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

[ShintaiDK]

So, now Desktop CPUs are going Tock to Tock, and Tick-Tock is now dead for Desktop CPUs?

You cant be surprised that server, mobile and tablet/smartphone is more important than desktop in terms of node access. Desktop is slowly dying after all.

 

[ShintaiDK]

Yeah, that does seem to be the 'new normal'. It makes sense with tiny IPC gains for ticks and increased difficulty in ramping new nodes.

It really doesnt have anything to do with node ramping. Its just the priority. And desktop was the loser of the 4 segments and came last. Its more important to make 14nm atoms than it is to make 14nm desktop chips.

Remember this? That slide sealed the fate of desktops nodewise in terms of "early" access to the limited capacity.

 

[ShintaiDK]

DDR3L is the most widely used already.

 

[NTMBK]

Remember this? That slide sealed the fate of desktops nodewise in terms of "early" access to the limited capacity.

Yeah that went well for them, I remember all of those 14nm Atoms in 2014 When is first device due out? Is it the Surface 3, in May 2015?

 

[Cloudfire777]

That does make sense but it seems like the Broadwell upgrade for those with the Z97 chipset is such a small number when it comes to the numbers Intel regularly does.

Hey, I'm not complaining, more options is great. Performance-wise we'll just have to wait and see if these new parts hit the price/performance curve for our individual needs and wants.

Been the norm with Intel after Sandy Bridge (2500K etc). Typical +5% clock for clock gains. Absolutely horrible. Almost sick to think those with 2500K havent missed out on much. And it launched in January 2011 haha.
Makes room for a very positive experience this time if Skylake is +20% though

 

[ShintaiDK]

Yeah that went well for them, I remember all of those 14nm Atoms in 2014 When is first device due out? Is it the Surface 3, in May 2015?

14nm Atoms started shipping in january.

http://www.anandtech.com/show/8831/intel-shipping-14nm-cherry-trail

 

[cbn]

You cant be surprised that server, mobile and tablet/smartphone is more important than desktop in terms of node access. Desktop is slowly dying after all.

Looks like Xeon LGA 2011-3 server will be the last to adopt 14nm node this round.

 

[ShintaiDK]

Looks like Xeon LGA 2011-3 server will be the last to adopt 14nm node this round.

Thats only due to validation time. And its quite normal.

 

[.vodka]

Been the norm with Intel after Sandy Bridge (2500K etc). Typical +5% clock for clock gains. Absolutely horrible. Almost sick to think those with 2500K havent missed out on much. And it launched in January 2011 haha.
Makes room for a very positive experience this time if Skylake is +20% though

On top of that, almost every 2500k/2600k hits about 4.5GHz on a reasonable voltage (1.3v), from then it's about how lucky you got on the silicon lottery. At those speeds it's up there with stock haswell, 4 years and counting. Amazing... I'd say Sandy deserves entering the hall of fame along with the Q6600.

Let's suppose the following:

Sandy -> Ivy (5%)
Ivy -> Haswell (15%)
Haswell -> Skylake (15%, including BW's 5.5%)

That makes Sandy -> Skylake at least a 38.8% jump per clock. In terms of relative clockspeed, 4.5GHz Sandy*0.612 = 2.75 GHz on our hypotetical Skylake. If then we consider the power savings... At least for me that's enough to consider upgrading my 2500k/P67 rig.

It remains to be seen how it overclocks... It should be better than Haswell in that regard.

 

[Deders]

Any hints of NVMe for Skylake?

 

[cbn]

Thats only due to validation time. And its quite normal.

On LGA 1366 validation time difference between Xeons and the consumer Bloomfield was only one quarter.

Examples:

i7-920 launched Q4 2008 --> http://ark.intel.com/products/37147/Intel-Core-i7-920-Processor-8M-Cache-2_66-GHz-4_80-GTs-Intel-QPI

Xeon X5550 launched Q1 2009--> http://ark.intel.com/products/37106/Intel-Xeon-Processor-X5550-8M-Cache-2_66-GHz-6_40-GTs-Intel-QPI

 

[ShintaiDK]

On LGA 1366 validation time difference between Xeons and the consumer Bloomfield was only one quarter.

Examples:

i7-920 launched Q4 2008 --> http://ark.intel.com/products/37147/Intel-Core-i7-920-Processor-8M-Cache-2_66-GHz-4_80-GTs-Intel-QPI

Xeon X5550 launched Q1 2009--> http://ark.intel.com/products/37106/Intel-Xeon-Processor-X5550-8M-Cache-2_66-GHz-6_40-GTs-Intel-QPI

i7 920 comes from server too.

Now before you mention something about Lynnfield. The LGA1366 was rushed so to say due to the dire need for IMC and QPI. Something you would never do in a controlled market.

 

[cbn]

i7 920 comes from server too.

Now before you mention something about Lynnfield. The LGA1366 was rushed so to say due to the dire need for IMC and QPI. Something you would never do in a controlled market.

Lynnfield also had a larger die size than Bloomfield.

P.S. Some of this delay on the current E5 Xeons might be due to validation, but I am thinking most of the delay we see these days has to due with the very large die size (top die variant on the E5-2600 v3 series is over 650mm2). I have to wonder if Intel decides to accelerate the launch of the future E5 Xeon by launching the smaller die versions first and they follow with the larger die versions later? (ie, launch the E5 stack in a staggered fashion rather than simultaneously?)

 

[Hulk]

On top of that, almost every 2500k/2600k hits about 4.5GHz on a reasonable voltage (1.3v), from then it's about how lucky you got on the silicon lottery. At those speeds it's up there with stock haswell, 4 years and counting. Amazing... I'd say Sandy deserves entering the hall of fame along with the Q6600.

