[News] Intel Core i7-8086K Listed, First 5.00 GHz Processor

[StyM]

https://www.techpowerup.com/244673/i...-ghz-processor

Intel is commemorating 40 years of its 8086 processor, the spiritual ancestor of the x86 machine architecture that rules modern computing, with a special edition socket LGA1151 processor, dubbed Core i7-8086K. The chip appears to feature a nominal clock speed of 4.00 GHz, with a maximum Turbo Boost frequency of 5.00 GHz, making it the first mainstream desktop processor from Intel to hit the 5.00 GHz mark, out of the box.

The Core i7-8086K is more likely to be based on a special bin of the 14 nm, 6-core/12-thread "Coffee Lake" silicon, rather than being something next-gen or 8-core. The retail SKU bears the part number "BX80684I78086K." The chip will be compatible with Intel 300-series chipset motherboards. Pre-launch listings put its price around $486, which is along expected lines, as it's 70-100 EUR pricier than the i7-8700K. Intel could unveil the Core i7-8086K at the 2018 Computex (specifically on the 8th of June), alongside the first motherboards based on its Z390 Express chipset.

[Particle]

If the title is going to make a point of saying 5.0 GHz is a first, it needs to be qualified. "First 5.0 GHz processor from Intel" is accurate. If talking about x86 at large, AMD released a 4.7 GHz base / 5.0 GHz turbo product five years ago in 2013. If talking about computer processors in general, the earliest one I'm personally aware of would be IBM's POWER6 from 2007/2008. The point though is just that the title isn't accurate despite the article being so since it states "the first mainstream desktop processor from Intel to hit the 5.00 GHz mark, out of the box".

[madcho]

5ghz first processor should be base clock too.

[Metroid]

If the title is going to make a point of saying 5.0 GHz is a first, it needs to be qualified. "First 5.0 GHz processor from Intel" is accurate. If talking about x86 at large, AMD released a 4.7 GHz base / 5.0 GHz turbo product five years ago in 2013. If talking about computer processors in general, the earliest one I'm personally aware of would be IBM's POWER6 from 2007/2008. The point though is just that the title isn't accurate despite the article being so since it states "the first mainstream desktop processor from Intel to hit the 5.00 GHz mark, out of the box".

Intel wants a free marketing ad, it hits 5 ghz out of box with just one core, so is pointless, to be precise, all cores should hit 5ghz and stay at that frequency. As it stands, 4.4 ghz for all cores, so is 100mhz more than the i7 8700k, so nothing to see here.

[iddqd]

5ghz first processor should be base clock too.

base clock is completely meaningless these days, since your CPU will never actually run at the base clock. what you really care about is the all-core turbo.

[madcho]

it runs turbo 5 seconds ... overheat, then go base clock

[dengyong]

it runs turbo 5 seconds ... overheat, then go base clock

Do you own one of these? How long have you been testing it and under what conditions?

[madcho]

No, I don't.
I am electronic engineer, and I understand well the path they took to do this.
They increase base clock with temperature/power limits and sensors, maybe internal electronic noise or delays too.

They use an built in microcontroler for doing that, they have implemented some advanced/fancy profiles.
Same principle with throttle function, when temperature/power is too high, they drop frequency under base clock. It was done a lot in Pentium 4. The problem that intel was sued for selling processors that can't keep up with base frequency.
They done different after that with a new way and they called it "Turbo". Base clock is lower, and Turbo clock aren't garanted ...
Today real average frequency now is more and more about dependent on process, and even mobo design.



Yeah mobo design is still important now, less and less off course, but when it was old time of pentium 4, the power stage that drived the hungry beast wasn't as efficient as today. A lot of heat from the power stage for the CPU, warm up air close to the CPU, and reduced the processor cooling performance.

Power stages are much more efficient today. However on a mobo, the CPU socket placement is still important today. The path of the air affect cooling performances, and so the real average clock.


