Oracle's new T5 Sparcs boost scalability in chip and chassis

Oracle is launching its much-awaited Sparc T5 processors for entry and midrange servers, along with Sparc M5 processors to effectively replace the iron it currently resells from server and chip partner Fujitsu. That Japanese supplier furnished Sun's and then Oracle's brawny-core Sparc Enterprise M midrange and big iron systems …

[Anonymous Coward]

M5? This won't end well

Just tell me they didn't let the chip's creator impress his own neurological engrams onto the damn thing, OK?

[Kiralexi]

lol

List prices have now been posted. $53k for the lowest-end 2-socket system seems... optimistic, compared to what IBM is selling 2s Power systems for.

CPU's look decent, though. I'm interested to see how the M5-16 is positioned relative to the T5-8, since core performance on the T5-8 should be slightly higher.

[Destroy All Monsters]

Re: lol

Once you install the Oracle Database on that, you won't feel the pain.

I wonder how the TCO is compared to Big Blue's stuff?

[Anonymous Coward]

Re: lol

Well how do you want the comparison to turn out? I'm sure you can find "expert" evaluations supporting both sides.

[John Riddoch]

Re: lol

Power 7 is 8 cores per socket, so you're getting double the number of cores in a T5-2 vs a 2 socket Power 7 system. Of course, performance isn't just about core count so it'll depend how each platform runs your workloads. I suspect Power 7 would blast single threaded workloads better, the T5-2 would chew up multi threaded apps better.

Also, who ever pays list price? Cost comparisons will depend on how much you can barter down Oracle & IBM against each other...

Core performance on the M5 should be far better than the T5 systems; it's got much more L3 cache per chip (48MB vs 8MB), fewer cores (6 vs 16) so a massive boost to L3 cache per core (8MB vs 0.5MB). Obviously, you need more of them to match the T5 equivalent system for raw compute capability, but anything which is memory intensive will prefer the M5 boxes.

[the spectacularly refined chap]

Re: lol

Core performance on the M5 should be far better than the T5 systems; it's got much more L3 cache per chip (48MB vs 8MB), fewer cores (6 vs 16) so a massive boost to L3 cache per core (8MB vs 0.5MB). Obviously, you need more of them to match the T5 equivalent system for raw compute capability, but anything which is memory intensive will prefer the M5 boxes.

Oddly enough that's the one I skipped over - it seems to me to be a conceptually flawed "me too" attempt to claim single-threaded credentials. Niagara and successors have always focused on overall throughput instead. The entire approach made a lot of sense - you can either spend increasingly large amounts of silicon on caches and various prediction stages gaining ever smaller returns or you can simply state "Cache misses and pipeline stalls are gonna happen. Do something else when they do."

The entire core design is premised on that attitude - it is madness to have so many threads per core otherwise. The M5 deviates from that basic approach with no alteration to the threads per core, no smarter pipeline, just an increase in the already relatively distant L3 cache. I'll be interested to see how it works in practice but I suspect it is a purely marketing-driven variant with little practical benefit.

[Jesper Frimann]

Re: lol

No IMHO you are wrong, the M5-32 is the more sensible system. This should really allow clients with legacy Mseries and older highend servers to migrate to a platform that makes sense for them. Again the M5 is 42 percent faster at same clock than the T5 on SAP2Tier app benchmark. That is a lot.

Look at the Oracle UNIX markedshare.. it's been dropping with catastrophic speed these last many years. The M5-32 is a step in the right direction for Oracle, but is it to little to late ?

It's brand spanking new, and still not able to beat the competitions rather ageing POWER7 (2010 technology) based POWER 795.

// Jesper

[Anonymous Coward]

Solaris versions

Worth mentioning a couple of things about what version of Solaris you can run. The T5 systems will run Solaris 10 update 11 natively either as a whole system image or LDOM guest; I believe LDOM control domains have to be Solaris 11.

M5 systems can only run Solaris 11 native, with Solaris 10 an option on LDOM guests (http://www.oracle.com/us/products/servers-storage/servers/sparc/oracle-sparc/m5-32/overview/index.html).

