Intel Skylake / Kaby Lake

[PeterScott]

I've seen evidence that the 7740X is binned better, its stock turbo bins are higher than the 7700K.

Intel setting the clocks higher, isn't evidence that it overclocks higher.

 

[crashtech]

Intel setting the clocks higher, isn't evidence that it overclocks higher.

I said it's evidence that it's binned better, which increases the odds of having a good OC sample. It's not possible to tell how many get over 5GHz by perusing published user results, since it's easy to infer that the worse a result is, the less likely it is to be bragged about.

 

[AdamK47]

Memory overclock-ability is higher on X299 as well.

 

[LTC8K6]

Memory overclock-ability is higher on X299 as well.

Yeah, that's another advantage over the 7700K.

 

[dullard]

I haven't really seen evidence that OC better than 7700K, it seems everyone and their dog could get their 7700K to 5GHz. How fast does the average 7740x run?

It is a niche chip that just came out. There isn't a lot of data to go on yet.

53% hit 5.2 GHz in this test:
https://videocardz.com/70338/intel-core-i7-7740x-overclockability

Gigabyte says:
"A good air cooler such as the Enermax ETS-T50 that uses a push-pull fan setup paired with a good CPU may be able to hit 5.3GHz or higher."
http://overclocking.guide/gigabyte-x299-kaby-lake-x-overclocking-guide/

 

[Zucker2k]

Intel setting the clocks higher, isn't evidence that it overclocks higher.

It's been clocked higher on HWBOT too.

 

[IntelUser2000]

The increase from 4.5GHz -> 4.7GHz should give about 5% of that gain. The larger shared L3 cache (12MB vs 8MB) should give the rest.

CFL is going to be awesome

I can't believe its just Turbo and cache.

Turbo frequency comparison:

http://www.cpu-monkey.com/en/compare_cpu-intel_core_i7_8700k-763-vs-intel_core_i7_7700k-664

7700K clocks merely 100MHz lower in Base frequency compared to maximum all-core frequency for 8700K. There has to be perf/clock improvement.

Cinebench scaling not 100%.

The gain never goes below 50%. There's an IPC improvement. The scaling is generally in the 85-90% range.

L3 cache and memory insensitive on Cinebench.

https://linustechtips.com/main/topic/599271-testing-cinebench-r15/

Skylake cache - putting aside HT, does having 8MB of L3 cache on the i7 help over 6MB cache on i5? Nope. Results were practically identical.

http://www.anandtech.com/show/5276/intel-core-i7-3820-review-285-quadcore-sandy-bridge-e/2
https://www.kitguru.net/components/cpu/luke-hill/intel-core-i7-6700k-i5-6600k-skylake-cpu-review/5/

ntel HD Graphics 730? Any information on that yet? I thought that the graphics were to be the same. Their stats are the same on the link except for the GPU generation gets a small bump to 10.5 from 10.

It's not exactly the same. I think if they call it the 730 generation they will call it "UHD 730":

https://forums.anandtech.com/thread...s-out-page-554.2428363/page-563#post-39018700

By the way, Skylake is Gen 9. Kabylake is Gen 9.5 because of the media enhancements. It looks like Coffelake doesn't bring any new features there. That'll be for Cannon and Ice.


How to explain this? Coffelake = Cannonlake CPU core + Gen 9.5 graphics? Maybe they have a surprise in store for us.

 

[Arachnotronic]

I can't believe its just Turbo and cache.

Turbo frequency comparison:

http://www.cpu-monkey.com/en/compare_cpu-intel_core_i7_8700k-763-vs-intel_core_i7_7700k-664

7700K clocks merely 100MHz lower in Base frequency compared to maximum all-core frequency for 8700K. There has to be perf/clock improvement.

Cinebench scaling not 100%.

The gain never goes below 50%. There's an IPC improvement. The scaling is generally in the 85-90% range.

L3 cache and memory insensitive on Cinebench.

https://linustechtips.com/main/topic/599271-testing-cinebench-r15/



http://www.anandtech.com/show/5276/intel-core-i7-3820-review-285-quadcore-sandy-bridge-e/2
https://www.kitguru.net/components/cpu/luke-hill/intel-core-i7-6700k-i5-6600k-skylake-cpu-review/5/



It's not exactly the same. I think if they call it the 730 generation they will call it "UHD 730":

https://forums.anandtech.com/thread...s-out-page-554.2428363/page-563#post-39018700

By the way, Skylake is Gen 9. Kabylake is Gen 9.5 because of the media enhancements. It looks like Coffelake doesn't bring any new features there. That'll be for Cannon and Ice.


