How well do you think Broadwell-K will overclock?

[cbn]

How well do you think Broadwell-K will overclock?

I'm thinking about 200 Mhz less on air (on average) than a Haswell i5 or i7 K processor.

The main selling point will be the iGPU. This to compensate for a denser process tech that needs to be used with the Broadwell Core in various products (Core M, Broadwell U, Xeon-D, Broadwell E5 Xeon) for various reasons.

 

[cbn]

I think we all realize Intel has incentive to increase perfomance per watt on the big core laptop chips, but Intel's E5 strategy is another factor to consider for the Broadwell core design/targets as well.

On the top Ivy Bridge E5 Xeon 2697 v2, the twelve cores had a base clock of 2.7 Ghz, but they could all turbo to 3.0 Ghz.

On the top Haswell E5 Xeon 2699 v3, the eighteen cores had a base clock of only 2.3 Ghz (and turbo on all cores was removed).

Now if Intel has plans to both increase core count on Broadwell E5 as well as increase base clocks, then I am thinking they would have increased density (more than expected over Haswell) to improve performance at low power consumption (per core). However, this will realistically limit maximum clock.

 

[Techhog]

ignore

 

[crashtech]

EDIT: I can see how my lame attempt at humor could be misunderstood. But this topic has kinda been beaten to death in other threads, we just have to wait for the actual CPUs now, imo.

 

[dragantoe]

It seems like every time I come onto anandtech the quality of reponses get worse and worse. I was hoping to see some speculation and hopefully learn something new about broadwell, but I guess not...

 

[cbn]

One question that was brought up in past was whether or not Intel uses the same process density on 14nm for all Broadwell cores.

I am working from the assumption the density (ie, metal layers) for all 14nm Broadwell cores is the same. Certainly it could be different (IIRC Cortex A9 was offered with two different densities for TSMC 40nm), but I am skeptical till I find confirmation.

 

[cbn]

It seems like every time I come onto anandtech the quality of reponses get worse and worse. I was hoping to see some speculation and hopefully learn something new about broadwell, but I guess not...

The reason I like this topic so much and why it is interesting to me has to do with the importance Intel puts on both mobile and the data center (server).

From my standpoint that is what guides the majority of Intel's decisions for Core design. Then we get what is leftover from that for desktop.

Crossing my fingers one day Intel beats this problem and we see a stronger return of x86 cores for desktop.

 

[DrMrLordX]

I predict 4.2-4.4 ghz peak clockspeed at non-suicidal voltages on air. Anything faster than that will take water or better.

 

[coercitiv]

It seems like every time I come onto anandtech the quality of reponses get worse and worse. I was hoping to see some speculation and hopefully learn something new about broadwell, but I guess not...

It seems that way because you just come and take a quick look, and expect to be compensated with valuable info instantly. It doesn't work that way, simply because this issue - 14nm process @ Intel - has already been discussed in other threads, speculations have already been made.

If you want the TLDR version, people are skeptic about Broadwell clocking equal or higher than Haswell, both due to the delayed launch and the way Broadwell U compares relative to Haswell U. Details bellow.

Back in 2013 Intel demoed Broadwell Y vs Haswell Y and showed a 30% power reduction in Cinebench at normalized performance. The result was impressive, and optimistic forum members even expected this advantage to increase as launch time grew near.

Fastforward in 2014 we saw Broadwell launch with surprising results relative to Haswell: on one hand it clearly offered around 30% power reduction at low frequencies, on the other hand the SKUs approaching 3Ghz seemed to offer lower energy savings, in the realm of less than 10%. Broadwell i5 showed massive improvements in turbo clocks consistency and brought better battery life in selected systems, Broadwell i3 saw a jump in base clocks, but the all the i7 had to show for itself was a few percents of better performance at arguably equal power usage.

Now keep in mind this is just speculation, because even if data was derived from existing products, there's a lot of variation due to different components used (different notebooks, different OEMs etc). The only confirmation to this line of thought is that 3Ghz+ Broadwell U products are consistently clocked a bit lower than their Haswell counterparts.

