Intel Kills a Top-of-the-Line Processor

itwbennett writes: In June of this year, Intel announced a processor branded as Broadwell-C. Now, the company has confirmed that the part was cancelled but would not give an official reason. Why did Intel kill the Broadwell-C? ITworld's Andy Patrizio speculates that it's a 'combination of increased cost, lower yield and potential product cannibalization' — cannibalization of the company's newly-launched Skylake processor, which the Broadwell-C outperformed.

Comments

Reminds me of Jupiter

[lophophore]

https://en.wikipedia.org/wiki/...

DDR4

[Holi]

Broadwell lacks ddr4 support. That is as good a reason as any to let the line die. DDR4 is out now and no one wants their flagship model to be relying on last years tech.

See a trend?

[Carewolf]

So the fastest Skylake is not as fast as the fastest Broadwell which was not even as fast as the fastest Haswell.....

Damn it Intel, get the high-end Skylakes out.

Real shame

[MatthiasF]

To have a fairly powerful OpenCL GPU and tons of eDRAM on chip, it could have been a pretty powerful parallel computing platform to pair with GPGPU.

They don't want Skylake to be fast

[Overzeetop]

i7 series parts top out at $1000, Xeon E5-4xxx series parts start at $1000

Why would you want your cheap consumer grade hardware cannibalizing your bread and butter business chips? The large cloud providers have already shifted to "consumer" hard drives to save money, knowing that their failure rates will be more than compensated for by lower unit costs.

Re:They don't want Skylake to be fast

[danbob999]

Because more than enough corporations will buy the "corporate" version of a chip even if it's slower and more expensive than the toy-like "consumer" version.

Re:They don't want Skylake to be fast

[ShanghaiBill]

The large cloud providers have already shifted to "consumer" hard drives to save money, knowing that their failure rates will be more than compensated for by lower unit costs.

Consumer drives do NOT have higher failure rates. The myth that "enterprise" drives are more reliable has been debunked by research done by Backblaze and Google.

Re:They don't want Skylake to be fast

[Anonymous Coward]

If you don't compare apples to bananas the picture is slightly different.
Let's take i7-5930K vs. E5-1650v3.

i7-5930K: 6C/12T, 15MB L3, 3.5/3.7 stock, fully multiplier and bclk unlocked, $594 boxed.
E5-1650v3: 6C/12T, 15MB L3, 3.5/3.8 stock, fully multiplier and bclk unlocked, ECC support, >64GB memory support, $586 boxed.

i7-5960X vs. E5-1660v3? Same deal.

Did I mention plenty consumer X99 boards support the 1xxx Xeons and e.g. ASRock Extreme also supports registered and reg. ECC DDR4 with em?

... yeah.

Re:They don't want Skylake to be fast

[CODiNE]

I can already feel the angry replies coming at your post.

The reason people react so strongly to this is that they are confusing two separate issues.

1. There are physical changes and firmware changes that can make one hard drive more reliable than another.
2. Companies are truly selling premium drives using these changes.

So yes, in fact, it is indeed possible for companies to create special enterprise-focused drives with carefully tuned hardware and firmware which will have a greater MTBF than consumer drives.

However, collected evidence seems to show that they are not in fact selling drives that show the expected increase in reliability.

Therefore, the assertion being made does not fly in the face of physics, nor is it saying that better drives cannot be made or do not exist.

The assertion is that enterprise drives are a case of false advertising or perhaps enterprise-focused drives that are not tuning the proper parameters. Maybe their engineers are making real changes in the hardware that simply doesn't create the more reliable product they thought it would.

When in fact, enterprise-focused drives do show a real world advantage in MTBF, then presumably back blaze and others will update their published data and reward the companies and drive models that show such improved performance. At this point however, they're generally not worth the extra money.

Re:They don't want Skylake to be fast

[Anonymous Coward]

Even though you might be right, why should anyone care? The matter of fact is: Hardware may break and you better be ready for it. It usually means using raid or other forms of redundancy. The only reason one should, imho, have in calling something enterprise is a featureset that is different (e.g. management consoles, high performance, hw redundancy).

