4-Way Sun Fire V40z Reviewed

Hack Jandy writes "Anandtech has a pretty thorough analysis of Sun's V40z 4-way Opteron server that fits in a 3U. Among some of the more noteable benchmarks include a 2 minute, 30 second Linux 2.6.4 kernel compile! Who would have thought only a few years ago that Sun would be the new champion of Linux and AMD?"

Comments

Who says they are?

[IANAAC]

Who would have thought only a few years ago that Sun would be the new champion of Linux and AMD?"

They're doing what they have to do to survive.

If they had their way, it'd be Solaris/Sparc all the way.

Re:Who says they are?

[saleenS281]

you confused "survive" with "grow and maximize profit". As if Sun is going anywhere anytime soon. They're going to die just like novell, BSD, and Microsoft are.
Sun realizes that the opteron provides nearly the performance of their sparc at a cheaper price... why not bundle it up and make MORE money since the cream of the crop for them is service. And more systems sold==more people buying service contracts. And lord knows cheaper prices==more systems sold.

Re:Who says they are?

[Anonymous Coward]

nearly?

show me, dollar for dollar, which sparc out performs an opteron processor.

if you tell me "well use this sparc costing X dollars" and it's some mind numbing number...then you've failed it.

assuming that said opteron is appropriate for a given job, peformance-wise, there IS NO SPARC that could compete when cost is a consideration.

fewer and fewer computational tasks require OR CA N AFFORD 200 processor sparc boxes.

when your client base shrinks to a few first world governments and high fallutin companies.....it's time to branch out.

sparc cpu WILL go away. yea sure, eventually they will all go away...but I MEAN SOONER, rather then later.

in 5 years or less, we'll see the writing on the wall, in 8 years, sparc will be a memory.

Re:Who says they are?

[codeguy007]

Novell is dying? I think you will find that Novell is making a comeback from the dead more than dying.

Re:Who says they are?

[Bert64]

A Sparcstation-20, which can often be acquired for free nowadays, will outperform any opteron for the same price!

Re:Who says they are?

[Bert64]

Sparc was never really designed for raw performance, but if you consider the performance drop as you increase the load on a system, sparc holds up much better than most other architectures and this is what sparc is designed for. Sparc also scales very nicely to large numbers of processors and is well proven in this field.
Also, Opteron is much newer than sparc, a lot of businesses won't trust something that hasn't been around a few years and is well proven.

Re:Who says they are?

[Octorian]

OpenBSD/sparc64 isn't a valid measure of performance. In my attempts at using it, I've noticed that OpenBSD on an UltraSPARC-IIi is SIGNIFICANTLY slower than Solaris pretty much all around. The Netra I was trying to run it on was way too sluggish. But once I dumped it in favor of Solaris 9, the machine suddenly became a lot faster.
(not to say your CPU comparison isn't accurate, but that the differences are a lot less than you make them out to be.)

Now OpenBSD/sparc (32-bit), on the other hand, tends to work very well on its intended machines.

Re:Who says they are?

[Anonymous Coward]

Sparc was never really designed for raw performance, but if you consider the performance drop as you increase the load on a system, sparc holds up much better than most other architectures and this is what sparc is designed for

That's a lot of hand-waving baloney. Sparc's got a whole bunch of "features" that end up actually working against that kind of scalability, like register windows for one and a historically small TLB. Stability and performance under high loads is much more a function of the system than the cpu - i.e. the i/o infrastructure, the memory subsystem and the OS itself. All of which are mostly independent of the cpu - you can put a crappy system around a sparc chip and a kickass system around an Opteron or vice versa.

Re:Who says they are?

[recursiv]

For fuck's sake, let me explain this for you:

He said that Sun is going to die like Novell, Microsoft and BSD
As in, not at all. But you frequently find people saying they will die. So, to die the same way as Novell, MS, and BSD are is to not die at all.

All clear now?