You can run Solaris 10 containers under Solaris 11, Solaris 8 & 9 containers have to run under a Solaris 10 native image, so either T5 whole system or LDOM on M5.

[frankgobbo]

Tasty

Awaiting delivery of our T5-8s (*20). Looking forward to it, they look like very nice machines. The virtualisation options available are great too.

TCO running Oracle's software stack will be much lower than HP, IBM etc - the per core licensing multiplier is much lower on Oracle hardware than the competitors (frankly a smart move).

Looks like the Sun of old is back in business, I can't wait to see it.

[Anonymous Coward]

Re: Tasty

"TCO running Oracle's software stack will be much lower than HP, IBM etc - the per core licensing multiplier is much lower on Oracle hardware than the competitors (frankly a smart move)."

Software licensing will be the same. They just raised the core multiplier for T5 to .5 in line with Xeon which is about right. Oracle is not about to lose software revenue, extremely high margin, to gain some server sales, low margin. If what you say is true, and Oracle is giving SPARC a preferential multiplier, then a large number users on other servers could move onto SPARC... which Oracle would like but mostly not care that much... and claim their "much lower" licensing/support rates.... which Oracle would not like and care a lot about.

[asdf]

SPARC lol

Who in their right mind is doing anything but migrating away from SPARC boxes at this point? That chip would have to absolutely destroy the competition to justify giving Oracle more vendor lock in at this point. Based on the numbers Oracle released last week of the hardware its shifting (or failing to shift) this isn't just my opinion.

[Anonymous Coward]

Re: SPARC lol

"Who in their right mind is doing anything but migrating away from SPARC boxes at this point?"

+1 for investing in anything by Oracle. Solaris servers are legacy boat anchors.

[asdf]

Re: SPARC lol

>Solaris servers are legacy boat anchors.

SPARC Solaris servers are. x86 Solaris can be decent especially when I get my hands on them and GNUify the shit out of the box (GNU findutils not only owns all the other UNIX(s) versions but makes them pee their pants and cry).

[Aitor 1]

Re: SPARC lol

yes but... I like my machines as standard as possible. Maybe they don't work as well as they could, but I hate to be the first one to have an obscure problem!!

[h3]

Re: SPARC lol

Shows how little you know - gnu findutils is nowhere near as fast as the version used by star.

You can get it as part of the schilly toolbox. (If it is locate you want you only need that if you don't know what you are doing.)

[Anonymous Coward]

Re: SPARC lol

I suspect the Solaris sys admins are hammering the down-vote button on their Zimmer Frames....

[Destroy All Monsters]

Re: SPARC lol

Matt, stop sockpuppetting as AC.

Or is there only POWER and the x86 and in the future, better ARMs? Well, maybe. But probably not.

There won't be Itanium though.

[Matt 21]

Re: SPARC lol

I have to say our lot have moved away from SPARC. Over priced and under performing is ho I see it. It's a shame as back in the 90s Sun held so much promise.

[Alan Brown]

Re: SPARC lol

"There won't be Itanium though."

There will be, but not the way you're thinking. Quite a bit of Itanium-ness is making its way into the X86 line at hardware level.

[Phil 4]

Re: SPARC lol

Look at IBM's last 4 Qtrs of Financial statements and question how is it that IBM's Power systems business is down so much? Maybe SPARC is BACK?

IBM reports 2012 first-quarter results

“Revenue: $24.7 billion, flat”

“Revenues from Power Systems were flat compared with the 2011 period”

http://www.ibm.com/investor/1q12/press.phtml

IBM reports 2012 second-quarter results

“Revenue: $25.8 billion, down 3 percent”

“Revenues from Power Systems were down 7 percent compared with the 2011 period”

http://www.ibm.com/investor/2q12/press.phtml

IBM reports 2012 third-quarter results

IBM total revenues for 3rd Qtr of 2012 of $24.7 billion were down 5%

“Revenues from Power Systems were down 2 percent compared with the 2011 period”

http://www.ibm.com/investor/3q12/press.phtml

IBM REPORTS 2012 FOURTH-QUARTER AND FULL-YEAR RESULTS

“Revenue of $29.3 billion, down 1 percent”

“Revenues from Power Systems decreased 19 percent compared with the 2011 period”

http://www.ibm.com/investor/4q12/press.phtml

[Matt Bryant]

Re: Phil 4 Re: SPARC lol

"Look at IBM's last 4 Qtrs of Financial statements and question how is it that IBM's Power systems business is down so much? Maybe SPARC is BACK?...." The IBM figures have been down pretty much inline with the overall decline in the UNIX market. Nothing to do with SPARC, indeed the Snoreacle server results have tumbled at four or five times the rate IBM's have declined.

