They want their RISCs back.
As part of the Sparc SuperCluster rollout yesterday, Oracle chief exec officer Larry Ellison let slip that the future Sparc T4 chip was coming out next year. Oracle and Fujitsu also announced a revved up Sparc64-VII+ processor for the Sparc Enterprise M series of midrange and high-end SMP machines, but did not provide any …
> In that new Oracle roadmap above the company is promising "+3x Single Strand" performance, which we take to literally mean that the clock speeds will be three times those of the past Sparc T2+ or now current Sparc T3s. That means between 4.8 GHz and 5 GHz clock speeds
I doubt they can bump the clock that much. In my opinion they will redesign/change multithreading model. With T2 and T3 the single thread performance is poor due to CMT constraints.
"With T2 and T3 the single thread performance is poor due to CMT constraints"
Actually it was a self imposed choice to increase throughput and reduce power usage.
You can call it a constraint if you like. Great idea for some loads, but obviously the wrong choice for single thread perf. It seems that it would be relatively simple to back off that choice a bit and increase the single thread perf a bit more.
As far as whether they can hit 4-5 GHz, I don't see why not. Especially if they back off from higher strand and core counts. However, I bet they are saying 3x the original T-Series GHz of 1.2 (if I remember right). That would mean only 3.6GHz perf, which isn't too shabby, but nowhere near 4.8-5.0 GHz.
Timothy Prickett Morgan writes, "which we take to literally mean that the clock speeds will be three times those of the past Sparc T2+ or now current Sparc T3s. That means between 4.8 GHz and 5 GHz clock speeds"
I don't think so - there was a lot of other technology surrounding SPARC besides more threads, more cores, and higher clock rate.
With papers still being presented on Rock technology (i.e. transactional memory, scouting threads, etc.) from Oracle - one might think that some of that technology will land into the new processors.
I don't think that Timothy Prickett Morgan, really nailed it.
If you compare the Old SUN roadmap:
with the newer Oracle roadmap for SPARC:
T3/Rainbow Falls with x2 cores compared to Victoria Falls is clearly listed as T-Series x2 Throughput on the Oracle roadmap in 2010.
T4/Yosemity Falls with half the number of cores to Rainbow Falls (Possible x2 faster cores) and 1.5 times the frequency with gotta be the T-Series in 2011 with x3 the thread throughput.
Then on the Oracle map there is a M-Series with 64 sockets listed in 2012. This corresponds very well to the YellowStone Falls on the old SUN map.
And finally on the Oracle map there is a T-Series in 2013 with 1-8 sockets, which gotta be cascade falls, which is also listed in 2013.
The real surprise here, if my deductions are right, is that M is going CMT processors also.
But otherwise it seems that Oracle is just executing on the existing "SUN roadmap".
It makes sense beefing up the hardware that sits beneath Oracle's software products, so that the on-premises licence revenue keeps rolling in. This should slow down their customers migrating away to cheaper options. Ultimately Oracle will have to join in on the platform as a service game to provide the value their customers will insist on, i.e. database as a service as part of a wider application platform running in the cloud which Oracle could create themselves. Not sure if bunging Oracle into Amazon EC2 will be compelling enough.
It means not higher Hz. It means that each thread is 3x faster. Strand can be translated to "thread".
Stronger threads will not be achieved by higher Hz, but by stronger cores. A redesign of cores is coming, where each thread will be 3x as fast as the T4 thread.
Earlier, only improvements of Niagara was more and more cores and threads. For the first time, the cores will be stronger, featuring out of order execution, larger cache, etc. A total redesign of cores.
There is no way to redesign Rainbow Falls in a year or two and get features you mention. Out Of Order execution is not something you simply add to In-Order processor such are T series. T series has a barrel thread execution which is pretty dependent on cores being in-order and highly multi-threaded.
Only way to do this is by revamping ROCK. ROCK has meaty cores (16 of them) and two threads per core, so maybe Larry will put some cash to move design to 45nm process and make it suck less than original 250W. That would be interesting to see.
Nice read. Lots of "innovations" by marketing dept:
- out of order, wow!!
- 28nm in T5 (intel, amd and ibm will have 16nm by then) :)
- "we do have a unique situation here where the operating system, processor development, virtualization software, and the applications all under one roof"... unique? except they do not own a single chip fab, so they depend on other's technology (taiwan), Oracle cannot innovate much here. IBM has everything from chip fab to applications.
- "virtualization software that can offer 100s of logical domains" -- reminds me of power5 partitions back in 2004... or mainframe lpars in 1980 :)
"except they do not own a single chip fab"
Really? That's your concern? So? Some would actually consider this a good thing. If the current fab has problems, use another one. IBM has the problem of supporting a relatively low volume chip manufacturing process, which must be extremely expensive for them. Manufacturing your own chips is not a feature (just ask HP).
