Broadwell to Skylake, will be similar to the transition from Prescott to Conroe

[Dave2150]

Broadwell for Desktop is cancelled. We've been talking about it for awhile.

Skylake leaks show it coming for back to school 2015.

Not sure who you've been talking to, but the rest of us have been discussing the 65w Broadwell desktop chips that Intel confirmed to be launching in Q2.

http://blogs.intel.com/technology/2015/03/gdc-2015/

 

[ninaholic37]

This thread made me Lol. Comparing Broadwell to Prescott = priceless :biggrin:

 

[erunion]

Not sure who you've been talking to, but the rest of us have been discussing the 65w Broadwell desktop chips that Intel confirmed to be launching in Q2.

http://blogs.intel.com/technology/2015/03/gdc-2015/

If you're going to have an attitude, you better at least be correct.

You said that skylake probably wont come until 2016 because broadwell desktop isn't out yet, which is not true. Broadwell K is coming 1 quarter earlier than SKL-S. And Broadwell K will continue to sell after SKL-S launches. Broadwell K will be niche 65w Iris Pro parts, not replacements for mainstream i5s and i7s.

 

[magomago]

how about we just wait for actual performance instead of incessantly speculating?

I understand that cpu tech has slow down and we are itchy for more. My 2500K in general is giving me little reason to upgrade (although stitching copious qtys of photos can always use more power, and if I could double my speed I'd totally do it), but lets not pretend what the results are before anything is actually known.

think about it...its a thread about skylake, and we are looking at performance charts of presscott lol. That shows how much info we really have on our hands.

 

[frozentundra123456]

Where are the quad core mobile chips? Is 28 watt dual core the fastest they are going to offer in laptops? Doesnt seem like a very complete map. Low power broadwell is not shown on there either and it is on the market now.

 

[erunion]

Where are the quad core mobile chips? Is 28 watt dual core the fastest they are going to offer in laptops? Doesnt seem like a very complete map. Low power broadwell is not shown on there either and it is on the market now.

Lots of products are missing. Dont let the inclusion of BDW-U fool you, its a desktop roadmap.

 

[Dave2150]

If you're going to have an attitude, you better at least be correct.

You said that skylake probably wont come until 2016 because broadwell desktop isn't out yet, which is not true. Broadwell K is coming 1 quarter earlier than SKL-S. And Broadwell K will continue to sell after SKL-S launches. Broadwell K will be niche 65w Iris Pro parts, not replacements for mainstream i5s and i7s.

My post which you quoted was correct - Broadwell is coming for desktop.

You were unaware of this, here, I'll quote your post where you said so:

Broadwell for Desktop is cancelled. We've been talking about it for awhile.

So, you were incorrect. Not me. I notice you edited your post now, though it's still visible in the posts of a few others who corrected your foolishness.

Regarding my comment of Skylake's release date - that's just my opinion, I think it will probably be delayed to 2016.

None of us actually know the official release date - Intel haven't set one. We just have a guideline of Q3 from leaked slides/sources - which they may or may not meet. We can all have our own personal opinion about whether it will launch on time or not.

Broadwell was also delayed - it should have fully launched last year with all skews available - instead it was only a few select skews in Q4, with the rest of the broadwell lineup launching in 2015.

Learn to differentiate from facts and opinions:

(Fact)It's a fact that Broadwell for desktop hasn't been cancelled and is launching this year, as confirmed by Intel's official website.

(Opinion)It's my opinion that Skylake could be delayed to next year for desktop. I may be right, I may be wrong.

 

[erunion]

My post which you quoted was correct - Broadwell is coming for desktop.

You were unaware of this, here, I'll quote your post where you said so:



So, you were incorrect. Not me. I notice you edited your post now, though it's still visible in the posts of a few others who corrected your foolishness.

Regarding my comment of Skylake's release date - that's just my opinion, I think it will probably be delayed to 2016.

None of us actually know the official release date - Intel haven't set one. We just have a guideline of Q3 from leaked slides/sources - which they may or may not meet. We can all have our own personal opinion about whether it will launch on time or not.

Broadwell was also delayed - it should have fully launched last year with all skews available - instead it was only a few select skews in Q4, with the rest of the broadwell lineup launching in 2015.

Learn to differentiate from facts and opinions:

(Fact)It's a fact that Broadwell for desktop hasn't been cancelled and is launching this year, as confirmed by Intel's official website.

(Opinion)It's my opinion that Skylake could be delayed to next year for desktop. I may be right, I may be wrong.

I wasn't aware of Broadwell K coming to desktops? Look at my post history, I've been talking about it for months.

My post edit added information, didn't delete anything.