 

[jji7skyline]

It seems that way because you just come and take a quick look, and expect to be compensated with valuable info instantly. It doesn't work that way, simply because this issue - 14nm process @ Intel - has already been discussed in other threads, speculations have already been made.

If you want the TLDR version, people are skeptic about Broadwell clocking equal or higher than Haswell, both due to the delayed launch and the way Broadwell U compares relative to Haswell U. Details bellow.

Back in 2013 Intel demoed Broadwell Y vs Haswell Y and showed a 30% power reduction in Cinebench at normalized performance. The result was impressive, and optimistic forum members even expected this advantage to increase as launch time grew near.

Fastforward in 2014 we saw Broadwell launch with surprising results relative to Haswell: on one hand it clearly offered around 30% power reduction at low frequencies, on the other hand the SKUs approaching 3Ghz seemed to offer lower energy savings, in the realm of less than 10%. Broadwell i5 showed massive improvements in turbo clocks consistency and brought better battery life in selected systems, Broadwell i3 saw a jump in base clocks, but the all the i7 had to show for itself was a few percents of better performance at arguably equal power usage.

Now keep in mind this is just speculation, because even if data was derived from existing products, there's a lot of variation due to different components used (different notebooks, different OEMs etc). The only confirmation to this line of thought is that 3Ghz+ Broadwell U products are consistently clocked a bit lower than their Haswell counterparts.

Do you think that skipping the 'tick' cycle for the desktop market will become a recurring theme for Intel? It seems like they would save quite a bit of money that way without any perceivable downside, especially if they can release higher clocked versions of their 'tock' CPUs instead like they did with Devil's Canyon.

 

[coercitiv]

Do you think that skipping the 'tick' cycle for the desktop market will become a recurring theme for Intel? It seems like they would save quite a bit of money that way without any perceivable downside, especially if they can release higher clocked versions of their 'tock' CPUs instead like they did with Devil's Canyon.

There are far better trained people on this forum who can give a more informed opinion on this matter, but it seems to me the tick-tock strategy is a very efficient way of minimizing risk while still providing the market with a rather steady flow of improved products.

 

[cbn]

The only confirmation to this line of thought is that 3Ghz+ Broadwell U products are consistently clocked a bit lower than their Haswell counterparts.

Yep, at 15 watts the Broadwell i7s are clocked slightly lower than the 15 watt Haswell i7s:

http://ark.intel.com/products/codename/42174/Haswell?wapkw=haswell#@Mobile

http://ark.intel.com/products/codename/38530/Broadwell#@Mobile

 

[coercitiv]

Yep, at 15 watts the Broadwell i7s are clocked slightly lower than the 15 watt Haswell i7s

It also applies to the 28W models.
http://ark.intel.com/products/84993/Intel-Core-i7-5557U-Processor-4M-Cache-up-to-3_40-GHz
http://ark.intel.com/products/83506/Intel-Core-i7-4578U-Processor-4M-Cache-up-to-3_50-GHz

However, keep in mind this info is no proof by itself, it's just one more thing to keep track of when comparing BW vs HW so far.

 

[Hulk]

It seems like every time I come onto anandtech the quality of reponses get worse and worse. I was hoping to see some speculation and hopefully learn something new about broadwell, but I guess not...

Did you read all of the Broadwell threads? All of the 14nm and other process speculation?

 

[alcoholbob]

So its a power hungry node that uses similar or more power for the same performance than 22nm? What kind of node shrink is that? :O

 

[jpiniero]

So its a power hungry node that uses similar or more power for the same performance than 22nm? What kind of node shrink is that? :O

One that's turned for Core M and Xeon D, not high end desktops? That said, on the U parts the base clock is higher and it can sustain the turbo basically indefinitely.

I'd say 4.1 or 4.2. Don't expect much out of Skylake-K either; The FIVR removal seems to have helped out the 35 W desktop parts and not >4 Ghz like people thought.