How to explain this? Coffelake = Cannonlake CPU core + Gen 9.5 graphics? Maybe they have a surprise in store for us.

IPC improvement comes from L3$ size growth and memory clock improvement. The core is Skylake.

 

[IntelUser2000]

IPC improvement comes from L3$ size growth and memory clock improvement. The core is Skylake.

Cinebench is not sensitive to L3 cache and memory clock.

20% higher revenue in a shrinking market is not happening and would turn it in to a growing market all by itself, not even Intel could do it,

Intel, couldn't do it. Nvidia did. It's up to a business to figure how to to grow your market. It's up to them to grow outside the market so they can grow total revenues. I agree, PC market growth of that degree may be very difficulty but as a company?

 

[Java Sunday]

IPC improvement comes from L3$ size growth and memory clock improvement. The core is Skylake.

Is it though? This is an improved process. I could see an ipc bump from process/design improvements.

 

[Arachnotronic]

Is it though? This is an improved process. I could see an ipc bump from process/design improvements.

IPC and process/physical implementation are orthogonal.

Better transistors mean either better frequencies at a given power consumption or better power consumption at a given frequency, all else equal.

So process improvements should ultimately translate into higher clocks, or the ability to stuff more cores/cache into a given area while keeping power in check. It will not affect the underlying performance-per-clock of the CPU core itself.

But that's OK. Nobody is buying IPC, frequency, or core count in a vacuum. They are buying delivered performance, and CFL-S seems on pace to deliver it in spades.

 

[Dayman1225]

https://twitter.com/intel/status/898254389948764160

"Are you ready for what's next? Watch LIVE on Twitter, 8/21 @ 8am PT, as we launch the #8thGen #Intel Core Processor."

 

[Ajay]

Well, that's only one test of an 8700K sample. That said, if those results bear out, I think Intel took a bit of time to tweak CFL's circuit design, at least while doing the hexacore layout, if not in general. Looks more like a tock than a tick due to process improvement alone.

 

[Arachnotronic]

Well, that's only one test of an 8700K sample. That said, if those results bear out, I think Intel took a bit of time to tweak CFL's circuit design, at least while doing the hexacore layout, if not in general. Looks more like a tock than a tick due to process improvement alone.

A lot of work did go into CFL, it wasn't a lazy effort, that's for sure.

 

[Zucker2k]

Only question left now is the thermals.

 

[raghu78]

IPC improvement comes from L3$ size growth and memory clock improvement. The core is Skylake.

We will know actual IPC improvement when reviewers test with locked clocks and same memory speeds. In that case perf improvement due to L3 cache size increase will be the only factor. I doubt Intel improved L1 and L2 cache latency on Coffeelake.

 

[Arachnotronic]

We will know actual IPC improvement when reviewers test with locked clocks and same memory speeds. In that case perf improvement due to L3 cache size increase will be the only factor. I doubt Intel improved L1 and L2 cache latency on Coffeelake.

The memory speed increase should also contribute to "IPC." KBL is officially rated for DDR4-2400; CFL is rated for DDR4-2666.

Given how bandwidth hungry the SKL core has proven to be (since Intel nerfed the L2 cache from 8-way to 4-way, and kept it at a small 256KB in the client core), the bigger L3 cache and faster official DDR4 memory will help.

In practice, since the KBL mem controller is so good, this is going to be less of a benefit for enthusiasts who wouldn't run a 7700K @ DDR4-2400 anyway.

 

[IntelUser2000]

We will know actual IPC improvement when reviewers test with locked clocks and same memory speeds. In that case perf improvement due to L3 cache size increase will be the only factor. I doubt Intel improved L1 and L2 cache latency on Coffeelake.

They won't be able to change the floorplan regarding the cores(which include L1 and L2 caches), except few tweaks they couldn't make it on the original silicon due to time restraints. That means few non-optimal functioning circuits that are re-enabled or deliberate empty space left for subsequent cores.

The memory speed increase should also contribute to "IPC." KBL is officially rated for DDR4-2400; CFL is rated for DDR4-2666.

Not on Cinebench. They show gains for both versions.