Let's suppose the following:

Sandy -> Ivy (5%)
Ivy -> Haswell (15%)
Haswell -> Skylake (15%, including BW's 5.5%)

That makes Sandy -> Skylake at least a 38.8% jump per clock. In terms of relative clockspeed, 4.5GHz Sandy*0.612 = 2.75 GHz on our hypotetical Skylake. If then we consider the power savings... At least for me that's enough to consider upgrading my 2500k/P67 rig.

It remains to be seen how it overclocks... It should be better than Haswell in that regard.

I have clock normalized Anand's benchmark scores from his Haswell review and it shows that Sandy Bridge to Haswell is 18% IPC improvement. If Haswell to Skylake could replicate that 18% that would mean a significant 36% from Sandy Bridge to Skylake.

If Haswell to Skylake is 20% then I might get on board. For me personally I'll be looking at two things. First, video encoding. Not only is it something I wait on but it is indicative of Sony Vegas Pro performance overall. Second, and I know this is going to sound crazy, but more compute with the iGPU would be helpful for preview performance in Sony Vegas Pro. Yes I know I could buy a lower end discrete GPU for much more compute. I did that and could never get the system as stable as with the Intel iGPU. Perhaps I just had a bad experience but I don't want to go down that road again when the Haswell GPU is nearly powerful enough for my needs. A little more and I'm there.

 

[Ajay]

It really doesnt have anything to do with node ramping. Its just the priority. And desktop was the loser of the 4 segments and came last. Its more important to make 14nm atoms than it is to make 14nm desktop chips.

Remember this? That slide sealed the fate of desktops nodewise in terms of "early" access to the limited capacity.

Yes, but I believe that the 'limited' initial quantities below 32 nm also gave pause to Intel's ambitions. So I'd say it's priority and a dose manufacturing reality.

 

[Ajay]

i7 920 comes from server too.

Now before you mention something about Lynnfield. The LGA1366 was rushed so to say due to the dire need for IMC and QPI. Something you would never do in a controlled market.

Yes, and HT and higher MT performance than Core2 Quads.

 

[CHADBOGA]

DDR3L is the most widely used already.

Is DDR3L simply DDR3 running at lower voltages than regular DDR3 and if that is so, does that mean that DDR3(L) has the same pin layout as DDR4?

My understanding was that it wasn't physically possible to put a DDR2 module into a DDR3 slot, but that may not be the case with DDR3(L) and DDR4.

 

[jpiniero]

My understanding was that it wasn't physically possible to put a DDR2 module into a DDR3 slot, but that may not be the case with DDR3(L) and DDR4.

Oh, it is. Unless you are talking about the UniDIMM DDR3L.

Seriously, if you are buying Skylake you are buying DDR4. Don't try to think you can cheap out of this.

 

[erunion]

This confirms the roadmap that implied all 95w skylake would be unlocked.

This pretty much guarantees that intel will continue( and expand upon) Devil Canyon's idea of using a different package for the unlocked chips. And 95w is the magic number that means solder is on its way! ( hopefully)

 

[VirtualLarry]

This confirms the roadmap that implied all 95w skylake would be unlocked.

This pretty much guarantees that intel will continue( and expand upon) Devil Canyon's idea of using a different package for the unlocked chips. And 95w is the magic number that means solder is on its way! ( hopefully)

Why would Intel go back to solder, on mainstream CPUs? DC wasn't solder, just a different TIM.

 

[erunion]

Why would Intel go back to solder, on mainstream CPUs? DC wasn't solder, just a different TIM.

Intel has traditionally soldered 95w and above chips. They stopped soldering mainstream chips with Ivy Bridge, which dropped to 77w. Haswell, even DC, were in the 80s; so no solder.

Unlocked Skylake could realistically have had an even lower TDP than DC due to being 14nm.
That Intel chose 95w exactly could signal that Intel intends to cool these parts past what is required by stock clocks.

 

[VirtualLarry]

Intel has traditionally soldered 95w and above chips. They stopped soldering mainstream chips with Ivy Bridge, which dropped to 77w. Haswell, even DC, were in the 80s; so no solder.

Unlocked Skylake could realistically have had an even lower TDP than DC due to being 14nm.
That Intel chose 95w exactly could signal that Intel intends to cool these parts past what is required by stock clocks.

Not necessarily. "95W" is a "standard" TDP class / category. Haswell and Ivy Bridge were initially "95W" too.

 

[Rvenger]

Ivy Bridge was 77w and Haswell was 84w, not 95w.

 

[Hulk]

Regarding what looks to be an extremely close tick-tock schedule for Broadwell/Haswell. We know the tick and tock teams work in parallel with a tick-tock period of approximately 12 to 18 months.

We also know that generally the new process (tick) has been taking longer than the new architecture (tock). Perhaps this is just a case of the tock being far enough ahead of the tick schedule-wise where it makes the most economic sense for Intel to move ahead with Skylake because it's ready.

To clarify what I'm stating. Broadwell has been shipping since around January. It is important for Intel to produce as much Broadwell mobile product as possible. There really is no reason to take some of that capacity for Broadwell for the desktop. There is no competition on the desktop but there is plenty in the mobile space. Maybe not directly with Windows laptops but certainly in the Broadwell Y space where they are trying to compete against laptops.

So demand is finally fulfilled with Broadwell mobile and they start producing desktop parts but only for a brief period because Skylake is ready and since the 14nm process yields are up there is no reason to wait on Skylake.

I'm actually kind of surprised they aren't skipping Broadwell for the desktop entirely with the Skylake launch being to close. But I guess they can get the desktop parts to system builders who were already using Broadwell compatible motherboards and now they can advertise them as having the newest generation Intel processor.

 

[mikk]

Does a higher base clock really help with anything?

Multithreading load for sure.