Re:I suspected

[DesScorp]

Sparc isn't dead...Sun just realized that they can't keep up with Intel and IBM in the chip wars by themselves. They've teamed up with a Japanese company (Fujitsu?) for future Sparc development. Sparc will be for high-end customers only. They're positioning Opteron for the cheap end.

Re:Curious

[caino59]

its free marketing.

the purpose here is too move product after all and make their shareholders money.

im not surprised at all

Re:I smell ...

[otis wildflower]

Hell, what server admin runs Linux, the lowest common denominator of Unix and Unix-workalies, on a real server?

Take a trip to NYC, walk out of the Wall St. 4/5 station, pick a tall building, go up on the roof, unzip your fly, and take a piss. Inside the building you hit you will find a company that transacts hundreds of thousands of dollars of business per MINUTE.

On Linux.

Better be quick though, as there's TONS of jobs moving across the Hudson :/

Re:Just been wondering about this

[byronmiller]

OSX doesn't have the capacity yet to make use of "big iron" (ie large memory systems). Sun has a decade + of experience in midrange computing.

Plus the price for this sun box outdoes the price i imagine we will ever see from the likes of apple

Re:Solaris and AMD

[ArbitraryConstant]

I agree.

I wouldn't want Red Hat or Suse or Gentoo on a production server, but I'd be happy with FreeBSD or Debian.

But I'd also be happy to run Solaris though. It has features that Linux and the BSDs don't have. Doesn't make it better for everything, but it's certainly worth looking at.

What was so good about these dead systems?

[Ars-Fartsica]

Was it a modern open unix? You can have that on x86.

Was it high performance? x86 outperforms all of your examples on a per-CPU basis.

Was it incredible graphics? These geezers don't have access to modern gpus.

Was it rugged hardware? x86 boxes are now equipped as good or better than any of your examples.

I'm not sure what it is you got out of using these systems that represents a legit advantage.

Re:What was so good about these dead systems?

[SoupIsGood Food]

Was it high performance? x86 outperforms all of your examples on a per-CPU basis.

This is a recent phenomenon, and has more to do with the politics of monopoly and inept business strategy.

In their heydey, MiPS, Alpha and PA-RISC were neck-and-neck in terms of performance, because all were funded and developed by vibrant companies at the top of their game. Sun was slower, especially in the benchmarks, but had other advantages (like its unreal low-latency).

Then along came Rick Belluzzo, who set both HP and SGI on the Itanium/WindowsNT deathmarch, killing off R&D for all three of the top-tier RISC/Unix architectures... once HP bought Compaq, they destroyed the old DEC R&D machine, and the Alpha with it, mostly out of spite.

What would have happened if HP hadn't decided to burn its bridges for Itanium? What would have happened if SGI had hired a CEO who decided to keep them on the RISC/Unix track and to keep Mips rather than spin it off?

You would see a top teir of premium processors, and a second tier of processors x86 could almost compete with. The way it was in '97, before "Merced" and "NT" were going to be the future of technical computing.

Was it incredible graphics? These geezers don't have access to modern gpus.

Modern SGI workstations, while laboring under an antiquated processor, have GPU subsystems you gamerboys can only have fond wet-dreams about.

Even still, past history shows that with a viable high-performance oriented platform, high performance innovation takes place that takes a few years to filter down to the commodity platforms: SCSI, Fiberchannel, crossbar connections for subsystems, wide datapath expansion cards (DEC's 64bit PCI comes immediately to mind), and GPU subsystems like anything from SGI or HP's Visualize.

Commodity gear has caught up, only because of Moore's law. The vendors essentially gave up their cutting-edge workstation and server markets to push their commodity systems, thinking they would offer higher margins and a wider customer base.

Instead, Dell took everything, slashing margins and eroding everyone else's share of the x86 pie.

Now Sun is making the same mistakes.

Understandable, though, as their SPARC R&D has been a complete mess. The Fujitsu SPARC chips are kinda sexy, but getting long in the tooth.