[Anonymous Coward]

Re: Phil 4 SPARC lol

"The IBM figures have been down pretty much inline with the overall decline in the UNIX market."

They have not been great, but they have been considerably better than Oracle/Sun and HP.

It is also just a product cycle issue. With any big iron, the numbers look really good when the new stuff comes out and people all upgrade within a year or so, which was 2010-11 for Power7, and then flatten or fall off because you are comparing a refresh cycle year with a non-refresh cycle year.

[Phil 4]

Re: Phil 4 SPARC lol

Although most of the press and analysts talk in terms of Revenue marketshare, this can be quite misleading, especially if you sell very expensive systems but NOT necessarily sell many of them. Selling 10 x $10M systems will show higher marketshare than selling 10,000 $9.5K systems.

~50% of IBM's UNIX/RISC revenue (~$2.5BN/Year) comes from just 3 very expensive systems that has had no real competition in last 3+ years. Power 770/780/795. The Sun/Oracle M-Series was Sun/Oracle's revenue generator systems trying to compete against high end Power7, but it came out in 2006/2007, over 6 years ago. Maybe that’s why Oracle HW has been in big decline these last 3+ years because theres been no real replacement to M-Series in 3+ years? However if you look at the volumes of these systems, its less than 10% of the total volume. For Oracle SPARC/UNIX, SPARC T4 represents ~65% of the volume but less than 40% of the revenue. Not enough to show revenue growth. And Oracle Engineered systems are still ramping up with longer sales cycles, and in many cases doesn’t replace high-end SMP systems.

Whats important though for ISV's, and customers, is *UNIT* marketshare. Why? Because more volume translates to more license sales, lower costs for smaller more powerful systems with lower pricing on components and ultimately a more healthy eco-system. Its only when you have enough volume, do you become really relevant, as has been the case for Xeon. SPARC/Solaris still has the #1 UNIX volume marketshare and with SPARC T5 and SPARC M5 as well as SPARC M10, Oracle now has the opportunity to gain not only more volume marketshare, but revenue marketshare. And realize it took IBM over a year to release the entire Power7 line, and 4+ months to release Power7+ line, and no Power7+ on Power795. Lets see if Power8 is released *next* year!

[Matt Bryant]

Re: Phil 4 Re: Phil 4 SPARC lol

"..... Its only when you have enough volume, do you become really relevant....." So what you're saying is, compared to T5's/M5's real competitors, Xeon and Opteron, Snoreacle's teeny-weenie market share mark it out as truly irrelevant?

[Anonymous Coward]

Re: SPARC lol

"Look at IBM's last 4 Qtrs of Financial statements and question how is it that IBM's Power systems business is down so much?"

You don't have to question it for very long. Power 7 came out in 2010, most people upgraded to the new gear in 2010 and 2011. Most people had already upgraded by 2012 and/or were waiting for 7+ to arrive. It is the natural product cycle.

Regardless, talk about pot and kettle. Sun has fallen off much worse, down 20-25% every quarter even if you include Exadata, Exalogic's growth, which has nothing to do with SPARC/Solaris.

If you look at the overall IDC Unix market share, IBM now has over 50% up from about 35% in 2008. Sun has about 18% down from about 25% in 2008.

[Matt Bryant]

Re: Destroyed All Braincells Re: SPARC lol

"Matt, stop sockpuppetting as AC....." Sorry, wasn't me, m8y, I was still too busy laughing at Larry describing his Frankenstein creation as a "mainframe"! Someone needs to tell Larry there aren't any mainframes running as webservers alone.

