anyone have a theory about sparc / the current project ?
how the current layoffs rumors are related to Sparc development or the Hardware business in oracle in general ?
All companies are doing their own hardware ( amazon , google ... etc ) it would be stupid to give up
doing your own hardware if you are trying to compete or even try to be a top player !!
@OxoWWr4-6mxx These top supercomputers (you mean Top500 list?) are "scale-out". Well, simple HPC nodes. They works as separate processors. Purpose of interconnect is just to give them enough work, and then (few hours or months later) collect results.
But in your database server you need proper scale-up solution. Just look how strong is interconnect in Oracle Supercluster, all supported by silicon. Nothing in x86 world can match it.
@OxoWWr4-5vos SPARC has some good ideas even today. Software in silicon. Secured memory. Real time encryption. There is no x86 processor which offers that.
On the other hand, most recent x86 extension is AVX512, which is good for video trascoding (in Youtube datacenter) or science computations, but it has no utilization in your database server.
You know, because there is difference between database server, datacenter and HPC cluster.
And don't get misleaded by Intel's revenue in their datacenter group. It is 15B today which is significantly more than Oracle's 4B, but there are also some problems.
From HPC side they are beign replaced by Nvidia GPU's and FPGA's. Not on in AI, but also for general purpose computing.
And from other side, there is ARM threat in datacenters. Qualcomm already has 10nm server ARM and they are delivering it to some customers today, if these customers will adopt it as low cost alternative to x86, then it can shrunk entire Intel's datacenter group.
Because, if their revenue will fall below let's say 10B, they will lose interest in developing new architectures. And will happen something similar to others divisions which didn't met expectations. For example ITANIUM, mobile, iot... Lack of inovations, move to older manufacturing nodes (to save cost), rebranding old processors...
And everyone who relay on x86 will be ducked just as Itanium customers back then.
@OxoWWr4-6dlc "They have better top to bottom control over engineered systems if they own X86 development." ... exactly which part of x86 dev Oracle owns?? last time I checked, x86 dev is fully controlled by Intel (and, yes, also AMD for a small fraction)
"There is no alternative to scale-up M7-4 and M7-8 servers. Only machine able to scale-up like that is IBMz."
I doubt that. They make the world's top supercomputers based on X86 architecture.
Markets dictate the sweet spot for # of sockets. So if they don't make it in x86, there aint much of a market for it. Supply and demand.
You clearly don't know how x86 cross licenses work nor do you appear to know Suns history with x86 dev.
They have better top to bottom control over engineered systems if they own X86 development.
SPARC had some cool ideas and innovations back in the day, but those days are long gone and you need to accept reality. X86 is by far the mainstream data center building block. While that might not be true down the road as ARM, NVIDIA and others make strides, it is KING today. AWS, Azure and other cloud platforms all depend on x86. You can't argue with that. SPARC is dead for all practical purposes. If it survives, it will only do so as a shell of it's former self, in a small niche and only by the graces of large legacy customers, or the government (think infiniband).
@OxoWWr4-5vos Oracle x86 will not survive. There is lot of other companies, which offers better x86 solutions for better price.
@OxoWWr4-5vos Someone already writed it: SAPRC makes more than half of hardware sales. You must add SPARC deployed and offering as cloud. And also, every SPARC is selling with some database license...
I don't know from where is rummors that Solaris ends this time. Last time i heard that it turns out, that there will not be Solaris 12 anytime soon, but development continues as 11.x releases.
TCO of SPARC is better. Power draw per single CPU is higher but total troughput is also higher. Also SPARC is cheaper to buy (since there is no extra supplier with 60% margin).
There is no alternative to scale-up M7-4 and M7-8 servers. Only machine able to scale-up like that is IBMz. There is no other CPU with software in silicon. M7 is able to process SQL at 160GB/s comprimed and encrypted database. There is no other processor able to do it. x86 peaks on fraction of that. So by dumping SPARC they will lose some very high end market.
Software ecosystem matters for example for personal computers or workstations, but these machines are selling as database servers and all you need is good database to run SQL and some other analytics. Off course, software matters even here, but if you need something extra then it will be written in Java, PL-SQL or other similar - interpreter - languages.
