Intel Skylake complete lineup - Coming in August - Pictures of CPU

[ShintaiDK]

Wow amd is that far behind huh? top end 5.0ghz barely even with i5 3.5ghz skylake.

Considering its multithreaded only makes it worse.

Even if we give AMD the bone and say 6600K runs 3.9Ghz Turbo all the time and the 9570 only runs 4.7Ghz.

9590 is 2.6% faster. But does so using twice the cores and 20.5% clockspeed advantage(most likely more).

In other words, the IPC difference between Skylake and the FX series is now 135% or more.

 

[dangerman1337]

Considering its multithreaded only makes it worse.

Even if we give AMD the bone and say 6600K runs 3.9Ghz Turbo all the time and the 9570 only runs 4.7Ghz.

9590 is 2.6% faster. But does so using twice the cores and 20.5% clockspeed advantage(most likely more).

In other words, the IPC difference between Skylake and the FX series is now 135% or more.

So basically if AMD want to catch up ideally they need a 40% extra IPC improvement. I really hope Zen is something very special. Assuming a 4 or slightly under years gap is it possible for AMD to do something like that? Unless Jim Keller pulls a rabbit out of a hat I doubt they are going to do that with a faux 14nm process.

At this point it seems Bulldozer was a waste; it was built on ideas that should have been held back until 2016/2017 (it seems DX12 would heavily benefit Bulldozer style Uarch).

 

[ShintaiDK]

So basically if AMD want to catch up ideally they need a 40% extra IPC improvement. I really hope Zen is something very special. Assuming a 4 or slightly under years gap is it possible for AMD to do something like that? Unless Jim Keller pulls a rabbit out of a hat I doubt they are going to do that with a faux 14nm process.

At this point it seems Bulldozer was a waste; it was built on ideas that should have been held back until 2016/2017 (it seems DX12 would heavily benefit Bulldozer style Uarch).

40%? They would have to stay at double the cores.

1-4 thread performance is simply impossible for AMD to reach against Intel. A quadcore Zen would have to be 2.35x (or more) faster per core than Pilediver to reach Intel.

 

[SPBHM]

I think the main competition for these products is not AMD, but the older stuff people own from Intel like Haswell and Ivy Bridge...

 

[ShintaiDK]

I think the main competition for these products is not AMD, but the older stuff people own from Intel like Haswell and Ivy Bridge...

Yep. Its been so for a lot of years. Lynnfield/Bloomfield and forward.

 

[Space69]

In other words, the IPC difference between Skylake and the FX series is now 135% or more.

Not quite. Remember that Cinebench is fp heavy.

 

[Chicken76]

Do we know if the dual-core SKUs (i3, Pentium, Celeron) are coming this year or not?

 

[Fjodor2001]

No, the performance gain seem to be around 20%.
If you look at ShintaiDK`s result, its 15% higher than the Haswell.

Then you are comparing with two CPU generations behind Skylake (i.e. skipping Broadwell).

In that case I think it's more fair to talk about performance increase per year. We don't know for sure when desktop Skylake will be released, but likely 2015Q3. Since Haswell was released in June 2013, it'll be a bit over two years between those CPUs.

So the yearly performance increase is still a measly ~6% or so...

 

[Arachnotronic]

Then you are comparing with two CPU generations behind Skylake (i.e. skipping Broadwell).

In that case I think it's more fair to talk about performance increase per year. We don't know for sure when desktop Skylake will be released, but likely 2015Q3. Since Haswell was released in June 2013, it'll be a bit over two years between those CPUs.

So the yearly performance increase is still a measly ~6% or so...

So don't upgrade every year?

 

[ShintaiDK]

Not quite. Remember that Cinebench is fp heavy.

Its not going AMD any better in INT loads. (32 BD cores vs 16 SB cores)

Looking up single thread Spec numbers doesnt help either. Note there are very limited submissions so not entirely acurate.

7850K gets around ~31.5. Assume at 4Ghz.
4690K gets around ~60.2. Assume 3.9Ghz.