Re:They don't want Skylake to be fast

[OdinOdin_]

Yes I'm sure you are correct, but... the lower TCO is in using consumer drives, they have lower replacement warranty periods but they must actually be lasting significantly well enough that the cheapest cost per month to ownership is in consumer drives.

This presumes you have factored in costs to replace, diagnose, deal with issues that might crop up more often due to partial/complete failure in units. I guess the mean variation is within 150%, when the consumer drive is 2 year warranty, the cost of replacement doesn't seem that high if you are doing it every 3.5 years.

Re:They don't want Skylake to be fast

[swb]

I wonder if the "enterprise" drives aren't just tuned for performance advantages when used in specific controller or SAN configurations. MTBF may be improved not because of specific advantages in actual mechanical reliability but in fault reporting schemes that allow the controller to better evaluate whether the drive has truly failed and to adapt to smaller scale faults versus failing the entire drive.

"Consumer" drives may have firmware which flags some errors more easily because common use cases can't adapt to some kinds of faults intelligently, so the only sane thing is to just fail the drive outright.

It would certainly be in third-party Enterprise vendor interests to be able to keep using failed drives. I had a customer who was losing SAN drives about every four months for about a year. The SAN vendor issued a drive firmware update and since then only one drive has failed in 2-3 years.

The whole MTBF/TCO question makes me wonder why we aren't seeing big flash arrays using modern consumer SSDs. The write endurance tests that appeared on Slashdot showed even older generations surviving extremely high numbers of writes and drives like the Samsung Pro series I believe have 10 year warranties.

If I was a Samsung I'd be paying someone to put together arrays using my consumer drives to see how viable they were in enterprise SAN configurations.

You bastards!

[DougOtto]

You killed Broadwell. You bastards!

Re:Maybe they found a backdoor

[denis-The-menace]

FYI: In theory, all newer Intel chips have Backdoors:

http://libreboot.org/faq/#inte...

Re:Is this news?

[Austerity Empowers]

I think it's news because many people feel the Broadwell-C was the better chip and that possibly SkyLake would be eclipsed. The things you can do when you're a monopoly are bad for the customer. I

Lies

[Tailhook]

Benchmarks of Skylake i7-6700K vs Broadwell i7-5775C show the Skylake CPU to be faster. Cheaper too. The Broadwell chip can perform better on some OpenCL tasks due to the Iris Pro integrated GPU, but non-GPU tasks handled by integer and floating point units, cryptography and media extensions are always faster on the Skylake CPU. The "which the Broadwell-C outperformed" part is stupid sour grapes from an unhappy little malcontent.

Re:Lies

[aliquis]

Skylake outperformed in doing the same instructions over and over in benchmarks but Broadwell outperformed in games.

Re:Lies

[Kjella]

Yes, pretty much the only thing it did better was integrated gaming though at a $2-300 premium you could use that for a significantly better discrete GPU. The only thing it was good for was a stylish AIO where you couldn't fit anything bigger, my guess is Apple didn't want it for any iMac so volume would be too small. Also Skylake is really small when it comes to die size, so I guess the profit margin is actually better than selling Broadwells.

Re:Lies

[rtb61]

Perhaps they might make a much smarter move like redesigning the CPU. Discrete GPU process, "hey I noticed you have a separate GPU doing the work I was designed to do, do you want me to do something else, like crunch numbers or do AI stuff".

Re:Lies

[aliquis]

Not at the same speeds.

At first this made me confused.

But yeah, Skylake is clocked higher and that's likely mostly why it beat Broadwell in benchmark tasks whereas Broadwell have more cache and and tweaks which obviously made it better for games even though lower clocked.

The speed / MHz I wouldn't call important though because it's a simple fact that Broadwell i5 and i7 C-cpus are lower clocked and that's how it is. What matter is how they actually perform not "but if!"

