Feeds

back to article Oracle nudges Sparc T5s back out to 2013

This time last year at Oracle's OpenWorld extravaganza, John Fowler, Oracle's executive vice president of hardware, revealed a faster-paced roadmap for Sparc processors that showed the Sparc T5 chip being pulled into late 2012 from its original early 2013 delivery date. It now looks like Oracle needs a little more time to put …

COMMENTS

This topic is closed for new posts.

It's plenty clear.

"It is still not entirely clear if the M4 chip is another name for a future Sparc64 processor from Fujitsu, as most of us expect."

It's Oracle. And not particularly high-spec.

http://www.hwsw.hu/kepek/hirek/2012/10/sparc2.jpg

http://www.hwsw.hu/kepek/hirek/2012/10/sparc3.jpg

2
0
WTF?

Re: It's plenty clear. - Oracle Lies

"Just double it" just because you double the cores does not mean you get double the performance. Maybe Larry is talking about the Oracle software licenses customers will have to pay and not chip performance.

How Fowler stands up on stage and says SPARC T is 2X every generation and Intel and IBM are only 20-30% is embarrassing. Intels Nahelem chip was clearly more like 3X and IBM's Power6 to Power7 was 5X.

Grant it the performance per thread was and pretty much still is horrible on the T chips but the chart is wrong.

If Oracle wants to actually show proof vs. false powerpoints they will release the MValues for their systems. IBM publishes rPerfs on all their systems. We use IBM's rPerfs to do capacity planning and decide on which models to purchase. SPARC is a trial required to find out what reality vs. marketing.

Fortuantely the sparc install base here is only refreshed on an exception basis, with a migration to x86/power as the rule.

1
4
Holmes

Re: It's plenty clear.

That are actually excellent pictures KiraLexi.

Note that they are referring to M9000-32.. The 32 socket version of the M9000. And as we all know the M9000 goes to 64 sockets. Again shown in the context of the roadmap that Phil 4 linked to, that they are misleading us. This is not x6 it's x3 the M9000, or x6 the M9000@32 sockets.

It's like saying our new model car has six time the horsepower of the old one when that one is only running on half it cylinders.

IMHO it's amazing that they still have clients.

// Jesper

0
4
WTF?

Re: It's plenty clear.

Jesper, the roadmap is a processor roadmap and not a system roadmap. The roadmap shows next M-Series CPU @ +6x throughput and +1.5x Thread strength compared to previous generation, CPU to CPU. Oracle is talking about per socket performance. Where do you get comparisons to M9000-64? The linked slides show 5-6x better performance *per socket* than M9000-32.

2
0
Stop

Re: It's plenty clear.

Phil 4, you are grasping at straws. The Oracle roadmaps you are refering to are talking about systems.

Try relating it to this article by TPM:

http://www.theregister.co.uk/2010/12/23/oracle_sparc_t_chip_roadmap_revisited/

Now if you really think that over a 5 year period that Oracle will be able to increase the socket throughput for Database load by a factor of 40 ?

Do you really think that Oracle will be able to

And last.. the slide says SERVER.. ... is it that hard to comprehend ?

And btw the x2 slides showed about performance.. that is supposedly better than the competition.. is per core performance.. going from a heavily multithreaded thin core..to a core that is somewhat decent.

It's hilarious.. is's basically saying my the top speed of my last tree vehicle's have have doubled each time.. Oh that sounds good, unless the fist one was a bicycle, the second a moped and the last a Eastgerman Wartburg.. then.. well it's not really that impressive.

It's amazing You sunshiners.. are more alike to cultists.. it's scary.. .. I am dead serious.. how can you decouple your critical sense so totally..

// Jesper

1
3
Anonymous Coward

Re: It's plenty clear. - Oracle Lies

Agree, if Oracle was serious about their performance claims, their database business would be in serious trouble because Sparc cores are priced at 1/4th the cost of Power and, now that it is back, Itanium. If Sparc could seriously out perform Power by 2x, 4x or whatever the number is today, every Power user could migrate today as reduce their Oracle DB costs to an 1/8th or 1/10th. Now, Oracle would never, ever allow this to happen to sell a couple of extra Sparc servers, so who is telling the truth. Their software business, which must judge performance fairly or give away 80% margin license business, directly contradicts Oracle's hardware claims.

0
1
Holmes

double what?

I wonder how Spart T5 performs in terms of performance per watt. Why didn't they mention that...

0
0
Silver badge

Take no bets

I'm expecting T5 to be the end of the road, unless someone takes it off them.