In fact, it's an insult to Frankenstein to describe Larry's new jokes that way, at least Frankenstein was a creation of the bits of others that supposedly made up a better man, whereas the T5 is the same old junk shrunk, and the M5 is just the same junk with a lump of cache bolted on the side. In the M5's case it's a laughable attempt to alleviate CMT's biggest design flaw, but without predictive technologies and still with the same wiener T5 cores. I cannot see this outperforming Xeon on real workloads, let alone Power or Itanium. IMHO, I give it a year and then Snoreacle will be swallowing their pride and badging Fudgeitso boxes again, the Athena chip looks a much better enterprise CPU.

[Aitor 1]

humm x86 anyone?

I would prefer to have a horde of 3500-4000€ 16 core xeon servers... but hey, I recognize it is not as easy to manage without adequate tools (and software really designed to be distributed).

I DO like solaris machines, but they seem to be too expensive.

[asdf]

Re: humm x86 anyone?

Works for Google and some of the other big boys. The day the SPARC route especially now under Oracle returns a lower TCO in any 3rd party study is the day Oracle might actually report its hardware sales not buried deep in the financials.

[Destroy All Monsters]

Re: humm x86 anyone?

"Commodity hardware is like violence, or XML ... if it doesn't work, you are not using enough of it"

[stanimir]

Re: humm x86 anyone?

Google/Facebook/Twitter other big boy do not care about consistency.

Who cares if you have receive a different search result ?

[stanimir]

Re: humm x86 anyone?

The main problem w/ x86 is the CAP theorem: out of Consistency, Availability and Partitioning you can pick only 2.

So you either sacrifice the Consistency or Availability as you go for the Partitioning . To mitigate the problem you might need to roll into 40Gb infinity band.

[Alan Brown]

Re: humm x86 anyone?

"The main problem w/ x86 is the CAP theorem: out of Consistency, Availability and Partitioning you can pick only 2"

Not to mention the elephant in the room which is power consumption. X86 is still way less efficient than everything else.

[asdf]

Re: humm x86 anyone?

>Not to mention the elephant in the room which is power consumption. X86 is still way less efficient than everything else.

ARM soon will be taking care of commodity boxes running power hungry and will destroy everyone else on performance per watt.

[Matt Bryant]

Re: asdf Re: humm x86 anyone?

"......ARM soon will be taking care of commodity boxes running power hungry and will destroy everyone else on performance per watt." Slight problem with that analysis - it's complete male genitalia. Remember a little company called AMD? Remember how they did that x64 thing and everyone said Intel was toast? Want to look at the market today and see if Intel have gone and AMD are running it? Intel are not stupid, they react to threats very well. The Atom CPUs are already a good start, and as the many disparate ARM designs scale up (as they will have to in order to be chips actually useful for doing more than playing Angry Birds on a tablet) it will find Atom is more than a match, especially with Intel's marketing muscle thrown into the mix. Those that write off Intel so quickly ignore the lessons of history.

[asdf]

Re: asdf humm x86 anyone?

Calm down Matt. I was just saying whether its ARM or Atom or whatever the commodity box market will ultimately prevail. If it was anybody but Intel I would say no way they get that x86 legacy POS architecture (instruction set was such balls even when it came out) to ever be power efficient but having more money than God and being consistently a generation ahead of everyone else in fabbing tech it might some day come true. Wow will they have worked an engineering miracle to make anything x86 based efficient power wise.

[Anonymous Coward]

Here are some benchmarks

http://www.oracle.com/us/solutions/performance-scalability/index.html

[Jesper Frimann]

Re: Here are some benchmarks

Thanks for the link.

Well ignoring the normal the normal Oracle Marketing bull. Then the M5-32 looks like a solid machine, although not really a match for the POWER 795, even though that is machine that IMHO needs a speed bump.

The T5's look solid, and it's nice that see that someone finally is taking on IBM in the UNIX marked, it's been a pretty onesided story for the last 3 years.

So hopefully we'll see some answers from the other vendors, although I have my doubts about HP.