On first generations with turbo it was around 1 to 5 seconds. Now it's maybe better, not because technique is improved, but because
1/ much more cores in the CPUs, so just one core at turbo can last long, full cores on turbo (with a cpu burn tool) it will never last long of course. Yeah they designed the base clock based on full cores without throttle and a small margin. It's obvious that if that could do better with all cores they would have did it on base clock ...
2/ process is a lot improved, helps a lot on thermal. Core are smaller and smaller with a lot of dark silicon. Easier to dissipate more with dark silicon, you just surround the cores with dark silicon.
3/ architecture improvements; with reduced base clock for same performances (better IPC), means less dynamic power consumption means cooler CPU. By More the frequency is far from the invicible wall of 5ghz clock, more it's "easy" to get large turbo clocks.
4/ Improvements in data transmission on the chip. New technologies came since first pentium. serials links with differentials pairs are everywhere now. It's easier to improve clock when the full chip is not going at the same clock, but every block has it's own clock domain. Much easier to synchronize.

Maybe some parts aren't totally exact, it's difficult to have real technical data on this. You can bet it's garded like a graal in companies.

If you find a processor that can keep up over 10 minutes with all cores at turbo frequency, (base clock is not very low and there is not a plane fan in front of it), I would be chock. It's just thermal electronics. It's same on a 0 to 5W processors (what I work with), than 100W. the math isn't different.

And off course I would be interested to read deep test on what it has become today, if there is any.

[dengyong]

No, I don't.
I am electronic engineer, and I understand well the path they took to do this.
They increase base clock with temperature/power limits and sensors, maybe internal electronic noise or delays too.

They use an built in microcontroler for doing that, they have implemented some advanced/fancy profiles.
Same principle with throttle function, when temperature/power is too high, they drop frequency under base clock. It was done a lot in Pentium 4. The problem that intel was sued for selling processors that can't keep up with base frequency.
They done different after that with a new way and they called it "Turbo". Base clock is lower, and Turbo clock aren't garanted ...
Today real average frequency now is more and more about dependent on process, and even mobo design.



Yeah mobo design is still important now, less and less off course, but when it was old time of pentium 4, the power stage that drived the hungry beast wasn't as efficient as today. A lot of heat from the power stage for the CPU, warm up air close to the CPU, and reduced the processor cooling performance.

Power stages are much more efficient today. However on a mobo, the CPU socket placement is still important today. The path of the air affect cooling performances, and so the real average clock.


On first generations with turbo it was around 1 to 5 seconds. Now it's maybe better, not because technique is improved, but because
1/ much more cores in the CPUs, so just one core at turbo can last long, full cores on turbo (with a cpu burn tool) it will never last long of course. Yeah they designed the base clock based on full cores without throttle and a small margin. It's obvious that if that could do better with all cores they would have did it on base clock ...
2/ process is a lot improved, helps a lot on thermal. Core are smaller and smaller with a lot of dark silicon. Easier to dissipate more with dark silicon, you just surround the cores with dark silicon.
3/ architecture improvements; with reduced base clock for same performances (better IPC), means less dynamic power consumption means cooler CPU. By More the frequency is far from the invicible wall of 5ghz clock, more it's "easy" to get large turbo clocks.
4/ Improvements in data transmission on the chip. New technologies came since first pentium. serials links with differentials pairs are everywhere now. It's easier to improve clock when the full chip is not going at the same clock, but every block has it's own clock domain. Much easier to synchronize.

Maybe some parts aren't totally exact, it's difficult to have real technical data on this. You can bet it's garded like a graal in companies.

If you find a processor that can keep up over 10 minutes with all cores at turbo frequency, (base clock is not very low and there is not a plane fan in front of it), I would be chock. It's just thermal electronics. It's same on a 0 to 5W processors (what I work with), than 100W. the math isn't different.

And off course I would be interested to read deep test on what it has become today, if there is any.

I have never experienced this problem but I do use high end water cooling on all my builds.

[Sparky]

Well, yeah, the better your cooling the longer/higher the turbo can run.

