Could Skylake present us a regression in single-threaded performance?

[meloz]

This forum has been slow -not much new CPU news- so let's speculate to pass our time.

I have been thinking about Intel's desktop CPU strategy for 2015. They will be selling Broadwell-K (old architecture, shrunk Haswell) and Skylake (their new baby) at the same time.

Now, we know our typical non-K Skylake SKU will be locked and not overclocakable, but even so a new Intel architecture usually gives a 5-10% performance increase in single-threaded scenarios. If Skylake does indeed give us such an increase, would that not obviate Broadwell-K's performance advantage even when overclocked? Why would anyone buy Broadwell-K when they can potentially get nearly the same single-threaded performance at much better performance/watt with Skylake?

Could it be, that in their pursuit of ever higher performance/watt Intel have decided to accept a regression in single-threaded performance, and designed Skylake as a 'weaker' core compared to Broadwell? Could Skylake actually be slower in single-threaded benchmarks than Broadwell (K or non-K) at similar clock speeds?

We know how Intel are becoming more mobile focused and want to push their core into as many gadgets as possible. It would also explain why they see a need to sell Broadwell-K in parallel to Skylake: Skylake should satisfy OEM and mainstream users, but for users who do not want a regression Broadwell-K should suffice?

What are your expectations about single-threaded performance from Skylake? I am not so much concerned about the overall performance (should be good with four or more cores) or the iGPU. I am not concerned about performance/watt either, which should also be better than Broadwell.

I am mostly concerned about single-threaded performance, which has started to stagnate these past few years, and now I worry it could even regress. :|

 

[Yuriman]

That hadn't occurred to me.

I'm not nearly as informed as some members on here, but my impression was that the simultaneous sales are a result of non-overlapping markets and delays in developing the process.

 

[NTMBK]

This forumindustry has been slow

FTFY. CPU progress has been slooooow lately.

 

[Deders]

It will depend on how much they can to clock it up to on release. Mobile parts will be slower because it's designed power saving, but if the transistors will take it then power isn't so much of an issue in desktops.

 

[inf64]

It will depend on how much they can to clock it up to on release. Mobile parts will be slower because it's designed power saving, but if the transistors will take it then power isn't so much of an issue in desktops.

I doubt there will be any regression in ST performance with Skylake, quite the contrary. If will likely outperform similarly clocked Broadwell by 5-10%.

 

[witeken]

I am mostly concerned about single-threaded performance, which has started to stagnate these past few years, and now I worry it could even regress. :|

Everything points toward a major upgrade. Intel really can't stagnate if it wants to keep its market share and customers (Apple), a regression would be ever worse.

So of course not.

 

[meloz]

Everything points toward a major upgrade. Intel really can't stagnate if it wants to keep its market share and customers (Apple), a regression would be ever worse.

Would most OEM customers care when performance/watt and iGPU are actually much better than Broadwell?

I doubt there will be any regression in ST performance with Skylake, quite the contrary. If will likely outperform similarly clocked Broadwell by 5-10%.

Here's wishing you are right. Skylake would then represent a good upgrade from my present Sandy Bridge PC.

 

[witeken]

Would most OEM customers care when performance/watt and iGPU are actually much better than Broadwell?

Ignoring that customers are pretty ignorant of technology, yes.

 

[Nothingness]

FTFY. CPU progress has been slooooow lately.

FTFY: x86 CPU progress has been slow lately

FWIW I can't believe Intel would ship a CPU with less single thread performance.

 

[beginner99]

Just look at devils canyon. It's a lot higher clocked than stock cpus. If the same applies to broadwell-k it will still be faster than a lower clocked skylake. I mean this applies even now. It makes little sense to upgrade from a SB that achieves 5 Ghz. If you ignore new instructions that SB is probably still faster than stock haswell which is 2!! generations newer.

Besides that broadwell-k will ship with beefier iGPU and more importantly the eDRAM/l4 cache that might offer some benefits in certain situations.

 

[witeken]

FTFY: x86 CPU progress has been slow lately

If by 'progress' you mean IPC improvement.