You're playing semantics, so this will be my last post to you.

Broadwell desktop is codenamed BDW-S. That is cancelled, as I've shown you. Two broadwell products will come to LGA 1150, i5- 5765c and i7-5775c. Both BDW-H and Iris Pro. That's it!

If you want to call two niche SKUs targeted at AIO and HTPCs, a broadwell desktop launch, go ahead. But I remain comfortable with my original assessment:

Broadwell desktop is cancelled and Intel will transition from haswell refresh straight to skylake in q3 2015

 

[NTMBK]

The other change that Kanter alluded to was a change in the interconnect fabric: from the current ring bus, to a 2D mesh. I don't really know many of the details surrounding this, although I'm fairly certain it'd be a good improvement to inter-core communication, as I recall ring buses being a pretty outdated concept.

It will make the most difference in super high core count Xeons. Ring bus does not scale well beyond where Intel are at today- the more cores you have, the more stops there are on the ring, which increases average latency to access an L3 slice (or snoop the private caches on other cores), increasing memory latencies in general. If I remember rightly, average latency scales linearly with the number of stops on the ring (i.e. number of cores).

This came back to bite them on the Xeon Phi- >60 cores, all with their own stop on the ring, all cache coherent. There was no shared LLC, so the increasing L3 latency wasn't a problem, but there was still a hell of a lot of coherency traffic flying around. Knight's Landing moves to a 2D mesh, with cores combined into two core clusters with shared L2. The biggest Ivy Bridge had three different overlapping ringbuses, and the biggest Haswell had two completely separate bidirectional rings connected by a switch- it was essentially a NUMA system-on-a-chip, and you could even configure it to behave that way in the BIOS to improve latencies on NUMA-aware apps. They were pretty obviously hitting the limit of the ringbus in their largest 22nm designs.

From my layman's understanding of the mesh fabric, average latencies should roughly scale with the square-root of the number of nodes in the mesh (as opposed to linearly with the ringbus). This should make it easier to scale to very high numbers of cores, but I doubt it will make any noticeable difference to consumer parts with ~4 cores and a GPU.

 

[Dave2150]

I wasn't aware of Broadwell K coming to desktops? Look at my post history, I've been talking about it for months.

My post edit added information, didn't delete anything.

You're playing semantics, so this will be my last post to you.

Broadwell desktop is codenamed BDW-S. That is cancelled, as I've shown you. Two broadwell products will come to LGA 1150, i5- 5765c and i7-5775c. Both BDW-H and Iris Pro. That's it!

If you want to call two niche SKUs targeted at AIO and HTPCs, a broadwell desktop launch, go ahead. But I remain comfortable with my original assessment:

Broadwell desktop is cancelled and Intel will transition from haswell refresh straight to skylake in q3 2015

You dug your own grave when you edited your incorrect post.

Why edit it, if you 'remain confortable' with your incorrect statement, hmm.

Broadwell for LGA1150 (desktop...) is not cancelled and is coming next quarter.

 

[crashtech]

Seems like we've all been left flailing a bit for lack of info. I thought there'd be more clarity by now.

 

[Hulk]

I hope you are not saying performance-wise Broadwell to Skylake will be like Prescott to Conroe.

If you are then I have one word for you IMPOSSIBLE!

All of the low hanging fruit was plucked long ago. IPC gains are very hard to come by.

Now for some facts. Prescott was a 31 instruction long pipeline processor. It was the last, dying breath of Netburst and Intel's quest to bring us 10GHz processors. Netburst was a failure, a wrong turn, and a huge misstep for Intel.

Conroe on the other hand was a return to sanity. Based on the earlier Pentium III architecture with a 14 stage pipeline Conroe was had fully TWICE the IPC as Prescott. Yes, TWICE. 100% improvement. Not 5% or 10% improvement but 100% per clock.

Skylake will be lucky to eek out 1/10 of that.

Now it would be absolutely fantastic if I were wrong though...

 

[III-V]

From my layman's understanding of the mesh fabric, average latencies should roughly scale with the square-root of the number of nodes in the mesh (as opposed to linearly with the ringbus). This should make it easier to scale to very high numbers of cores, but I doubt it will make any noticeable difference to consumer parts with ~4 cores and a GPU.

Ah, that is a very good observation. You are most likely correct.

 

[Enigmoid]

You dug your own grave when you edited your incorrect post.

Why edit it, if you 'remain confortable' with your incorrect statement, hmm.

Broadwell for LGA1150 (desktop...) is not cancelled and is coming next quarter.

You two are arguing over schematics.