And even if you have some application in C++ or assembly for x86. Rewriting and optimizing it for SPARC need similar effort like rewriting it and optimizing to new x86 ISA extension
Also there is some reason, why other companies are putting so much effort into developing their own architecture. Why Microsoft want ARM PC? Why HPE is building The Machine on ARM? Why Google want Open POWER? Oracle already has all the advantages today and it absolutely does not make sense to save few money* by dumping certain part of their hardware division.
*not much because hardware developers will sty since Oracle is developing ASICs and some Solaris developers will have to stay because they will have to optimise linux kernel for their needs.
@OxoWWr4-5nix and @OxoWWr4-5mot
yes it's the end of Oracle sparc hardware fur sure, x86 might survive. Whatever is left in SPARC HW/SW down the road is only kept around to dangle in front of legacy customers and maybe hook them in for more $ down the road, but the gravy train will end once the ROI inverts. With the recent news, I would be looking for a transition/migration strategy if I still run on SPARC.
Without SPARC they are just generic supprier of x86 PC's
SPARC has nothing going for it other than being power hungry. Sun blew it when they had control and Oracle kept blowing it. If you want to build a newclear power plant behind your data center with SPARCs, more power to ya (pun intended).
Also, without software, HW means nothing. The recent demise of SOLARIS and the late term abortion of Linux on SPARC is more proof in the puddin that the game is over.
I am surprised they aborted ZFS though. That could have been a very useful piece for their cloud strategy.
X86 is the main stay in the rest of the world. Nothing wrong with that. SPARC is dead, let it go.
@xoWWr4-5mot
You mean it's the end of Oracle hardware which is a small part of the Oracle behemoth.
@OxoWWr4-5ziz ...if it is true, then it will be end for Oracle too. Without SPARC they are just generic supprier of x86 PC's, and since they will not be able to compete againts HPE, DELL... they will not last long.
what is the future of sparc ?
dead, none, get out while you can
Looks like M9 is on the chopping block . Meetings behind closed doors is going to kill it before it sees the light .
Reminds me of killing Obama care with no replacement . Anyone know if they have a plan for a different project or the HW business is done !
What about M8?!
Oracle
Oracle’s Safra Catz: We Didn’t Buy Sun to Sue Google
Reuters
May 16, 2016
Credit Oracle for giving it's (Sun) Systems Group more than enough rope,
Given the circumstances of the acquisition no one would have faulted them
for ending SPARC and Solaris right after the deal closed.
The writing has been on the wall for several Quarters.
There are limits based in economic reality. Stockholders have
become impatient. Markets have changed and business plans have
had to be altered and Cloud Migration accelerated lagging behind
the growing competition necessitating more acquisitions.
As recently reported by the press Oracle layed off many employees
primarily from VP Fowler's (Solaris/SPARC) hardware systems division,
@OxoWWr4-2ryb Based on, " last year x86 managed to get close (mainly with Exa) but SPARC is still more than half.", SPARC HW revenue seems to be between ~$2B to $3B (out of total HW revenue of ~4B), which is not a tiny number (lot of units and customers), so very difficult to walk away from and hence need a plan/product roadmap to support these SPARC based customers for a long time.
So Oracle HW revenue numbers are comparable to AMD's ($4B) so Oracle's R AMD has a headcount of ~8200 (includes everything as in Engineering, Sales, Support, HR, Accounting, etc.). Before the recent reshuffle/reorg, just MH's SPARC Systems group had a headcount of less than AMD's total headcount, however, just for SPARC i.e. not counting headcount for x86 based systems.
Yes, $2B is totaly overshooted. Costs is $10k's to $100k's for masks and setting up production and then $500 - $5000 per manufactured wafer. All depending on mask count.
Most of money you spend is to pay people. From layout engineers, design engineers, verification engineers... to ISA analytics... all jobs are $60,000 to $100,000+ per year.
But oracle already has these people, and they will remain in, even if SPARC will be cancelled (for developing storage, networking ASICs or SWiS). So there will be minimal savings in canceling SPARC.
Btw.: as a example of CPU development costs you can take AMD. With revenue of $4B and R&D expenses just around $1B they are able to produce several products per year (CPU, GPU, APU, embedded SOC...) at very leading edge technologies (20nm, 14nm, 16nm, 7nm in development) in three different foundries (Samsung, TSMC, GlobalFoundries- which is AMD + IBM spinofs).