2
5
Go

Re: Take no bets

Bogus FUD. Do you really believe Oracle, with its massive R&D investments in SPARC and software on silicon will stop now? They wouldn't release a brand new public roadmap http://www.oracle.com/us/products/servers-storage/servers/sparc-enterprise/public-sparc-roadmap-421264.pdf

with future CPUs showing up in 2015-2016 if they planned to stop developments after SPARC T5? Larry wants to control the whole stack and make certain its #1 in performance and value to customers. Remember his stated goal. He wants to beat IBM on the high end. And he wont stop till he gets there. And now with Fujitsus SPARC64X getting prepared, IBM's in for a double-wammy.

4
2

Re: Take no bets

Maybe you're right. Competition is good for all customers.

But look again at the Oracle roadmap. It does not commit to anything. There is no functionality or any basic description of each future chip. What do they commit to? A name of the chip? Do we know at least the number of cores or threads? +2x Throughput? Throughput of what? Of SPARC III? They can make up basically anything to fit this "roadmap".

1
0
Thumb Up

Re: Take no bets

Atleast Oracle has the balls to put out a public roadmap with anything on it! You don't get that from either Intel or IBM. Isn't the bottom line of CPU advancements all about performance improvements? Well, the way I understand the roadmap, based on what Oracle already announced with SPARC T4, the next T-Series, expected to be the SPARC T5, shows +2.5x throughput and +1.2x single-threaded performance, which I believe would be compared to previous generation (SPARC T4). So maybe its not very detailed on what these really mean, nor what technologies are under the hood, but clearly, a little more than a year later than SPARC T4, you'll see significant improvements in performance-which is quite a feat. if you look at SPARC T3 versus SPARC T4 on real world app performance like Siebel or Java for example, SPARC T4 delivers almost 50% faster Siebel performance and 41% faster Java jvm performance. And seeing that SPARC T3/T4 are still holding several world records on several benchmarks like TPC-C, TPC-H, SPECjEnterprise2010, Specjvm, Oracle E-Business Suite, Peoplesoft, Siebel, JD Edwards, etc, SPARC T5 performance gains is good news.

4
2
fch

Re: Take no bets

well, "you don't get that from either Intel or IBM" - not quite so. IBM maybe, but Intel's roadmaps as far as one of their tick-tocks ahead are usually very well published (by Intel, in fact) and much talked about. Who cares that you can't buy Haswell-based Xeons till mid-2014 ? Everyone knows they're coming, the instruction set enhancements as well as throughput/latency figures for the CPU as well as some chipset details are out there, not just in leaked NDA presentation slides but pretty much all over the tech press.

On the other hand, maybe you meant "Itanium" when you said "Intel" ?

0
0

Re: Take no bets

Intel x86? And POWER? Well, both of them can not show 100% performance increase every other year. Larry Ellison is a hard CEO: "just double it", no matter how technically difficult it will be to increase performance that much.

For instance, the newest Haswell had the graphics and power consumption as main targets, the cpu performance is not a priority. Haswell will only be 10-20% faster than Ivy Bridge, which is not much compared to 100% every other year. The POWER7+ is only 20% faster than POWER7, despite double cpus.

.

.

"It wouldn't be surprising to see KSplice hot splicing come to Solaris, too."

Actually, KSplice is coming to Solaris. But Solaris had a similar technique decades ago, way back in Solaris 8, where you could hot patch the kernel. Regarding DTrace in Linux, the inventor of DTrace, Bryan Cantrill says that DTrace in Linux is not a good port. Yet. Totally useless as of today.

2
2

Re: Take no bets

Phil, I've already seen many roadmaps from IBM and Intel, with much more details on upcoming chips.

BTW, you say "SPARC T4 delivers almost 50% faster Siebel performance and 41% faster Java jvm performance". Cool. Now look at the picture with John Fowler and the slide titled Outpacing the Competition. "SPARC: 2x performance each generation". Shouldn't it make you wonder?

On the other hand performance improvement of POWER6 to POWER7 was about 5x... not close to what tis presentation says.

2
0
Anonymous Coward

Re: Take no bets

Phil 4 - while I admire your confidence in Oracles commitment to SPARC, I would be very surprised if future SPARC processors amount to anything more than a process shrink and very minor tweaks (i.e. more cache due to additional die space). The R&D required to deliver a new generation of processors is not sustainable while x86 processors are rapidly eroding your market.

As long as the x86 server market continues to be a high volume commodity market, x86 will continue to improve performance and features faster than the more specialised processors that it is competing with.

Re: controlling the stack: if you are playing in a commodity market when the alternative is spending billions on R&D that may yield very poor returns? If you are unhappy with your current vendors product, change vendors. If your customers are asking for a wider range of SKU's to meet their needs, add or change vendors... The R&D/support requirement for this is significantly less than the requirements for processor development.

These comments apply equally to Itanium as Intel cannot afford to subsidise a product range that is struggling to survive when they have to figure out how to compete with ARM as mobile devices replace desktops.

I believe IBM will last another round before having to consider the same path due to their use of POWER across multiple product lines that generate significant software/services revenue but it will depend how good the x86 competition is.