// Jesper

[Phil 4]

Re: Here are some benchmarks

SPARC M5 is not a match to the Power 795? You’ve got to be kidding me! SPARC M5 is PCIe Gen3, Power 795 is still PCIe Gen1! The I/O on Power795 is the same as the previous Power 595 introduced back in 2008! SPARC M5 is 8-socket glue less and supports up to 4 hardware domains for full electrical isolation, not possible in 795. The list price of a 32-socket SPARC M5 with 4TB of RAM has list price of just over $1M, while a *base* config P795 with just 32 cores active and *1TB* of RAM is roughly same list price. Sure discounting will be high here, but IBM will need to discount a minimum of 70% to match Oracles SPARC M5 *list prices* And then theres the performance comparisons.

[Matt Bryant]

Re: Phil 4 Re: Here are some benchmarks

Aaaaaaannnd our Honourary Sunshiners Of The Year Golden Blinkers Award goes to...... Phil 4! Honestly, chap, you need to stop reading the Snoreacle press releases and start reading some industry news. Start here (http://www.theregister.co.uk/2013/03/20/oracle/) - do so in private as your subsequent crying may upset your colleagues.

/SP&L

[Jesper Frimann]

Re: Phil 4 Here are some benchmarks

Matt has a point, one thing that all our clients, that run Oracle, is talking about is Oracle license cost reduction, some of the projects we have done, have actually paid for themselves, we are talking Hardware, labour etc etc. in Oracle license reduction cost, over 2,3,4 or 5 years.

Which kind of like puts Oracle in the same position as IBM 20-25 years ago, but again that is also what Larry wanted, to make Oracle into a copy of the old monopoly IBM. He just didn't learn read what happened to the IBM of that period.

// jesper

[Jesper Frimann]

Re: Here are some benchmarks

@Phil 4

And in those PCIe G3 slots you'll put PCIe Generation 2 adapters, which is basically what Oracle is shipping, nice for future adapters.. irrelevant for now.

Furthermore where do you get your prices, I haven't seen any anywhere, not even from the Industry analyst tools we use ? Well you most likely use internal Oracle prices.

And did you include Services in your comparison ?

Again for a highend server the Oracle warrenty is 1-Year Hardware Warranty with 2 Business Day On-Site Response, you've gotta be kidding, both SD2 and POWER 795 has same day, and for the POWER 795 it's even 24x7.

And from my response to your post in another thread, you now should have learned that the Oracle Service on their servers cost more than then actual servers themselves. Again 3 years support on a T5-4 cost 1.5 times that of the actual server cost.

Again Oracle sells you the server cheap, and then turbo charges you for the service.

Larry wasn't kidding when he claimed that he wanted to turn Oracle into the IBM of the 70/80'ies, he most likely plans to dump his stocks in good time before Oracle goes the way of IBM in the 90ies.

// jesper

[Anonymous Coward]

Congratulations Oracle

Seriously. They have stayed the course on the "Throughput Computing" initiative of yore and these boxes are impressive as hell. They continue to simply shatter standard benchmarks and bury the competition. Not an easy thing to do really. The team did great work and should be proud of their accomplishments. Not an Oracle fanboi, just giving credit where credit is obviously due.

[Jesper Frimann]

Re: Congratulations Oracle

No, they do not shatter the competition, and you do sound like a fanboi, when using such wording :)

One thing that is good that Oracle have fixed, is putting hotswap adapters back in their servers. That was nice.

But IMHO, the T5-8 might be a bridge to far when seen as a compute node.

IMHO it's might be to big for a server that does not support hotswap/upgrades of component other than adapters and power. I know how much time it takes to empty large servers for virtual machines, either through moving running virtual machines, cluster failovers or simply closing down virtual machines.

// Jesper

[Anonymous Coward]

Re: Congratulations Oracle

"Throughput Computing" - hmmm - is that like the Oracle Secure Computing Initiative - you know those 'Unbreakable' servers (that were promptly hacked) - laugh a minute that was....

[Anonymous Coward]

Re: Congratulations Oracle

Run a PoC with your actual workloads on IBM's Power and Oracle/Sun's SPARC... see how it comes out.

[TheVogon]

Re: Congratulations Oracle

"Run a PoC with your actual workloads on IBM's Power and Oracle/Sun's SPARC... see how it comes out." or put them on SQL Server and buy yourself an island somewhere sunny with the change....

