3930X, not a 3960X!

I have direct access to a quad 6168 system built last summer with 64GB ram. My experience compared with dual E5530's run side by side benchmarks highly threaded high precision geometric workflow showed the intels per core 1.5x faster than the amds (this is with hyperthreading turned ON).

At the time it was encouraging to see the amds scale robustly to so many cores. The 48 core amd beat the intel on 2 of 3 tests, one test having a section that was single core only, that made the intel faster.

Now, this 3930X thrashed the above dual intel system, being almost 2x faster per core in 2 of 3 tests. The quad 6168 still on mixed I/O load, the single socket 3930x wins on parallel/single mix, they both are close on the pure parallel cpu task.

Note the two server systems have access to ~900MB/s sustained storage (tekram pcie raid6 hardware) whereas the 3930X only has a 120MB/s sata disk. Opening up IO on the 3930X might change the game.

Btw I will concede IO throughput to the amd (the board has dual intel gigabits). The bus seems to handle more throughput (both network and disk) more smoothly than the dual E5530 system. I'm sorry I can't quantify this, it's not something on our critical path.

[leeenux]

i don't care about how many cores do you use. i only care for speed per money. and speed per watt energy usage.

less clock speed means in fact a higher energy efficiency!
more cores means more energy efficiency!

Intel's mindless fans tend to ignore these aspects except when it suits their current argument. Heck, go look at the mindless drivel on Anandtech, they've released their 2nd consecutive Bulldozer Opteron article where they've whined about how it's too hard to make workloads scale past 8 cores, so their just going to use Cinnebench and declare Xeon the CPU to buy, even though less than one percent of servers are purchased for moderately threaded FP heavy loads.

It's starting to look like Intel is buying their reviews from Phoronix too, read the first paragraph:

Besides finishing up the Phoronix Test Suite 3.6-Arendal release this weekend, on Saturday I began running some new Intel CPU benchmarks. In building the Linux 3.1 kernel for x86_64 in a default configuration (make defconfig), I've now managed to trim down the compile time to less than sixty seconds on a single-socket desktop system. Similar speeds can be achieved out of multi-socket servers and other configurations, but this is the first time I'm seeing such kernel build speeds out of a single processor -- the AMD FX-8150 Bulldozer doesn't come close.

OK, that's fine, except you aren't providing any reference of how much faster than Bulldozer it is, but just using strong language to suggest Bulldozer is completely inferior, without offering any actual facts to back it up. It sounds to me like you've already benchmarked both on similar configurations, bar graphs and hard numbers or STFU. I have an FX-8120, I'll gladly do a timed kernel compiling if you'll provide the exact parameters you're using to compile it on the Intel system.

[NoEffex]

Intel's mindless fans tend to ignore these aspects except when it suits their current argument. Heck, go look at the mindless drivel on Anandtech, they've released their 2nd consecutive Bulldozer Opteron article where they've whined about how it's too hard to make workloads scale past 8 cores, so their just going to use Cinnebench and declare Xeon the CPU to buy, even though less than one percent of servers are purchased for moderately threaded FP heavy loads.

It's starting to look like Intel is buying their reviews from Phoronix too, read the first paragraph:



OK, that's fine, except you aren't providing any reference of how much faster than Bulldozer it is, but just using strong language to suggest Bulldozer is completely inferior, without offering any actual facts to back it up. It sounds to me like you've already benchmarked both on similar configurations, bar graphs and hard numbers or STFU. I have an FX-8120, I'll gladly do a timed kernel compiling if you'll provide the exact parameters you're using to compile it on the Intel system.

You can typically do math to gauge how much worse it is. Bulldozer gets stomped by the mid-lower-end i7's/i5's and since this stomps those then it's going to completely obliterate bulldozer.

I had bought a bulldozer-based system because bulldozer was supposed to be actually a run for having a lot of performance. The thing sucked. The "8-core" CPU is pretty much a hyper-threaded quad-core but slightly worse. It is really not worth buying. Bulldozer is legitimately a disaster and the only things AMD has going for them are 1. server CPUs (Heavily threaded applications such as databases thrive on high-core opteron systems, like 32-core or 48-core) 2. graphics cards (The ATI cards are similar in performance to the NVIDIA cards in that the highest-end AMD card MIGHT be better than the highest-end NVIDIA though that could just be false). Desktop CPUs is not their thing and I heavily regret ever purchasing one.

When the sandy bridge e came out I saw an opportunity and bought the 3930k (not paying an extra arm or leg for slightly better performance). It is better in every aspect. The bulldozer I had basically could not handle anything I run (IntelliJ being chief among those applications) and it just was not pretty at all, let alone play any games. The i7 can handle several vms, 3-4 intellij windows, and still not show any signs of slowing down. It's significantly better and IMO more worth the money. A couple hundred dollars more (Or maybe even less than that) for well over twice the power is worth it in a heartbeat.

[bnolsen]

You came in here very confrontationally spouting things that reviewers all disagree with, and that I've personally verified some of the results of these reviewers myself.

In usd terms the 3930x is 599 and the 6272 is 539. Not a big difference, these new intels are still at introductory prices.

And don't ever use supermicro boards. ever, esp with linux. I've had bad experiences with all 4 models (multi socket). They are bad about fixing crap bioses, etc. Tyan is the way to go, at this level they generally make a superior board and try to make sure it works.

Every single benchmark I've seen shows the 627x performing only slightly better than the 61xx series. Are they all being paid off to make amd look bad??? And the xeons they test against are still last gen intel tech, and they beat the amds.

Core for core amd can't beat intel. AMD may scale slightly better than intel but they can't scale well enough to make up for the really crappy throughput.

[haplo602]

Someone having similar Opteron system to what you mentioned should answer and confirm that.

It does not beat in energy efficiency for sure though.

The funny thing is, I skimmed over the 2 BD benchmark article graphs (original one and compiler perf comparison). Timed kernel compilation is not there (at least not on the normal one, maybe it is there for paying users).