Correct me if I'm wrong, but hasn't Intel canned all Broadwell desktop CPUs (including Broadwell-k) except for a 65W unlocked Broadwell Iris Pro part? Which, I guess is going to be Iris Pro and Broadwell-K at the same time?
Or am I reading the roadmaps wrong?
Haven't seen anything like that. Any links?
A lot of sites have been reporting that Broadwell-K (And H??) may be canceled due to power consumption issues. But not sure on reliability. But the rumor is that with Skylake coming, Broadwell is just too late to be worth putting more money into.
http://www.eteknix.com/intel-broadwell-k-running-too-hot-possible-cancellation/
http://egmr.net/2015/02/intel-debating-whether-broadwell-k-binned/
According to the official roadmap, Intel will have released Broadwell-K, Skylake-S in the middle of this year. Both are 14nm process, the former for overclockers, the latter for the mainstream user, divide and rule.
But now the bad news. As Broadwell family only exists on the desktop, Broadwell-K traced to overheating problems, the original set of 65W thermal design power has been unable to do so, it will reach a high frequency of about 88W, in fact one of the new models have been set in the 95W.
It is said that the problem comes from Broadwell architecture itself, but it can not be said to be a design flaw, this situation is said not unexpected.
As for the new 14nm process is not also have a responsibility, not yet confirmed, anyway, when he had just been born 22nm provoke a lot of trouble, it is easy to cause the processor temperature is too high, and this surprisingly similar.
In fact, Intel desktop processors in high-end models have been around 90W, the suddenly reduced to 65W, could have been a big surprise, the results back.
It stands to reason about 90W is not unacceptable, especially love overclocking generally do not care about power, but this is obviously not just a problem of high power consumption, not to mention the decline in consumption means high overclocking capabilities.
Allegedly, Intel executives are currently discussing whether to completely abolish Broadwell-K.
Taking into account the root causes of the problem, I am afraid to spend time is not completely resolved, but not the economy, so the possibility of cancellation or high.
Trouble is, we have not heard the news Skylake-K. Although mainstream version have, overclocked version should not be a problem, but now it seems, if you like overclocking, must wait longer to have the new stuff up.
Oh yeah, there are plans early next year, the birth of a new fever platform Broadwell-E, scheduled to 140W or whether it will face overheating problems?
Yeah . . . those links (and their source, mydrivers.com) look, ummm, reliable?
Your assertions are based on nothing. Kirk Skaugen already confirmed Skylake for desktop will come in H2. It has nothing to do with AMD. Intel will obviously not release products as long as it's an uneconomical node.The lack of information about Broadwell-K or Skylake (desktop) is rather frustrating.
I fear 14nm has severe problems for 4, 6, 8 core CPU's, as we haven't seen any of them yet, not even leaked ones.
We had leaks of Haswell, Devil's Canyon many weeks/months before release, it's looking more likely that Intel won't release any new desktop CPU's this year. I guess we have AMD to thank for that - Intel know they can keep making the cheap, high yield rate 22nm Haswells and still sell bucketloads of them. Why bother with 14nm yield problems on desktop parts, when there is absolutely no competition on the desktop?
Like I said, not sure on reliability. But since there has been little news on H or K, and no leaks of the chips themselves, kind of lends itself to that theory that we wont ever see them.
Your assertions are based on nothing. Kirk Skaugen already confirmed Skylake for desktop will come in H2. It has nothing to do with AMD. Intel will obviously not release products as long as it's an uneconomical node.
Intel released BDW-Y/U first because those series will benefit most from the die shrink: desktops will benefit most from the Tock, so it makes more sense to skip BDW-S.
And IIRC, we got information about DC (the SKUs for HSW-R) only after Intel announced DC.
Oh, I have no problem believing desktop Broadwell may get cancelled. I'm just not buying the reasoning of the article. As TDP is determined by Intel, claiming that chips are overheating to a higher TDP doesn't really make a lot of sense.
They may mean that in order to reach its performance goals, Intel had to set the TDP higher than originally anticipated. But the articles don't cite any other sources and the explanations all seem to be based around the TDP of certain chips. It appears more likely that the articles are speculation that is based on a misunderstanding of TDP.
Higher TDPs at same frequencies than Haswell.
There are already hints that point to higher TDP at higher frequencies, compare Haswell mobile SKUs voltages at say 2.9 to a Core M at the same frequency, BDW voltage is 12.7% higher, if the parasistic capacitance has not been reduced then TDP would be 27% higher at same frequency.