I don't agree they are poor over-clockers though. Swedish Sweclockers thought they was but they only looked at the resulting clock-speed and yeah, the 6700K can likely do 5 GHz something the 5775C likely won't. But the 6700K stock is 4 GHz whereas the 5775C is 3.3 GHz so in percent they both over-clock quite well and likely by about the same amount/percent.

How much (as a ratio) you can over-clock should be what's important, not what frequency you end up with.

Re:Lies

[aliquis]

In the comparison I saw, Broadwell-C outperforms by an inconsequential amount in games that are CPU bound (5%). So yes, with Broadwell-C you can pace Civ 5 etc at 125fps instead of 119fps. Those are meaningless numbers from a playability perspective.

I wouldn't call them meaningless. Regardless no-one said there was a huge difference or that you should bother.

Even the i7 4790K isn't far away from the i7 6700K Skylake one so you could even go with that which is also cheaper.

And speaking about prices and old models the i7 5820K would likely give 6700K a run in modern game titles too.

If you go further back the i7 4770K, i7 3770K and possibly the i7 2700K isn't terrible and far away either, per generation at least. You likely got like less than 10% of extra performance for each new one. And that's when you ignore 5775C (as in 6700K not being 1.1*1.1 times better than the 4790K.)

That a supposedly uninteresting chip with better integrated graphics clocked at 3.3 GHz can outrun the newer chip on the new platform clocked at 4.0 GHz in gaming was a surprise for people.

Speaking of over-clocking and the 5775C vs 6700K the 5820K likely over-clocks very well as-well (The metal capsule/lid over it is supposed to be soldered on that one if I remember correctly, the 8-core one over-clocks nicely and the 6-core one should have an easier time I suppose unless lower quality.)

Re:Lies

[aliquis]

$200-$300 premium?

I7 5775C in Sweden: 3949 SEK.
I7 6700K in Sweden: 3629 SEK.
Difference: 320 SEK = $39 - VAT = $31.2

So more like $20-30 premium...

And that's before you consider the Z97 + DDR3 vs Z170 + DDR4 difference.

Asus Z97-P 790 SEK.
ASUS Z170-P D3 1129 SEK
ASUS Z170M-E D3 1012 SEK

Z170 = 222 SEK more expensive.

2x4 GB DDR3 474 SEK.
2x4 GB DDR4 545 SEK.

i7 6700 + Z170M-E D3 + 2x4 GB DDR4 = 5186 SEK
i7 5775C + Z97-P + 2x4 GB DDR3 = 5213 SEK
Difference: 27 SEK = $3.3

Re:Lies

[Kjella]

$200-$300 premium?

I7 5775C in Sweden: 3949 SEK.
I7 6700K in Sweden: 3629 SEK.
Difference: 320 SEK = $39 - VAT = $31.2

For the integrated graphics, for gaming you're better off with a <$100 CPU and a >$200 GPU instead of the 5775C. Basically you are paying way too much to have the graphics integrated into the processor. I wasn't comparing it to the 6700K, it's a significantly better CPU but you need to pair it with a dGPU to match gaming performance and then you're in another budget category.

Re:Maybe they found a backdoor

[Hadlock]

It's called Intel vPro technology, which lets you run a hardware session of VLC from the BIOS, on the wall side of the power control. It's pretty impressive shit for enterprise but you can do a whole lot more with it obviously.

Memory exploits on the WinTEL platform

[nickweller]

How does this relate to memory exploits on the WinTEL platform?

One Source

[Princeofcups]

At one time the open software community was proud of porting their software to every hardware platform. Now people don't even know or care that there are alternatives to x86/x64 architecture. Nor do they know about the days when hardware shipped crippled, unless you paid the upgrade cost to remove a jumper. I fear that those days are returning.

Re:One Source

[Anonymous Coward]

Everything is locked down these days, including increasingly x86/x64. Thanks, smartphones.

Re:One Source

[laffer1]

Two events happened to kill interest in SPARC and POWER:

1. Apple switched to Intel. Macs were cheap, lowend POWER systems. Many ports to IBM hardware started as Mac ports.
2. Oracle bought Sun. The few remaining workstations disappeared. The cheap 1u servers from the .COM era were killed.