Depending on how you phrase the question either could be true. There will be a few desktop chips but the majority of broadwell for desktop has been canceled. Iris pro broadwell seems to be thrown in as an afterthought, likely as lower bins for a few broadwell quad GT3e chips that will appear in the imac or on mobile.

 

[crashtech]

Intel can't feel good about completely leaving enthusiasts high and dry this time around. Seems unprecedented. Isn't there a bit of a sense of disbelief about this?

 

[Sweepr]

I think they wouldn't have released Devil's Canyon if they knew that a (probably) lower-clocked Skylake wouldn't be able to outperform it. Just my 2 cents.

 

[NTMBK]

Intel can't feel good about completely leaving enthusiasts high and dry this time around. Seems unprecedented. Isn't there a bit of a sense of disbelief about this?

They must be quaking in their boots at the prospect of AMD stealing all those sales :awe:

 

[NTMBK]

I think they wouldn't have released Devil's Canyon if they knew that a (probably) lower-clocked Skylake wouldn't be able to outperform it. Just my 2 cents.

At that time they may have still expected 14nm to clock better than it has.

 

[crashtech]

They must be quaking in their boots at the prospect of AMD stealing all those sales :awe:

Obviously they aren't worried about it from a business perspective, but certainly it's a blow to the pride of some there if there just isn't anything at all that can hold a candle to DC.

 

[frozentundra123456]

I hope you are not saying performance-wise Broadwell to Skylake will be like Prescott to Conroe.

If you are then I have one word for you IMPOSSIBLE!

All of the low hanging fruit was plucked long ago. IPC gains are very hard to come by.

Now for some facts. Prescott was a 31 instruction long pipeline processor. It was the last, dying breath of Netburst and Intel's quest to bring us 10GHz processors. Netburst was a failure, a wrong turn, and a huge misstep for Intel.

Conroe on the other hand was a return to sanity. Based on the earlier Pentium III architecture with a 14 stage pipeline Conroe was had fully TWICE the IPC as Prescott. Yes, TWICE. 100% improvement. Not 5% or 10% improvement but 100% per clock.

Skylake will be lucky to eek out 1/10 of that.

Now it would be absolutely fantastic if I were wrong though...

That (Prescott to Conroe type improvement) was what the article linked in the OP said.

I dont believe it will be close to that either, but there is a "source" for such speculation. I would be more than happy just to see 20 percent vs Haswell, since it is 2 generations.

 

[III-V]

That (Prescott to Conroe type improvement) was what the article linked in the OP said.

I dont believe it will be close to that either, but there is a "source" for such speculation. I would be more than happy just to see 20 percent vs Haswell, since it is 2 generations.

Broadwell's about 5-6% from some relatively minor changes, so I think 20% is within the realm of possibility. Still, I am thinking it will be about 10-15% compared to Haswell.

 

[erunion]

Intel can't feel good about completely leaving enthusiasts high and dry this time around. Seems unprecedented. Isn't there a bit of a sense of disbelief about this?

Everyone was "high and dry" when we got stuck with haswell refresh. Intel did throw enthusiasts a cookie in the form of devils canyon, though.

I actually see no broadwell as good news. 2014 was a slump because broadwell wasn't ready, but with skylake in 2015 that puts Intel back on track (roughly). A way better alternative than artificially delaying skylake to sell us a year old Broadwell lineup in 2015.

 

[Hulk]

Broadwell's about 5-6% from some relatively minor changes, so I think 20% is within the realm of possibility. Still, I am thinking it will be about 10-15% compared to Haswell.

I would bet it will be more like 5% again. In fact I bet Intel has all of the gains for the next 3 or 4 processors laid out in a very methodical manner. Since there isn't any competition there isn't any reason to "let it all go" at once.

What they seem to be doing it over engineering one or two parts of the CPU to increase IPC so that the bottleneck is not only gone but opens things up so much that another area becomes the bottleneck, which is then opened up on the next iteration.

Intel could have done this all at once but there was no need.

For example, new branch predictor for Sandy Bridge.
Then Ivy gets dynamically partitioned internal structures and prefetcher improvement.
Then Haswell gets a wider execution engine...

 

[jihe]

Broadwell to Skylake will be like Prescott to Conroe........



















Pricewise.

 

[jpiniero]

I would bet it will be more like 5% again. In fact I bet Intel has all of the gains for the next 3 or 4 processors laid out in a very methodical manner. Since there isn't any competition there isn't any reason to "let it all go" at once.

True, at least for the products coming out now. But I imagine it's going to be more difficult at Intel to get anything that increases power consumption for future products, even if it meets the famous 2% for 1%.

Take the removal of the FIVR. If that ends up hurting Skylake mobile's power consumption, it would be a disaster. The decision to remove it happened a long time ago.