[AliG]

On first generations with turbo it was around 1 to 5 seconds. Now it's maybe better, not because technique is improved, but because
1/ much more cores in the CPUs, so just one core at turbo can last long, full cores on turbo (with a cpu burn tool) it will never last long of course. Yeah they designed the base clock based on full cores without throttle and a small margin. It's obvious that if that could do better with all cores they would have did it on base clock ...
2/ process is a lot improved, helps a lot on thermal. Core are smaller and smaller with a lot of dark silicon. Easier to dissipate more with dark silicon, you just surround the cores with dark silicon.
3/ architecture improvements; with reduced base clock for same performances (better IPC), means less dynamic power consumption means cooler CPU. By More the frequency is far from the invicible wall of 5ghz clock, more it's "easy" to get large turbo clocks.
4/ Improvements in data transmission on the chip. New technologies came since first pentium. serials links with differentials pairs are everywhere now. It's easier to improve clock when the full chip is not going at the same clock, but every block has it's own clock domain. Much easier to synchronize.

Maybe some parts aren't totally exact, it's difficult to have real technical data on this. You can bet it's garded like a graal in companies.

If you find a processor that can keep up over 10 minutes with all cores at turbo frequency, (base clock is not very low and there is not a plane fan in front of it), I would be chock. It's just thermal electronics. It's same on a 0 to 5W processors (what I work with), than 100W. the math isn't different.

And off course I would be interested to read deep test on what it has become today, if there is any.

I agree with virtually all that you said, but I would like to point out that the newer silicon isn't necessarily always better from a leakage perspective. If I recall correctly, Ivy Bridge had serious issues due to the 22nm process being very leaky at high clocks/temps. FinFET obviously helps a lot, but it's hard to say how turbo speed will be impacted as we go to smaller and smaller processes. I think there's realistically going to be a point in the near future where we need to significantly redo the dielectric materials.

[iddqd]

Isn't necessarily better? That's an interesting way of saying "always worse".

[AliG]

Isn't necessarily better? That's an interesting way of saying "always worse".

It's not always worse. 14nm FinFET process > 22nm process without FinFET. It's not as simple as saying things will always get worse as the gates get smaller. The actual engineering of the gates themselves matters too.

[iddqd]

That's why they had to come up with finFET - with planar transistors, the gate leakage and sub-threshold leakage would be unreasonable. But in general, as you make the channel shorter, it is more vulnerable to the aptly-named short-channel problems. All channels suffer from imperfections around the corners (so, close to each depletion region); but with a long-enough channel it doesn't matter; the strong E-field imposed by the gate will keep things under control in the middle of the gate, and the imperfections around the edges don't really matter. As you shorten the channel, two things occur: (1) the E-field is weaker, as there's less gate now, and (2) the imperfection regions are brought closer together.

As the channel becomes very short, the two imperfection regions start to overlap, which causes disastrous effects; eg: punch-through - at this point, you are leaking 100% of your current from source to drain, and the gate is unable to control the channel at all! The transistor no longer functions as a switch.

FinFET attempts to deal with this problem by building the channel vertically; where it was previously a flat piece of doped silicon resting between two depletion regions, it now sticks out of the substrate vertically, with the gate wrapping around it. This achieves two things: (1) we once again have more gate surface area - which can put up a stronger E-field to better control the current, and (2) we bring as much of the channel as possible away from the troublesome depletion regions, thereby (partially) mitigating the short channel imperfections (while still having a shorter channel; it's just tall now).

But you cannot fight physics forever. Right now, we surround our channel with the gate on 3 sides (the bottom is still just sitting on top of the substrate). And yet, short channel effects are still a big concern, and continue to get worse as we shorten the channel. The next obvious step, indeed already attempted by semiconductor researchers, is to surround the channel on all 4 sides! Thus we have the next thing, the Gate-All-Around transistor. We'll likely need it around 4-3nm, otherwise - again, we'll end up with transistors that cannot be turned off once they're turned on.

[madcho]

That's why they had to come up with finFET - with planar transistors, the gate leakage and sub-threshold leakage would be unreasonable. But in general, as you make the channel shorter, it is more vulnerable to the aptly-named short-channel problems. All channels suffer from imperfections around the corners (so, close to each depletion region); but with a long-enough channel it doesn't matter; the strong E-field imposed by the gate will keep things under control in the middle of the gate, and the imperfections around the edges don't really matter. As you shorten the channel, two things occur: (1) the E-field is weaker, as there's less gate now, and (2) the imperfection regions are brought closer together.