I have two Sun Netra T1 servers in my basement. I used to support SPARC64 on MidnightBSD and even had bought some used Ultra 10 systems for that purpose early in the project. However, it was hard to find parts that were cheap and reliable to keep them running. Then oracle shutdown free access to documentation and firmware updates. I gave up. I haven't powered on the systems for awhile. My goal is a desktop system and without workstations, there is no point in making a desktop OS for a line that only has serial ports.

Re:One Source

[drinkypoo]

Power 7 and Power 8 chips outperform top-of-the-line Intel chips. Pretty much through brute power--more hertz and more cache--but they're faster nonetheless.

I'd love to own a Power box, but they're too expensive,

They're horribly, incredibly expensive. Nobody but nobody is more proud of their kit than IBM any more.

Re:One Source

[thegarbz]

Nor do they know about the days when hardware shipped crippled, unless you paid the upgrade cost to remove a jumper.

This still happens and not just for IT. The difference is it's no longer a jumper but rather some discrete hash or hard coded settings.

Re:Is this news?

[alvinrod]

It doesn't make sense for Intel to kill Broadwell-C when they can just charge more for the performance premium. It's not as though Intel is under any obligation to sell the chip for less money because it's not the newest technology.

I'll still be waiting

[AndyKron]

Doesn't matter how fast the processor is, I'll still be waiting, waiting, waiting.

yield problems

[edxwelch]

Iris Pro was first introduced with Haswell with much fanfare. There were both as desktop and laptop versions. But in reality the products barely existed. There is only a handfull of design wins, which sold in very small quantities.
Now, we have a product based on Broadwell Iris Pro that's cancelled. At this point it's looking like Intel were having major yield problems with the production of the on chip eDRAM.

Re:Maybe they found a backdoor

[qpqp]

a hardware session of VLC from the BIOS

Awesome, so now I can watch your pr0n collection remotely. ;)

It's a K

[Blaskowicz]

What's more, the chip is called Broadwell-K not Broadwell-C. Even though model numbers end in C.

Re:Maybe they found a backdoor

[Blaskowicz]

My linux installation has backdoor software named something like "ssh", I'm extremely concerned that they would let such a glaring hole through. What can I do about it?

Better question

[Moof123]

A better question is why we have plateaued on performance so badly. 7700k vs 4770k is a wash at best after 2 years for power and performance (way less than a Moore's law cycle would lead you to expect). Consumer grade processors are stuck at 4 cores, and now we get to pay for a bunch of low end GPU die area that will never get used. I don't get it.

Give me a 6 core with no GPU over a 4 core with a low end GPU any day.

Re:Better question

[ark1]

Why? Lack of competition.

Re:Better question

[esperto]

Simple, no competition, AMD is almost dead and can't put out anything close to what intel can do, and intel it self will not drive prices down on a monopoly, so we will get same performance at the same price just with a different sticker, until the day mobile processors actually can get close to what a desktop can do.

Re:Better question

[aliquis]

ZEN is supposed to increase IPC speeds by 40% at the same clock. It will also get SMT but I guess that may already be accounted for.

Whatever those clock speeds will be and how many cores they will have we don't know. Manufacturing process likely 14 nm in both cases.

AMD R&D budget is much smaller though.
Though Intel also build their plants.
Both invest in graphics but I guess Nvidia graphics R&D budget may also be higher than AMDs? =P

Suck to be AMD.

Re:Better question

[GigaplexNZ]

AMD usually overpromises, but even if we assume they get a 40% IPC improvement, it still won't be enough to catch the current Intel chips. At least it's a step in the right direction.

Re:Better question

[eWarz]

Wait, there is a 7700k now? Oh no, there isn't. Just like this article, it's a complete fabrication.

Re:Better question

[mister_playboy]

Correct, but the X99 platform is going to offer much lower performance per $ spent for most users, even most power users.

Make sure you have a use case for 6 or 8 cores before you buy in.