 

[CHADBOGA]

I have no doubt that Skylake will be an improvement over Broadwell in IPC, I'm more concerned at what clockrates it will be capable of running at.

 

[Nothingness]

If by 'progress' you mean IPC improvement.

Yes

But as I have already told, I'm amazed by the fact Intel engineers can still improve IPC of the Core architecture, given how great it already is.

 

[witeken]

Yes

But as I have already told, I'm amazed by the fact Intel engineers can still improve IPC of the Core architecture, given how great it already is.

Apple seems to have been able to improve IPC by 20% or so with Cyclone 2. So given that it was already on par with Haswell, Intel can't stay behind.

 

[crashtech]

Crystal ball says exactly 5% improvement, making everyone in the poll wrong.

 

[witeken]

Crystal ball says exactly 5% improvement, making everyone in the poll wrong.

5% would be as much as Broadwell, which is a Tick.

 

[sm625]

Well they added two execution ports with haswell. It is safe to assume they will do something similar with skylake. Perhaps more, if they dont dump a whole bunch more transistors into AVX. And I hope they dont, because they are dumping an awful lot of transistors into AVX, and its still really not being used. I assume intel realizes this too, and so they are hopefully giving that a rest for now and focusing more on general computing performance. If they do that then we could see huge jumps in things like javascript, which will makes skylake scream through benchmarks. Another thing they could easily do, with so many execution ports, is add another thread to hyperthreading.

 

[BigDaveX]

I have no doubt that Skylake will be an improvement over Broadwell in IPC, I'm more concerned at what clockrates it will be capable of running at.

Early rumours aren't very encouraging - the engineering samples in the wild right now supposedly take 95W to achieve a 2.3GHz stock speed (2.9GHz turbo). I can only assume that either Intel is packing one mother of an IGP in Skylake, or they figured AMD are so uncompetitive that they can dust off their Tejas samples and pass them off as new chips.

 

[witeken]

Early rumours aren't very encouraging - the engineering samples in the wild right now supposedly take 95W to achieve a 2.3GHz clock speed (2.9GHz) turbo. I can only assume that either Intel is packing one mother of an IGP in Skylake, or they figured AMD are so uncompetitive that they can dust off their Tejas samples and pass them off as new chips.

Yes, it are engineering samples. Here's something you can do: search for the clock speeds of engineering samples of Haswell/Ivy Bridge/Sandy Bridge.

 

[AtenRa]

If by 'progress' you mean IPC improvement.

If by IPC you mean legacy

 

[ShintaiDK]

Apple seems to have been able to improve IPC by 20% or so with Cyclone 2. So given that it was already on par with Haswell, Intel can't stay behind.

Cyclone is not on pair with Haswell. Not even close.

 

[ShintaiDK]

If by IPC you mean legacy

And what would the alternative be?

 

[witeken]

Cyclone is not on pair with Haswell. Not even close.

Maybe.

Interesting quote, BTW (from IDC).

 

[Lepton87]

If Skylake does indeed give us such an increase, would that not obviate Broadwell-K's performance advantage even when overclocked

For that to be true skylake would have to provide a higher improvement in IPC than we are used to seeing. It's been about 7-10% since SB and I think it's not going to change. 4790K is clocked way closer to its limits than any Intel CPU after P3 1.13GHz which was clocked beyond the silicon limit So, if we get that and Broadwell K is only going to overclock about 10% and Skylake is going to be clocked as close to silicon limit as K version of HW than they very well may be similar in performance.

 

[mikk]

Early rumours aren't very encouraging - the engineering samples in the wild right now supposedly take 95W to achieve a 2.3GHz stock speed (2.9GHz turbo). I can only assume that either Intel is packing one mother of an IGP in Skylake, or they figured AMD are so uncompetitive that they can dust off their Tejas samples and pass them off as new chips.

Nonsense. For a first ES this is a standard procedure. First Ivy Bridge ES in the wild clocked at 1.8 or 2.0 Ghz. How much these CPU consumes you can't know unless you did a power consumption test (would be meaningless nevertheless), once again 95W ist just a standard TDP rating.