As the channel becomes very short, the two imperfection regions start to overlap, which causes disastrous effects; eg: punch-through - at this point, you are leaking 100% of your current from source to drain, and the gate is unable to control the channel at all! The transistor no longer functions as a switch.

FinFET attempts to deal with this problem by building the channel vertically; where it was previously a flat piece of doped silicon resting between two depletion regions, it now sticks out of the substrate vertically, with the gate wrapping around it. This achieves two things: (1) we once again have more gate surface area - which can put up a stronger E-field to better control the current, and (2) we bring as much of the channel as possible away from the troublesome depletion regions, thereby (partially) mitigating the short channel imperfections (while still having a shorter channel; it's just tall now).

But you cannot fight physics forever. Right now, we surround our channel with the gate on 3 sides (the bottom is still just sitting on top of the substrate). And yet, short channel effects are still a big concern, and continue to get worse as we shorten the channel. The next obvious step, indeed already attempted by semiconductor researchers, is to surround the channel on all 4 sides! Thus we have the next thing, the Gate-All-Around transistor. We'll likely need it around 4-3nm, otherwise - again, we'll end up with transistors that cannot be turned off once they're turned on.

Agree, 3D FET transistors was a large improvement, leakage was heavily reduced with that. The race is like FET current, quadradic improvements for years. After the improvements will be slooooow and slower, and definetly hard. The insulation of SIO2 (for gate) is already around 5 atoms thick for decades already. There is no way to improve this.

Next computers may be DNA based or quantic, ... , not only transistors. And the improvement margins ares less and less on hardware. But large software optimisation is needed. A well optimised code can do a lot of work even on ?Controllers.

4-5nm will be the last commercial node. And Maybe available in 2027-2030. (And few sample or limited quantity before).

[AliG]

That's why they had to come up with finFET - with planar transistors, the gate leakage and sub-threshold leakage would be unreasonable. But in general, as you make the channel shorter, it is more vulnerable to the aptly-named short-channel problems. All channels suffer from imperfections around the corners (so, close to each depletion region); but with a long-enough channel it doesn't matter; the strong E-field imposed by the gate will keep things under control in the middle of the gate, and the imperfections around the edges don't really matter. As you shorten the channel, two things occur: (1) the E-field is weaker, as there's less gate now, and (2) the imperfection regions are brought closer together.

As the channel becomes very short, the two imperfection regions start to overlap, which causes disastrous effects; eg: punch-through - at this point, you are leaking 100% of your current from source to drain, and the gate is unable to control the channel at all! The transistor no longer functions as a switch.

FinFET attempts to deal with this problem by building the channel vertically; where it was previously a flat piece of doped silicon resting between two depletion regions, it now sticks out of the substrate vertically, with the gate wrapping around it. This achieves two things: (1) we once again have more gate surface area - which can put up a stronger E-field to better control the current, and (2) we bring as much of the channel as possible away from the troublesome depletion regions, thereby (partially) mitigating the short channel imperfections (while still having a shorter channel; it's just tall now).

But you cannot fight physics forever. Right now, we surround our channel with the gate on 3 sides (the bottom is still just sitting on top of the substrate). And yet, short channel effects are still a big concern, and continue to get worse as we shorten the channel. The next obvious step, indeed already attempted by semiconductor researchers, is to surround the channel on all 4 sides! Thus we have the next thing, the Gate-All-Around transistor. We'll likely need it around 4-3nm, otherwise - again, we'll end up with transistors that cannot be turned off once they're turned on.

Isn't that verbatim what I said (regarding the gates themselves and dielectrics needing to be redesigned)? There's multiple people with graduate EE degrees here, you don't need to justify a simple mistake by trying show how smart you are

[iddqd]

Well, the point I was trying to make is that leakage always increases as channel length decreases. You can come up with more optimal transistor geometries, but that doesn't really change the fact that it'll still keep getting worse. I'm sorry I decided to back that up with facts.