Intel denies chip line being killed

[Fallen Kell]

Intel responded to Anandtech's inquiry into the killing of the chip line and denies that it is dead and in fact is wondering where the bad information has come from:

http://www.anandtech.com/show/9639/the-death-of-intels-broadwell-is-greatly-exaggerated-socketed-broadwell-continues

DDR4 when there's GDDR5

[tepples]

DDR4 is out now

I thought DDR4 came out in 2000, and the PlayStation family was up to GDDR5.

Re:DDR4 when there's GDDR5

[aliquis]

thought DDR4 came out in 2000, and the PlayStation family was up to GDDR5.

First one was a joke on your part.

I don't know about the second one but there exist no spec for DDR5, GDDR5 isn't based on and improved upon DDR4. It's based and improved upon GDDR4 which in return is on DDR3.

One is the chimpanzee the other the human so to speak.

Re:Is this news?

[tsotha]

It isn't news because it's pure speculation. There's no support for the idea Intel killed it for nefarious monopolist (as if Intel were a monopoly) reasons. If they were really playing those kinds of games they would never have greenlighted the project to start with. It's far more likely either the project wasn't meeting expectations or they have some nearly-finished technology they want to incorporate into the next top-of-the-line part.

Re:Maybe they found a backdoor

[_merlin]

vPro also doesn't work at all on my Dell Precision workstation. It's supposed to be supported but they must have fucked something up in the implementation as it's completely impossible to enable.

Re:Is this news?

[GigaplexNZ]

They didn't, IT World got it wrong.

Yet another post from timothy that won't get fixed

[nyet]

Yet another post from timothy that won't get fixed

Re:Maybe they found a backdoor

[cfalcon]

Can this be disabled?

Re:Maybe they found a backdoor

[cfalcon]

On Linux, you can kill sshd and disable it so it won't start on your next boot. Depending on your security needs, this may already be the case.

How can I make my Intel processor not take external commands? How can I "turn off the daemon"?

Re:Maybe they found a backdoor

[cfalcon]

AC:
> The motherboard also has to support vPro technology for the backdoor to work. vPro support only in the CPU will not do much. Most gaming and consumer motherboards do not support vPro.

So look for a mobo without "vPro", or disable "vPro" in teh BIOS?

Re:Correction

[TechyImmigrant]

According to the source the original article has been corrected. The Skylace-C was killed not the broadwell-c.

But they seem to have done a search and replace. The logic of the article evaporated.

Please correct the summary!

[kelemvor4]

From: TFA: CORRECTION: Due to a reporting error, an earlier version of the story misidentified the product that has been retired. It is the Skylake-C and not the Broadwell-C that has been discontinued.

Why should they

[Kuruk]

Not like AMD is releasing a faster processor. They can cool their jets and profit anyway.

Power.

[DarthVain]

The reason is changing product uses. The reason there has been little progress in performance is that chips today are being optimized for power rather than performance. The reason is most people buy laptops now, which have a battery, which means power conservation is a bigger priority than ever increasing performance to which only a very small subset of users actually ever utilize.

Back in the day, you just poured more power into the chip, and so long as you could cool it enough from melting you didn't really care how much power it sucked down as you were typically plugged into a wall outlet.

Now with most consumers using laptops it just isn't a priority anymore. Considering for most performance hungry users the end use was computer games, and on a laptop you likely aren't going to have a discrete GPU except at the very upper end, there is little use for CPU performance either.

I don't think it is because they *can't* achieve greater performance, it's just that it isn't the market they are building these things for. And before you start talking about various product lines, these things are all designed at the same time, and the overarching goal is reduced power consumption while keeping a reasonable amount of performance. As a desktop user it's depressing. The same can be said for all the emphasis on CPU/GPU integration and its marginal gains, all driven by the ubiquitous laptop.

Re:Correction

[nyet]

See also http://slashdot.org/comments.p...

/. has never been good about errors in their articles. Not sure what the editors do, but editing /. stories is definitely not their primary job.

Re:Maybe they found a backdoor

[denis-The-menace]

If you do, the CPU won't work at all b/c it can't load the Binary Blob to fix mistakes in the CPU factory.