A 2500K scores between 43 and 50. Depending on OS, compilers and so on.

 

[Cloudfire777]

Anyone who knows if DDR4 finally will go down in price when Skylake hits the market or how long it will take? I`d like DDR4 on my upcoming build but they are bloody expensive right now (not talking about generic DDR4)

 

[scannall]

Anyone who knows if DDR4 finally will go down in price when Skylake hits the market or how long it will take? I`d like DDR4 on my upcoming build but they are bloody expensive right now (not talking about generic DDR4)

I would imagine that it will a few months after Skylake is on the market.

 

[Fjodor2001]

So don't upgrade every year?

Have I said that I would, or even intended to do that?

Does it even matter what my specific upgrade plans are, when we're discussing the Skylake lineup in general?

 

[Cloudfire777]

I would imagine that it will a few months after Skylake is on the market.

Ugh, wait even longer? Don`t know if Im capable of that.
Intel should work with Corsair or any other DDR4 manufacture and make a bundle pack with Skylake processors. Maybe even have a Intel 750 there as well.

Intel Skylake + Intel PCIe 750 400GB SSD + 8/16GB DDR4 for a great price
That would be awesome

 

[xthetenth]

So basically if AMD want to catch up ideally they need a 40% extra IPC improvement. I really hope Zen is something very special. Assuming a 4 or slightly under years gap is it possible for AMD to do something like that? Unless Jim Keller pulls a rabbit out of a hat I doubt they are going to do that with a faux 14nm process.

At this point it seems Bulldozer was a waste; it was built on ideas that should have been held back until 2016/2017 (it seems DX12 would heavily benefit Bulldozer style Uarch).

If it gets close to that target, it would probably benefit AMD greatly because competing on price or core count is a lot more viable without a giant IPC gap. I think it would potentially be possible given historical precedent to get that much advantage over a bad architecture, but I also think that we're deeper into small increases requiring significant work so I'd be pleasantly surprised if it were that big. I'm hoping for competitive, not flawless.

 

[Cloudfire777]

Then you are comparing with two CPU generations behind Skylake (i.e. skipping Broadwell).

In that case I think it's more fair to talk about performance increase per year. We don't know for sure when desktop Skylake will be released, but likely 2015Q3. Since Haswell was released in June 2013, it'll be a bit over two years between those CPUs.

So the yearly performance increase is still a measly ~6% or so...

Broadwell is Haswell, its the same architecture. They got a minor performance gain mostly because they jumped to 14nm.

Even if you take Broadwell in to consideration, its still only around +5% over Haswell, which means Skylake is +15% over Broadwell if we look at the recent benchies. Although I still like to see more evidence from other sites

We know when Skylake launch. Its in the OP. Did you read it? The processors will be launched in month 8, aka August.

 

[frozentundra123456]

Then you are comparing with two CPU generations behind Skylake (i.e. skipping Broadwell).

In that case I think it's more fair to talk about performance increase per year. We don't know for sure when desktop Skylake will be released, but likely 2015Q3. Since Haswell was released in June 2013, it'll be a bit over two years between those CPUs.

So the yearly performance increase is still a measly ~6% or so...

The 15% improvement was vs Devils Canyon, which brought a 13% improvement vs the original Haswell. So in the two year time period, the increase is nearly 30%.

BTW, even if you accept your 15% figure, how does 15/2 suddenly become 6?

 

[Fjodor2001]

Broadwell is Haswell, its the same architecture. They got a minor performance gain mostly because they jumped to 14nm.

Similar, but not the same. Tweaks have been made to the uArch, just like was done on Ivy Bridge and other node shrink CPU generations. Add to that the effects of the node shrink itself, often brining higher CPU frequency.

Previously the Tocks used to bring much higher performance improvements than Ticks. But the last few Intel CPU generations the Tock vs Tick difference has not been that great.

 

[Fjodor2001]

BTW, even if you accept your 15% figure, how does 15/2 suddenly become 6?