 

[Ajay]

We will know actual IPC improvement when reviewers test with locked clocks and same memory speeds. In that case perf improvement due to L3 cache size increase will be the only factor. I doubt Intel improved L1 and L2 cache latency on Coffeelake.

Apparently you missed @IntelUser2000 posts above. There are lots of things that can change vis-a-vis CPU cores besides caches. Anyway, Intel didn't didn't have time to tweak KBL, but likely had time to update the CFL core - or, the data from cpu-monkey is bogus (but when added to the slides from China - probably right). Intel likely did allot of work on Cannon Lake, and very well may have found a way to fit some of those changes into CFL.

 

[StinkyPinky]

Can't wait for the 8700K. Expecting big things from this CPU.

One thing I never get about modern cpus and marketing though. If the 8700k really does turbo 6 cores to 4.3, why not advertise it as a 4.3 cpu?

 

[dullard]

Can't wait for the 8700K. Expecting big things from this CPU.

One thing I never get about modern cpus and marketing though. If the 8700k really does turbo 6 cores to 4.3, why not advertise it as a 4.3 cpu?

Because base speed DOES matter.

Consider a chip in its most typical environment: an OEM box with poor cooling, slow fans or not enough fans, with a thick layer of dust collected in the machine, inside a terrible PC desk cabinet (with few openings to let the heat out, if any). All that and it must still run in that 50°C India example. This is when CPUs run at base all the time and rarely at turbo.

Just because enthusiasts and reviewers with overblown cooling systems run at turbo more often than not, does not mean that the bulk of the CPUs are lucky enough to be in that position.

That and CPUs are listed both with their base and turbo speeds, so your whole premise is off. https://ark.intel.com/products/97129/

 

[Ajay]

Can't wait for the 8700K. Expecting big things from this CPU.

One thing I never get about modern cpus and marketing though. If the 8700k really does turbo 6 cores to 4.3, why not advertise it as a 4.3 cpu?

Because it's probably dependent on thermals. 3.7 GHz (or whatever) is guaranteed - turbo clocks are not. What if you live in AZ and don't have air conditioning

 

[IntelUser2000]

One thing I never get about modern cpus and marketing though. If the 8700k really does turbo 6 cores to 4.3, why not advertise it as a 4.3 cpu?

I think it has to do with lawyers.

Base clock speed is basically a worst case scenario. It could be dropping off Turbo clocks when iGPU is utilzed fully. It could be dropped when AVX2 is used. It could be dropped due to a very intensive program or just a bad thermal solution.

In order to understand you have to understand how Turbo 2.0 works in the first place. Turbo 1.0 was basically clocking higher that depended on the amount of cores active. Now, that's not as flexible and in most cases the gains can't be as great as it can be.

Enter Turbo 2.0. Heatsinks take time to heat up, while power consumption rises instantly. The delta between the two allows the CPU to clock higher. This is most apparent in mobile, where chips can't handle all core max frequency for long and most fall off to a point that's few hundred MHz above Base. Of course in desktops that's not much of a problem since it uses oversized HSF.

Intel likely did allot of work on Cannon Lake, and very well may have found a way to fit some of those changes into CFL.

I'm not saying that its certain that Coffelake has backported Cannonlake core, but the gains seem too good for 50% increase in cores(and programs that don't scale linearly) and 4.4% increase in max single core frequency.

It was rumored at some point some of the 14nm-based cores have backported Cannonlake cores. I believe in the saying if you see a smoke on the smokestack, there's something burning. Not all rumors are true, but its possible to discern with reasonably high accuracy which are true and which are not. Willow Trail does not exist as a product today, doesn't mean the slides with Willow Trail is a rumor, it just never became a product.

The "job" we are responsible as forum goers is figuring out the reasons for why its such a case. Sometimes the decisions made by these companies do not make sense.

 

[jpiniero]

Turbo 2.0 can run OVER the TDP limit for a brief period if designed to do so.

Coffee Lake does not appear to have the two main new features of Cannonlake (AVX-512 albeit only the 2x256-bit units and Gen10 graphics) so it having any changes from Cannonlake doesn't seem very likely.

My guess is that it's a canned benchmark where cache really helps.

 

[IntelUser2000]

Turbo 2.0 can run OVER the TDP limit for a brief period if designed to do so.

Key point. Brief period. Long term clocks and TDP has to be at or below the assigned level.

Throw away the concept of looking at desktops. It's all about mobile if you want to understand. TDP exactly equals power use in a system where Turbo 2.0 is in use.