The unknown is how much the parasistic capacitance was reduced, reviews on Core M are not accurate enough to draw a definitive conclusion, but first impressions are that they ll be lucky if they manage to just get the same perf/Watt than Haswell on dynamic conditions.
Not only 3W less, also faster!
BOOM!
Higher TDPs at same frequencies than Haswell.
You still keep the lies going?
BOOM!
The articles didn't say anything about Haswell.
Also, query: we know that Core-M (at least at base frequencies) must use less energy than Haswell at the same clock. We know this because either (1) Core-M wouldn't work in fanless systems, or (2) fanless systems based on Haswell would have been out much sooner.
So, is your theory limited to higher frequency parts? As frequency increases, does energy consumption of Broadwell accelerate faster than Haswell?
My post was clear,
You are talking about exact roadmaps/launch dates.Kirk Skaugen also said that Broadwell would be released for desktop in Q1. It's March soon, surely if Broadwell-K was really releasing so soon, we'd have new leaks, information, engineering samples as we've had previously... We don't, so it's safe to assume this date has slipped, or been cancelled altogether.
Seeing as his statement about Broadwell's release date has slipped, it's also feasible that the Skylake date of '2H'2015 could also slip.
Honestly, I very rarely find anything you say clear. Usually I attribute this to the language barrier plus my limited understanding of the science/math involved.
That being said, can you answer my question with a yes or no?
Is it your belief that Broadwell uses more power than Haswell at higher frequencies and less power than Haswell at lower frequencies?
Or, is it your contention that Broadwell is fundamentally less energy efficient than Haswell at any frequency?
If the latter, how do you explain that, for the first time, Core processors are operating in a fanless form factor? Why did OEMs not create fanless Haswell systems?
My post was clear, at a said frequency power will increase as the square of the voltage ratio, at 2.9 HW need 0.976V while BDW ask for 1.110V, the ratio is 1.127 and the square of this ratio is 1.27, as said at equal parasistic capacitances it will yield 27% higher TDP and it s very unlikely that parasistic capacitance was that reduced because reducing the distances between conductors in one hand and transistors terminals on the other hand will increase thoses capacitances, capacitances between conductors are proportional to the distance, the closer the conductors the more the capacitance.
Your use of words makes me think that you don't quite understand the capacitance tradeoff because it's not as one sided as you imply and you seem to be using "distance" in the parallel plate capacitance model in a weird way.
EE101:
1) Wire to wire capacitance:
- What increases it: Process typically narrows the spacing between them causing capacitance to increase.
- What decreases it: Process also allows the wires to reduce in length and height thus reducing the cross section between the wire
2) Transistor gate capacitance
- What increases it: Process traditionally thins oxide thickness or increases dielectric constants to maintain better current control thus causing the gate capacitance to increase.
- What decreases it: Process also typically improves Ion/um which reduces the transistor width (due to drive strength) and length (due to process shink) which reduces the cross section and thus reduce capacitance
Agree overall but your post is not easily understable for whom does not have at least a basical formation.
All wirings cant be that shrinked, for high current you have to take account of electromigration, for signal related connections you have to take account of resistance that would increase the charging/discharging time constants of the driven devices.
Also Finfets have intrinsicaly higher input capacitance than planar, this was undoubtly compensated on their 22nm by a better transconductance that allowed to reduce supply voltage since a better conduction at a given gate voltage means higher Ion and the capability to charge the input capacitances faster.
I think that it s this part of their design that went underperforming, not that their process is bad but to me it s obvious that dynamic power is higher than with Haswell, i dont think that it s due to strain capacitances but to the transistors input capacitances as well as lower transconductance than expected, hence the higher supply voltage to increase gate voltage and hence Ion, this is correlated by the 12.7% higher voltage at 2.9GHz, indeed i have no other explanation for this voltage caracteristic.
I don't really understand your explanation and I haven't really followed the discussion..
As for the frequency-voltage curve, that's affected by process and design. Imagine a curve which represents all possibilities frequency and voltage design point that the CPU design team is targeting. If you pick one design (which represents the lowest voltage to hit that frequency) and start to adjust voltage/frequency from a fixed design (how much voltage do you need to hit a higher frequency, how much voltage can you drop for a lower frequency), that SECOND curve will always be suboptimal compared to the first curve. How's that for a plausible explanation.
Haven't seen anything like that. Any links?