ARM and MIPS may disrupt this further, but we will wait and see how they do in the high-end server space when they enter it.

1
0
Holmes

Re: Take no bets

@Phil 4.

Do you really think that a system coming sometime in 2013 that will be 6 times faster than an M9000-32 will be any any what so ever threat to Highend POWER systems ?

It's going according to the data linked to above to be a 384 core machine, running at 3.6GHz. it's lucky it can perhaps nudge itself in between the POWER 770 and POWER 780 using POWER7+, but coming even close to the POWER 795 of today is unrealistic.

To little to late.... kind of like the Oracle mantra.

// Jesper

0
1
FAIL

Re: Take no bets -- tom 99

Actually, I think they're pretty clear to state that those numbers are relative to the previous version.

0
0
Happy

Re: Take no bets - nope, doubling down!

Anon --- I would be very surprised if future SPARC processors amount to anything more than a process shrink and very minor tweaks (i.e. more cache...

Process shrinks are not that easy. When you take the same paths, you can't necessarily get linear clock speed increases out of them since the distance of a path does not uniformly decrease in length (depending on the routing.) The amount of time it takes for systems to perform the routing of these massive chips it completely outrageous.

There appears to be an M4 & M5 both in test - possibly the end-2013 and end-2014 chips on the roadmap.

The latest Oracle road map has a T-series chip past the T5 in mid-to-late 2014.

Considering that POWER-mongers have been wrongly suggesting SPARC is not long for this world - it appears that SPARC and POWER are going to be the last chips standing, in the high-end arena.

It looks like you might be surprised!

1
0
fch

Re: Take no bets

when Larry says "just double it", then consider that doubling the length of a yacht won't double its speed, but will more than double its cost.

It all depends on what exactly you double, and there are many ways of doubling "performance". As there are even more benchmarketeering ways of measuring doubled performance.

Sun sailed that course for many years with this idea of "sum(more cores) > sum(fast cores)". It seems Larry got converted, like if there's not much else to show then at least show you can double "it" - find a suitable "it".

I don't doubt there's some market for these, just it's shrinking not growing; what you can do these days with a 20k$ x86-based server you couldn't do with a cluster of ten Sun E10k's fifteen years ago. "High-end" computing is becoming a commodity and that's not a trend which will reverse any time soon.

Solaris, on the other hand ... give me more. Site licenses for Solaris, for example, would be a great way of knocking crimson headwear peddlers out in places. Relay that to Larry if you would ? Something like, we'd be happy to more than double our use of Solaris - though, if and only if we can cap the license costs ...

0
0

Re: Take no bets

"...Sun sailed that course for many years with this idea of "sum(more cores) > sum(fast cores)..."

Yes, as opposed to IBM that mocked that course, saying "1-2 strong cores are better than many weaker cores, because databases runs best on strong cores". Where are those 1-2 core IBM cpus running at 7-8GHz today? Maybe IBM should mock POWER7+ cpus with 16 cores running at an even lower clock speed.

When IBM competitors do something, it is the worst idea in history. When IBM does the same thing years later, it is "innovative".

1
1
Thumb Down

Re: Take no bets

It's always a balance between throughput and single core performance.

And you really should check your data.. now the T1 runs at 9.6 GHz is from Q4 in 2005 when Scott McNealy in his desperation made up that way of doing calculations. At that time when the IBM processor to match it was the POWER5.

So if you take the best specint2006 rate result of the POWER5+ (cause there aren't any POWER5's) and compares it to the best of todays new POWER7+ you get this:

POWER5+ at 8 chips with a total of 16 cores running at 2.2GHz is 248, that is 15.5 per core and 31 per chip. The best per core POWER7+ result will give you 58.3 per core and the best per socket score will give you 383.2 per chip.

Now that is a factor of 3.8 per core and 12.4 per chip compared to POWER5+

And you estimate the POWER5 processor the numbers would be a factor of 4.8 and a factor of 15.6.

how is that not delivering on both per core throughput and chip throughput. Something that Oracle/Sun have failed to do.

// Jesper

0
1

Let's not accept all marketing...

From the article:

"Oracle's edge will still be that it owns the whole stack. "We are the only company in the world that can take application knowledge down to the silicon," Fowler said as he ended his keynote. Until Microsoft or SAP get into the hardware biz, or IBM or Hewlett-Packard buys SAP, he has a good point."

I'm not sure that this is totally right... IBM for example does have most of the things inhouse... Be it OSes or databases or analytics software or management, compilers etc. So in that sense this is really a true statement.

On the other hand, IBM has much more on the silicon side. For more advanced semiconductor and architecture research as well as fabs. So they are far better positioned to excel in the silicon side (see example their eDRAM).

1
0
Pint

Children will be disappointed on Christmas day

Oh dear, they're totally going to miss the holiday season sales now! :-p

0
0
This topic is closed for new posts.