 

[crashtech]

So its a power hungry node that uses similar or more power for the same performance than 22nm? What kind of node shrink is that? :O

Intel's first stab at 14nm? It's obviously proving to be quite a challenge.

 

[Techhog]

I predict 4.2-4.4 ghz peak clockspeed at non-suicidal voltages on air. Anything faster than that will take water or better.

Are we going to eventually get the point that getting an i5 to 4GHz will require LN2!?

 

[cbn]

Do you think that skipping the 'tick' cycle for the desktop market will become a recurring theme for Intel? It seems like they would save quite a bit of money that way without any perceivable downside, especially if they can release higher clocked versions of their 'tock' CPUs instead like they did with Devil's Canyon.

I think what will happen next is mainstream will increase to beyond four cores.

If Intel releases a Skylake Xeon-D SoC I'm thinking that could become the next LGA socket (call it Skylake-D) --> http://forums.anandtech.com/showthread.php?t=2429616. Then after that the quad core + iGPU and 8C SoC LGA sockets merge together as one SoC based LGA socket.

 

[cbn]

So for Skylake and Cannolake:

Skylake-S= 4C + iGPU
Skylake-D= 8C
Skylake-E= 12C or greater
Skylake Xeon E5= 12C or greater. 18C or greater, 24C or greater

Cannonlake-S= 4C + iGPU
Cannonlake-D= 8C or 10C
Cannonlake-E= core counts increase over Skylake E
Cannonlake Xeon E5= core counts increase over Skylake Xeon E5

Next stop after Cannolake? MorphCore or continued evolution of Core uarch?

 

[jpiniero]

Cannonlake-S= 4C + iGPU
Cannonlake-D= 8C or 10C
Cannonlake-E= core counts increase over Skylake E
Cannonlake Xeon E5= core counts increase over Skylake Xeon E5

It's very unlikely Cannonlake comes to the desktop. My guess is that it brings back the FIVR thus making the two incompatible socket wise plus they need to eat more 14 nm wafers. It'd be nice if they brought the PCH ondie, but it seems like that won't happen until After Cannonlake. MorphCore would be nice.

 

[NTMBK]

It will be a mediocre overclocker.

If it overclocked better than Haswell, we would see a greater than 4/4.4GHz replacement for the 4790k. We're not, so... yeah.

 

[frozentundra123456]

It will be a mediocre overclocker.

If it overclocked better than Haswell, we would see a greater than 4/4.4GHz replacement for the 4790k. We're not, so... yeah.

I am not expecting much either, but remember the launch clocks of the 4770k were considerably lower than the clocks of the 4790k, although max overclocks did not change much more than a couple hundred mhz. If the leaks are correct, I actually see the stock clocks of the 6700k as a good thing, because they are higher than the stock clocks of the 4770k. So maybe we will see a skylake refresh with something like 4.2/4.5 base turbo.

Now this is for Skylake, Broadwell, who knows, but IMO that is a pretty much lost generation on the desktop anyway. Broadwell k does not really make sense to me. I can see a lower clocked, lower TDP part with iris pro for Brix like devices, maybe SFF lite gaming boxes where you still want intel quality cpu performance and a better igp (what is the market for that???), but why make it a K model.

Edit: I guess what I am saying is that if I wanted good cpu performance I would go with 4790k, hex core intel, or wait for skylake. If I wanted good IGP performance, I would go with Kaveri and save a lot of money. At the price I expect it to go for, I just dont see where BDW k fits in.

 

[DrMrLordX]

Are we going to eventually get the point that getting an i5 to 4GHz will require LN2!?

Probably not. There will be changes in materials before we get to that point. We won't be on silicon forever.

 

[jpiniero]

but why make it a K model.

It's not a K model. it's a C model. As I've mentioned before, my theory is that Intel intended the unlocked models to have Iris Pro exclusively going forward but changed their thinking at some point to do both C and K now.

So maybe we will see a skylake refresh with something like 4.2/4.5 base turbo.

Skylake Refresh will probably just be a platform update. Maybe the non-i7 K will get an extra 100 mhz or something. Don't expect much.