Your other comments are just childish. We can all go over marketing terms and say we had this first, or we had it better, but it's all just nonsense. SPARC has supported domains before any of the other Unix vendors... so, who really cares?
In 2015 Oracle will release a Solaris server with 16.384 threads and 64TB RAM. That machine will crush seriously. That config sounds sick today, but in the future everyone will offer such servers.
Back in the days, the Niagara with 8 cores was shocking. It was first. Now everyone offers 8-core cpus.
Suppose that this is not good news to HP but people should know that Oracle has changed their licensing policy quite radically.
Intel Itanium multiplier has been increased from 0.5 to 1 and SPARC VII+ decreased from 0.75 to 0.5.
Change took place already 5 days ago and there has been no news in el Reg! :)
Yes, basically they made it cheaper to use their own hardware.
Although there is a loophole in the Itanium per core licensing scheme. It clearly reads:
(For servers purchased prior to Dec 1st, 2010)
(For servers purchased on or after Dec 1st,2010)
So basically this price raise is only valid for new _machines_, bought after Dec 1st 2010.
Which basically means if you have a current Montvale/Montecito based machine that can be 'upgraded' to use Tukwila then the actual machine have been bought before 1st of December.
So the wording suggest that upgrades of processors goes free, as you are not buying a new machine.
Furthermore the licensing scheme doesn't make much sense, just look on how the 'per core' licenses for the T series processors seem to go up and down
T1 -> 0.25
T1(+) -> 0.50
T2 -> 0.75
T2(+) -> 0.50
T3 -> 0.25
"So basically this price raise is only valid for new _machines_, bought after Dec 1st 2010."
Your point? So? It makes sense to protect your customers investments. I think Oracle should be applauded.
"Furthermore the licensing scheme doesn't make much sense, just look on how the 'per core' licenses for the T series processors seem to go up and down"
Seriously? You don't understand? Only a marketing droid would have trouble understanding this...
Do you really need it explained? I will explain, but I'm guessing that you really do understand... but then, what's your point?
"Your point? So? It makes sense to protect your customers investments. I think Oracle should be applauded."
Sure, I am not complaining. I am simply sharing an observation with the readers here that might save them a lot of money. One that I am going to use in my role as an architect that does Oracle solutions.
Hence if you are buying a new Oracle License and are going to put it on a Itanium machine, then upgrade one of your existing machines to Tukwila rather than go buy a new one. I mean the savings can be easily be in the hundreds of thousands USD.
"Seriously? You don't understand? Only a marketing droid would have trouble understanding this..."
Well software licenses usage is meant to reflect how much capacity or clicks or whatever term you want to use, you are using the software for.
What doesn't make sense to me, as an oracle customer/partner, is that there haven't been a consistent mapping between T-processor throughput/capacity and the number of licenses you'd have to pay, for the five years that the T series machines have existed.
Sure if I were a 'marketing drone' I'd say Sure it's cause they wanna sell new hardware, but that is not my focus. I couldn't care less. But this constant changing of licenses, does take up to much of my time, discussing with our clients on how to protect their Oracle SW license investment, and minimize their cost. And sure that means that I am sold as consultant for quite some hours, but I should be using my time on making solutions.
I told you that Larry will make sure customers switch to Solaris. I wrote something like:
"Sun had 30.000 customers. Oracle has 350.000 customers. If Oracle can make a tiny percent to switch to Solaris, then the Solaris trend is rising upwards.
Earlier, there where only technical advantages to use Solaris, but Larry will make sure there are also business advantages. He will also lower the pricing for SPARC servers to make everyone switch"
But two days ago, Larry officially said he has great respect for IBM and their excellent products, but he mocked HP and said Oracle is after HP.
Cazzo posts, "Can I get a workstation in a pizza box case with one of these chips please?"
Take the crippled chip, with bad cores, and stick them in a pizza box.
Maybe something daring, like a Octagon Pizza Box, to fit into a corner of a desk, support 3 1080p monitors out-of-the-box, bluetooth keyboard & mouse, multiple internal disks, and option for a touch-slate control panel on a 4th disconnected monitor.
Target Oracle database & systems high-end DBA desktops - every customer will need one for their senior DBA or CIO who wants to see all the metrics for his/her global systems.
OK, so it looks like Larry and Co twigged to what the customers were saying over the last few years - multi-wheiner-core and multi-threaded Niagara was not giving them the throughput required because the majority of their business apps still needed strong single-threaded performance. It looks like the webserving niche where Niagara shines just isn't enough to keep Niagara alive. It also looks like Larry doesn't expect the majority of business apps to be going to the multi-threaded nirvana the old Sunshiners told us they would, otherwise why would he be trying to make Niagara look more like Pee7 and Tukzilla?