Opteron is a last-gasp stopmeasure for Sun. It will probably do little except irritate their longterm Solaris/SPARC customers.

SoupIsGood Food

Re:4-year-old dupe :)

[Eil]

Has the kernel really grown THAT much?

Yes and no... Using kernel compile times as a benchmark is categorically useless you quote the exact config file in the analysis.

A few weeks ago, I tried to compile a GNU/Debian Linux 2.6.x kernel on a Pentium III using the default kitchen-sink config. After about an hour and a half of just sitting there waiting for the damn thing to finish (this was on-site maintenance of a critical mail server), I halted the build and took my chances at configuring it by hand, hoping that I wouldn't forget some option that caused the machine not to boot. After paring it down, the new kernel plus a few modules were fully built in less than 10 minutes. (And it even booted fine.)

Opetorns're for low end customers

[reachbach]

Sun's selling linux/opteron boxes only to low end customers. Remember, a linux box comes kick ass cheap and does not have half the features of S10. But for the serious ones, Sun still offers S10 on Sparc(heard of the 32 way Niagara?that's what you would call a beast of a server.A server for real internet workloads). The take home points:
1)Sun sells Linux too(surprise,surprise!!).
2)It does this for the low end guys
3)Sparc is still the defacto chip for any serious high end customer.
4)Sun's amd boxes will be far superior to those of IBM & HP. Why? 'cos HP & IBM don't have their own industry standard OSes, while sun has a beauty in the form of Solaris10 that will give you better value for money on your AMD64 processors.
Finally,learn to accept the truth.Call a spade a spade.S10 is simply a superioir OS to any other OS that exists on this planet today. Embrace it or be left behind. Use DTrace if you like S10 or be content with using top and gather cobwebs snuggling up to a cute penguin.

[ And the Sun never sets forever... :-) ]

Re:They will lose

[SunFan]

Look at Alpha - fastest platform in its day but it had the stink of death even though a well-heeled company (more than one through acquisitions) was being it.

Even thought the quality of Sun's marketing dept. is certainly open for debate, it is clearly better than DEC's was.

What is a "high end" chip anyway?

One thing that differentiates UltraSPARC from Opteron is that UltraSPARC is designed to scale to over 1000 CPUs in a system. Opteron's sweet-spot is up to 8 CPUs. Otherwise, both CPUs have similar characteristics, such as ECC support, etc.

A lot of work can get done with 8 CPUs, but for everything else, there's UltraSPARC, POWER, and Itanium.

Re:They will lose

[afidel]

Well, a high end chip is one designed for throughput as well as number crunching on tiny data sets. In fact the biggest problem for most computational problems today is not cycles but memory bandwidth, and a Sparc system delivers memory bandwidth in spades for a large number of processors. The Sun machines are unfortunatly for Sun not needed by that many people as many classes of large jobs have had architectures designed that allow them to run on piles of comodity wintel/lintel servers. Sun realizes this and want to be the guy that supplies you with those comodity boxes as well as the big back end database server that feeds them all. Another fine example of a high end chip is the PA-RISC chip which does checksuming in every component and which runs all calculations through either two CPU's or two cores to make sure that hardware errors don't produce data errors. That's not something that tends to produce the fastest chip on a given process but there are companies willing to pay for it, which makes it a high end chip.

Re:I suspected

[vrai]

Can you fit more than eight Opterons in a single machine? Can the CPUs be hot swapped? Do they have the proven uptime record of UltraSparcs?

If the answer to any of these questions is 'No' then I forsee a continued market for Sparc hardware. Banks spend millions on new Sparc kit every year - for both new and legacy applications. Contrary to popular Slashdot belief, not every task is suitable for clustering. The bandwidth between nodes is still far too small, and the network induced lag far too great.

When you can get five-nines uptime out of a thirty processor Opteron box - then it'll be time to retire the Sparc range. Until that day comes they'll always have a market.