[Jim 59]

Oracle

Great to see the new systems and continued Solaris roadmap. It's just a shame how Oracle is treating existing customers, at least here in the UK. Some shops are transitioning to Linux just to get away from poor and expensive support. Not chucking Sun boxes out, just replacing them with Linux when it comes to refresh time. Others are buying OracleT4 kit though. (I've been a Sun admin for 20 years.)

[apleszko]

M5-32 is not the only one with 32TB RAM

@TPM, in fact, your own article about Fujitsu's SPARC servers mentions that a fully configured M10-4S can also reach up to 32TB within a single NUMA system, but with 1024 cores instead of 192 from M5-32.

http://www.theregister.co.uk/2013/01/25/fujitsu_racle_athena_sparc64_x_servers/

Sure, Fujitsu M10-4S reaches 32TB RAM with 64 sockets instead of 32, nevertheless is 32TB within a single image.

"At the moment, Oracle is shipping only one box based on the Sparc M5, with 32 sockets, called the Sparc M5-32 (obviously). Fully configured, this big-iron box weighs in at 192 cores, 1,536 threads, and 32TB of main memory. No one has as much memory in a single image today – not IBM, not Silicon Graphics, not HP, and not Fujitsu."

[ToddR]

Re: M5-32 is not the only one with 32TB RAM

SGI have much larger, ccNUMA, systems than 32TB

[Phil 4]

Re: M5-32 is not the only one with 32TB RAM

And SGI's systems run enterprise software like Oracle DB?

[Kebabbert]

Re: M5-32 is not the only one with 32TB RAM

ToddR

If you read wikipedia on ccNUMA, it says that ccNuma servers are just a cluster. This means that the SGI altix servers with 1000s of cores and 64(?) tb ram, is just a hpc cluster. Sure, they might use single image, by using a hypervisor that tricks linux into believing it is single image, but it is not a smp server.

Ibm p795 and oracle servers are smp servers, not a cluster. Have you asked youself why ibm and oracle and hp, releases unix servers with only 32 or 64 cpus, whilst linux servers have 2048 cpus? A hint: one are clusters (disguised) and the other are smp (one big fat server).

It is much more difficult to build a big smp server with32-64 cpus, than a cluster. It takes decades of research and development, to build scalable smp servers.

Linux also has smp servers, the biggest on the linux market has 8 cpus. There are no 16 cpu linux smp servers for sale, because linux can not scale to 16 cpus, smp wise. But there are 2048 cpus linux servers, disguised clusters they are. Can you show a link to a 16 cpu linux server that is smp? No. For a reason, linux can not scale smp wise.

[Jesper Frimann]

Re: M5-32 is not the only one with 32TB RAM

@Kebabbert

http://www-03.ibm.com/systems/power/hardware/795/

Just to mention one of many.

// Jesper

[Matt Bryant]

Re: Kebabfart Re: M5-32 is not the only one with 32TB RAM

".....they might use single image, by using a hypervisor that tricks linux into believing it is single image....." You're missing the big point - the business doesn't care. What they care about is the business requirement, if that leads to a technical requirement to run one application over 2048 cores, they don't care if those 2048 cores are due to a hypervisor "tricking" Linux or it being one SMP image - indeed, they might value the additional resilience of a grid over a single SMP instance - all they care about is which does the job to the required performance level and at the least cost and disruption to the business. Major fail to see the bigger picture.

[asdf]

Re: Kebabfart M5-32 is not the only one with 32TB RAM

>You're missing the big point - the business doesn't care.

>Major fail to see the bigger picture.

Unless his business is selling SMP hardware/contracts in which case that is the business and he sees the bigger picture by talking up SMP and poo pooing commodity hardware.

[Kebabbert]

Re: Kebabfart M5-32 is not the only one with 32TB RAM

@Matt Bryant,

You are missing the point. The business DO care. You can not run SMP workloads on a cluster. There is a reason Linux is targeting 4-8 socket SMP servers, and 2048-4096 cpu servers (clusters), with nothing in between. There are no 16 nor 32 cpu linux servers for sale. If it were easy to create SMP servers, then we would see many cheap 32 / 64 / 128 cpu linux servers for a fraction of the price of the IBM P795. But the only 32 cpu servers, are from Oracle, IBM and HP. There are no such linux servers.