So, is this the redesign of Niagara to make it a better able to meet the x64 competitors, or is this squeezing some Rock tech into Niagara in an attempt to get out of the x64 space and make it viable against Pee7 and Tukzilla? The latter implies that Larry is looking for a replacement for Fudgeitso's SPARC64, which does seem to be fast approaching a development dead-end, but I'm not sure a Frankenstein-like hack-job of Niagara with bits from the Rock corpse is a viable answer.
Welcome Matt. Your snarky style has been missing from the discussion. You imply that Niagara must change it's goal of throughput to meet the further demands of single point performance. Does this mean that when Intel and IBM add cores and threads they are abandoning single point perf? Seems like a fairly naive assumption. As David points out above this has been in the roadmap for some time and the designers have been designing it for some time.
Just cause the changes to the T processors have been on the drawing board for quite some time doesn't mean that the overall observation that Matt has made isn't right.
It looks like the T4 is moving towards fewer fatter cores with better single threaded throughput. You might even say that the T4 core is moving in the same direction as POWER7 and Tukwila cores. With Tukwila being the pure single threaded monster it is and POWER7 being the throughput/single threaded hybrid.
And the fact that the T-core is going that way, rather than going to 24/32 or whatever cores on a chip, might very well be that SPARC64 is going to be replaced by T5/6.
I don't think that Venus looks like replacement for the current 4 core SPARC64 VII+ who has SMT and runs at 3GHz at a 65 nm. process. Venus is 8 cores with no SMT and it only clocks in at 2GHz at a 45 nm process. Which basically means that for commercial throughput workloads then it would more or less tie with the current VII+ for throughput.
Again it only strengthens Matt's point.
Matt does have a point, but it's on the top of his head.
I'm not really sure what's wrong with T5 replacing SPARC64. SPARC is SPARC. It's not like Power or Itanic that require a recompile. It's all just SPARC. If it runs faster, then that's all just good for Oracle customers. It's like saying that IBM should not come out with Power8, cuz that means the end of Power7... what?
Who's following who into the singlethread or multithread market, who cares? IBM and Intel chose to keep pushing thread perf. Sun chose to ignore thread perf and go for many threads. Now IBM and Intel are adding more threads, while Oracle is increasing thread perf. They both end up in the same place. Who cares how they got there, other than FUD throwers like Jesper and Matt?
"....Your snarky style..." Actually, it's more of a "falling-off-my-chair-laughing" response to the Sunshine.
"....Does this mean that when Intel and IBM add cores and threads they are abandoning single point perf?...." No, both Intel and IBM have improved single-thread performance whilst squashing more cores into less sockets. The original Niagara design was many more cores but they had very poor single-threaded performance, but then Niagara wasn't meant to compete with Itanium or Power, that was Rock's job. When Rock crashed and burned, Sun had to fall back on SPARC64, a decent chip but getting very long in the tooth. Now it looks like Larry is resurrecting the idea of beefing up Niagara into some kind of semi-Rock in an attempt to compete with Pee7 and Tukzilla, which implies SPARC64 is dead soon.
".....As David points out above this has been in the roadmap for some time...." Yes, that would be the roadmap that so underimpressed Sun's customers that Sun's sales dived into oblivion. Larry has taken his time getting a roadmap out and has then done nothing more than change the logos on the old, failed one. Whilst you may think the Sunshine factor will mean companies will blindly fall into line, the recent experience of Sun's decline is liable to make the current Sun base rather more wary. It will also do nothing for all those customers that have already started or completed migrations off Sun kit - they aren't going back for another system's lifecycle at least, so that's three-plus years their new vendor has to make them comfy before Larry even gets another look in. I said a while back that Larry had about a year max to save the Sun hardware bizz, and so far, in my opinion, he hasn't done nearly enough. Simply revisiting the Sun graveyard and dusting the corpse with some cheap Oracle licences does not look like a winning strategy.
/SP&L < did you miss that too?
Well, Niagara T3 has already several world records in several benchmarks.
And, it broke the TPC-C world record too, with 30.2 million tmpc. IBM has 12 million tmpc, and HP has 4 million tmpc.
I would hardly say that T3 needs to catch up. It is the other cpus that need to catch up on Niagara.
"....you make a lot of mistakes.
Niagara T3 did not break any record. It was 108 Niagara T3s. Single one of them is too slow to beat any competitor...."
It seems that you dont understand what I write. What kind of mistakes do I make? I explain it again, just for you: "T3 has several world records" and provide links if you question what I say:
Look at "References" at the bottom
It can look like this:
Biting the hand that feeds IT © 1998–2019