It's not 15/2. Its 1.15^1/2 = 1.07 => 7%. And that is if it's exactly two years between the CPU releases. We know it'll be a bit longer than that, so it'll be closer to 6% per year.

 

[Fjodor2001]

Just wondering, would the performance be better at the top end if Intel focused more on designing high performance process tech instead of prioritizing mobile?

Or do they already have one lower power & low performance process tech, and a separate high power & high performance process tech, both on 14 nm? I know e.g. TSMC has separate process tech for different types of chips, and Intel has had it on previous process tech nodes too IIRC.

 

[IntelUser2000]

Just wondering, would the performance be better at the top end if Intel focused more on designing high performance process tech instead of prioritizing mobile?

Can't say that's going to be a win at all.

For the past 2 decades basically we got increases from increasing TDP every year. Unless we do that we won't get that so its not realistic.

Plus process increases are being less and less which means whatever extra gains they want is directly TDP. Traditionally, doubling core power I think resulted in square root of that, so about 40%.

 

[Fjodor2001]

Can't say that's going to be a win at all.

For the past 2 decades basically we got increases from increasing TDP every year. Unless we do that we won't get that so its not realistic.

Plus process increases are being less and less which means whatever extra gains they want is directly TDP. Traditionally, doubling core power I think resulted in square root of that, so about 40%.

Yeah, but top end Skylake will be 95 W TDP. That is compared to 84 W Haswell, and 77 W Ivy Bridge. But performance has not increased that much.

 

[IntelUser2000]

Yeah, but top end Skylake will be 95 W TDP. That is compared to 84 W Haswell, and 77 W Ivy Bridge. But performance has not increased that much.

The performance gains are same from 975EE to 2600K. But back then, they did not release any Westmere parts. So the gains are greater than it looks. This time, Intel released 4790K. TDP-wise, it might be because of process. We are seeing IMO a time when gains are going to be really hard to come by. Back in the Pentium 4 days, a simple shrink brought immediate 20% gain, and 60% over the lifespan.

They could have just released 4790K as an incremental refresh at 3.6-4.1GHz(100MHz bump), then we would have seen BIG gains just like back then.

Lower end and lower power we'll see much bigger gains. I am expecting 20-30% in the 65W range and 30-40% for 35W. It should be really good for Y and M parts coming later this year, probably even for quads(but that's 2016 story).

 

[Fjodor2001]

The performance gains are same from 975EE to 2600K. But back then, they did not release any Westmere parts. So the gains are greater than it looks. This time, Intel released 4790K. TDP-wise, it might be because of process. We are seeing IMO a time when gains are going to be really hard to come by. Back in the Pentium 4 days, a simple shrink brought immediate 20% gain, and 60% over the lifespan.

They could have just released 4790K as an incremental refresh at 3.6-4.1GHz(100MHz bump), then we would have seen BIG gains just like back then.

Lower end and lower power we'll see much bigger gains. I am expecting 20-30% in the 65W range and 30-40% for 35W. It should be really good for Y and M parts coming later this year, probably even for quads(but that's 2016 story).

Yes, it looks like that. Top end is not seeing much gains anymore.

But you're not thinking that is because mobile is prioritized when designing the process tech for new nodes?

Instead it's maybe that they just design process tech for new nodes, and it just happens that when shrinking process tech nodes nowadays it provides more benefits in the low power & low performance range than the high power & high performance range?

 

[IntelUser2000]

But you're not thinking that is because mobile is prioritized when designing the process tech for new nodes?

Maybe, maybe not. But considering how high the TDPs are now, expecting that we'll go back to "high performance focus" is unrealistic. Also, the curves always favored lower clocked devices, meaning they'll always have better perf/watt. That means higher and higher means less and less gains. Same with GPUs. As soon as it became entirely bound by TDP, the gains started flatlining.

That's probably also why it looks like ARM is catching up. They are at the very low end of performance, so ramping up is easy, while Intel always plays to "eke out" every last bit. Now "easy" is relative, but its "almost impossible" versus "pretty hard".