If you could run SMP workloads on a cheap cluster, there would be no market for P795 / m9000 / HP superdome(?). Why would anyone pay huge amounts for a Unix server, when you can get a cheap Linux cluster? No one would. For instance, the P595 that IBM used for the old TPC-C record costed $35 million list price. One single f-cking 32 cpu server. Why not buy 128 cheap PCs and stuck on a fast switch with Linux instead? SMP servers are different from clusters.

Here is an article on one of those 2048 cpu Linux servers (which turns out to be a cluster)

http://www.theregister.co.uk/2011/09/20/scalemp_supports_amd_opterons/

"...Instead of using special ASICs and interconnection protocols to lash together multiple server modes together into a shared memory system, ScaleMP cooked up a special hypervisor layer, called vSMP, that rides atop the x64 processors, memory controllers, and I/O controllers in multiple server nodes. Rather than carve up a single system image into multiple virtual machines, vSMP takes multiple physical servers and – using InfiniBand as a backplane interconnect – makes them look like a giant virtual SMP server with a shared memory space..."

.

.

@Jesper,

Yes, I know about the IBM P795. If you read my post again, you will see that I specifically mentioned the IBM P795. Although Larry Ellison mocks it and other SMP servers, I think the P795 is mighty cool. It is probably the most powerful server for sale today. It would be cool to just see and touch one of those beasts in real life! :) I hope Oracle can beat it in two years, when Oracle releases the 16.384 thread 64TB RAM SPARC server. But IBM is not waiting, IBM will release POWER8 equipped servers then. It will be an interesting future for all of us! :)

[Matt Bryant]

Re: Kebabfart M5-32 is not the only one with 32TB RAM

".....You can not run SMP workloads on a cluster...." Yes you can, as was proved years ago with simple cases such as SETI@home. A more up-to-date example would be the idiotic LOIC used by the Anons to bundle up thousands of PCs in place of the large SMP systems they do not have. Yes, most grid solutions are aimed at replacing supercomputers, but I am surprised that a chronically blinkered Sunshiner such as yourself so quickly forgot Sun's own attempt at grid, Sun Grid (now Oracle Grid Engine).

"....There is a reason Linux is targeting 4-8 socket SMP servers....." because it is the largest customer market segment for Linux, simple as. Commercial Linux companies like Red Hat are simply concentrating their resources on the market segment they expect to make the most money from.

"....But the only 32 cpu servers, are from Oracle, IBM and HP....." <Cough*UNISYS*cough><cough*Fujitsu*cough><cough*SeaMicro*cough> Damn, Kebbie, please do some reading, all this coughing is killing me!

"....Why would anyone pay huge amounts for a Unix server, when you can get a cheap Linux cluster?....." I know, I have asked my bosses this question but the answer seems to be they like proven, off-the-shelf stacks of commercial and proprietary software because they have confidence in them. You could say exactly the same about mainframes - why would anyone still buy one when UNIX can do the job cheaper and better?

[Kebabbert]

Re: Kebabfart M5-32 is not the only one with 32TB RAM

@Matt Bryant,

SETI@home is HPC workload. HPC are clusters. There is a difference between SMP servers and HPC clusters. HPC clusters can not replace SMP servers. That is the reason you need to buy Oracle/IBM/HP if you need big SMP servers. And, IBM Mainframes are also SMP servers, they are not HPC clusters.

[Peter Gathercole]

Re: Kebabfart M5-32 is not the only one with 32TB RAM

I would dispute that seti@home is an HPC workload. It is a distributed workload, partitioned into units that can be worked on in isolation from each other.

There is a huge difference between a distributed workload and a proper HPC workload, and people like weather agencies, atomic research institutions etc. would be only too happy to explain.

Proper HPC needs a huge interconnect, so that a single model can be broken down into multiple threads spanning many systems, all passing data between each other.

BTW, IBM P7 795s are quite cool, but P7 775s are even cooler (literally, water cooled cooler!). I know, because I work with a couple of clusters worth.