"I will state the obvious - its a matter of ROI. We spend $2b every time the wafer is re-spun."
Agree with ROI comment, in general; investment in time and resources (engineering/development, money, etc.) have to make business sense. However, not sure how and where the $2b (as in billion, I think) number is coming from. It costs, maybe $2 to $20m (million) to re-spin a wafer, depending on number of layers/masks, which need to be re-spun, not counting all the testing, validation, productization, etc., costs, before RR (Revenue Release).
I will state the obvious - its a matter of ROI. We spend $2b every time the wafer is re-spun. Without enough external demand, its difficult to keep respinning it. The brakes, even if we decide to end it, is not going to be obvious and immediate, it takes some time for the ecosystem to stop outputting another product in the pipeline.
Now, lets think from customer perspective. I put myself in an executives shoe, and Oracle tries to sell me exadata. They try to sell me cloud. I have choices from competitors, and these huge layoff rumors. What do i do ?
Large finance, govt institutes still need on-prem extreme performing DB machines, so exadata is here to stay as long as mainframes. But, that uses x86, except an odd-ball here or there.
So, going back to original question -- will SPARC last ? I cant see a reason, and Cloud, Storage Appliance (x86) and exalogic (x86) conspiring against it. With Solaris Dev teams RIFed, it cant be more obvious. We will be lucky to get another 1 or 2 major SPARC based h/w releases, but thats just because most of the work has already been done, we will pump it out. Most likely will be offered as a freebie for cusotmers buying our cloud, as i dont see anyone with sane mind investing in it from customer perspective.
Interesting opinions.
My sources saz, that Skylake and Cascadelake platforms will not be adopted by them neither, so which processors they will use?
Regarding x86 and SPARC share: x86 was used mainly in storage servers (now there is also S7). Database servers was combination of that. If x86 was needed, they provided it, but main platform is SPARC. In Superclusters (most lucrative parts) there is SPARC only for clear reasons.
Also big part of SPARC machines was deployed on premise (cloud).
...and, there are also for example security issues with x86... I almost forget it but it is important, since it makes it unusable for some applications.
Regarding performance, general performance is comparable to x86. x86 has advantage in wide vector units, SPARC has advantage in clock speed (2x higher under full load). So for example SPEC scores might be comparable. Real advantage of SPARC is in large scale databases, mainly because of better work with memory, DAX engines and scalability (scale-up). And here comes 2x-5x-20x advantage of SPARC. M7-4 and M7-8 has no match in x86 world, and if you need that performance, you have no choice.
Regarding software ecosystem, i see you noticed that there will not be big release of Solaris anytime soon. I also bet, that you noticed, that it is transforming to different model of update releases.
At the end i will try answer, why Oracle is building another SPARC. It is similar to, why HPE is building The Machine on ARM. Main reason is that Intel is too greedy, and it is cheaper to build your own processor. Second reason is that it is customized to your needs. You can integrate into it anything on the ISA level. It is something Intel will never do for you. And last, but not least, you are not dependent on Intel. They promised lot of thing but lot of it they did not delivered. Last thing with which lot of customers has problem is 3Dxpoint. Or silicon photonics (if you need silicon photonics in your ASIC, you have to ask TSMC).
All of RMs tools could easily be swapped out with industry standard tools from synopsis and cadence. They provide almost no benefit in an ML context. They'll all be gone.
Theyre people's initials for fool who is trying to participate in a conversation you clearly know nothing about.
And your question about volumes - those exist, are hard to collect, and obviously won't be shared by me or others On here.
What is MS / RM?
"MS's org will be 20% of its size after M9 is terminated. RM's group should be gone. RL's group may get lucky and focus on ML at a small fraction of its current size."
How can MS / RL's team can get anything done without CAD flow/tools help from RM's team ?
If M9 is terminated, won't the customers stop buying M7 and M8 based systems since there is no
roadmap for sparc chips (cpu's) and HW systems ?
expecting to sell in next 2 to 5 years ?
Amount of time and effort that was invested in developing M7 and M8 based systems,
there better be a huge demand (10's of thousands if not 100's of thousands units /
couple to few Billions of $'s worth) for these.
Does anyone have # of Units / $$ Number for Sparc based HW Systems (T4/M4, T5/M5/M6 and M7) sold after Oracle acquired SUN ? Oracle publishes HW revenue numbers in their quarterly / yearly reports but has never published the breakdown of x86 and Sparc based HW numbers; seems like those numbers are dominated by x86 based HW Systems.
Regardless of the "stellar" performance (single thread, multi-thread, throughput, benchmark numbers, etc.) of Sparc based HW,
if Oracle can't sell these in enough volumes with long term contracts, just to recover the development costs, let alone make profits, "winning" performance benchmarks is meaningless.
Bottomline, Best cpu/architecture/OS/HW/Systems are meaningless if they can't be sold.
I share a similar theory about RL s org continuing with ML if they survive. However, why do you think RL's org will survive but RM's won't?
Sparc design is effectively done. Most of the people left are for support for existing life cycle process. More layoffs will happen in Sparc HW over time until they decide to stop fully supporting.
"All these companies do not make hardware they buy it."
We're talking about TPU2 style products not x86 boxes. If you don't know what's being talked about don't comment.
They canceled Solaris and now Linux for SPARC. Without any software to run on the SPARC servers, I anticipate they'll cancel SPARC hardware soon too. Last week they reorganized all hardware systems to combine SPARC with x86.
All these companies do not make hardware they buy it. Same road as oracle will go they are getting rid of most hardware support before end of the year all Europe is gone except Romania just a bunch of newbies. So it will not be weird if at one point oracle will just sell off the hardware business.
Best case
MS's org will be 20% of its size after M9 is terminated. RM's group should be gone. RL's group may get lucky and focus on ML at a small fraction of its current size.
None of the listed companies in OP are doing unique core architectures. They do workload accelerator ASICs which do not require large teams.
With no Linux story there is no Hardware story. This will all become clear soon under current organization.
Ask around the SC office and see what people think.
That's a good question.
If anybody in the upper management had really thought enough about this, they would have taken the path you suggested. Why didn't they?
The answer is obvious once you look at how thE rest of Oracle's acquisitions have worked in the past. The rest of the acquisitions simply needed the buying part of the process for the product to start making money for Oracle. Granted that Sun was perhaps the largest acquisition, but even Sun had a few products (aka licenses) that saved money from day one. And the mgmt was too complacent to worry about the ones that needed some integration process. So basically, they didn't care enough.
The other point to notice is that - Sun wasn't making good processors when they got bought. They were bought after a string of failures. My suspicion is they took 5-6 years to catch up in hardware alone, by which time it became too late to start actually integrating hardware with software because by then software was flailing and trying to fight it's own battle in the cloud.
"Integrating" takes a ton of mgmt effort and vision- something at the level Jobs was capable of. I don't see any "leader" in Oracle with that kind of commitment or vision.
"Oracle doesn't have to do sparc to make its own hardware"
lets assume Sparc is not 3x or 5x faster , lets say its the same at 1x.
why wouldn't Oracle fully adopt its own hardware in its data centers for the cloud, and save a lot of money instead of buying the competitors hardware and pay premium price for it ?!
lets say there are so many issues and problems to do that. why there is no plan or strategy to go there since the savings are huge if you use your own hardware !
something is missing here ....
Oracle doesn't have to do sparc to make its own hardware. In fact, the one part of hardware that has consistently been making money is the x86 exadata line.
Don't believe what you hear about Sparc being 3x or 5x better at one application or the other. Customers see right through that sh--.
The crux of the matter is Sparc offers reasonable performance at an exorbitant price, not accounting for costs in re-development for Solaris ( I had some hopes when they started Linux but that is gone now). This, along with typical customer reluctance to change means that no new customer shall ever buy Sparc unless they got fooled ( Or through some form of corruption). That's simply not a sustainable model.
Given these factors, Sparc seems to have survived because Larry was afraid of publicly admitting that Sunday hardware was a bad bet. And as MH and SC take over, it looks like they don't want to let this farce go on for too long.
I don't have any insider info but my guess is they ll kill most of sparc eoy ( months after m8 announcement). If they were kind they would kill it right now once and for all but I don't see them doing that tbh.