[FUDZILLA] No Broadwell for Desktop

[dbcoopernz]

And two year cycle for desktop updates.

http://www.fudzilla.com/home/item/32524-broadwell-won’t-make-it-to-desktop

We have heard people talking about it but we managed to confirm it now. Intel will make Broadwell mobile only and it won’t make a desktop version like it used to before.

The desktop version of a new CPU is codenamed Haswell refresh, which gives you idea of what is coming from Intel on the desktop side. Haswell refresh is just a slightly improved and optimized Haswell, while Broadwell will hit tablets, convertibles, two-in-ones and the traditional notebook market. Intel spends a lot of time addressing this market, which has already fallen on hard times.


Our sources are telling us that Intel will do desktop chip every two years, which is a huge slow down from the current roadmap, making it much easier for AMD to grab some breathing space in the desktop segment.


This is definitely not great news for desktop enthusiasts. It will make things slower and boring on the desktop side but GPUs will still come out every year, making GPU updates for gamers look like a good way of boosting performance.


Skylake is the next Intel chip that is set to have a desktop version and it is a tock, a new architecture processor that can probably bring more performance than what you would expect from the Broadwell 14nm shrink. Skylake is 14nm and it is set to come at some point in 2015.


Typical non-desktop users won’t benefit much from new processor unless they use heavy video encoding and rendering and watching Full HD movies and even 4K movies won’t be much of a task in the future.


According to Intel’s own figures, the average PC is now more than four years old, the PC market is mature and upgrade cycles are getting longer, which explains the shift to a two-year cycle in the desktop segment.

 

[MisterMac]

Well - everyone knew this more or less.

(Even if it's fudzilla reporting).



What i hope is - they'll start shifting more server tech to enthusiast levels
Perhaps move i7 brand to E only - and make a cheaper lines and scale more cores.


Look at the top DIE SKU's scaling of Ivy E5-EP lines.
We need that our in enthusiast segment - so the software ECO will be interested in catching up.

 

[frozentundra123456]

So really this is not a complete doubling of the upgrade cycle, but just a skipping of the "tick" on the desktop. Still disappointing, but perhaps there will be more progress with each each new model.

Perhaps it will give amd some breathing room, but if they are going all apu as it seems, I doubt they will pass Intel in CPU performance.

 

[ShintaiDK]

So really this is not a complete doubling of the upgrade cycle, but just a skipping of the "tick" on the desktop. Still disappointing, but perhaps there will be more progress with each each new model.

Perhaps it will give amd some breathing room, but if they are going all apu as it seems, I doubt they will pass Intel in CPU performance.

Desktop sales are just becoming too irrelevant. AMD is more or less heading the same path with APUs only. Because you cant use a FX style CPU in a laptop. And if you note, AMDs 2015 CPUs will be 45W and 65W on the desktop. Even 100W SKU is dead.

Mobile chips are also becoming "desktop". See Mac Mini, Intels NUC and so on.

But if you really want, you can buy a Broadwell E3 Xeon and put in a 9 series board.

 

[USER8000]

Another thing is that DDR4 is probably going to start being cheap enough,to use as system RAM in 2015 too. Hence,I suspect both AMD and Intel probably are looking at this too.

 

[frozentundra123456]

Desktop sales are just becoming too irrelevant. Same is more or less heading the same path with APUs only. Because you cant use a FX style CPU in a laptop.

But if you really want, you can buy a Broadwell E3 Xeon and put in a 9 series board.

Even more concerning is what will happen when it becomes impractical to continue die shrinks. Not sure when this will happen, but it could be in another cycle or two.

Maybe Intel will finally be forced to add more cores to the mainstream desktop. Or unfortunately, they just might say desktop is good enough and not bother.

 

[ShintaiDK]

Another thing is that DDR4 is probably going to start being cheap enough,to use as system RAM in 2015 too. Hence,I suspect both AMD and Intel probably are looking at this too.

Yep, its obvious only Haswell-E will carry DDR4 in 2014.

 

[ShintaiDK]

Even more concerning is what will happen when it becomes impractical to continue die shrinks. Not sure when this will happen, but it could be in another cycle or two.

Maybe Intel will finally be forced to add more cores to the mainstream desktop. Or unfortunately, they just might say desktop is good enough and not bother.

Die shrinks will continue for a long foreseeable future. And we will get them every second year from Intel.

I think you misunderstood the statement. CPUs and node shrinks is not irrelevant, nor becoming it. The desktop is becoming irrelevant. Hence laptop, servers, tablets and smartphones will get the latest node shrink before the desktop.

And you can buy more cores with the E series. Nothing is hindering you to get 6 cores today and 8 cores next year. Why do you think AMD is going from 8 cores to 4 again on the desktop? There is no sales in "moar cores" for average joe.

 

[blackened23]

There have been so many mixed messages regarding Desktop LGA broadwell that I don't know what to think. The one fact remains clear, though: it doesn't make sense for intel to release "Broadwell" and "Haswell refresh" at the same time - so this does sound plausible. It would be downright stupid for intel to do a "Haswell refresh" simultaneously with a LGA broadwell chip.

The other fact is that Broadwell is designed as a very low leakage 14nm chip for mobility and efficiency - these will serve the server, enterprise, data center, and mobile markets VERY well. On the desktop LGA? Not so much. It will be about as well received as Haswell was (and I think Haswell was a good chip, but people had inflated expectations). Here's what would happen with Broadwell LGA: reviews hit. People whine about everything and anything. They can't get 2ghz overclocks past the base stock clockspeeds. IPC increase is "only" 10%. It was the same story with IVB and Haswell, but I don't think everyone understands that intel has to design the uarch for efficiency and not all-out IPC.

It is what it is I guess. Intel is clearly gunning for markets that sell. Currently, joe average consumer doesn't buy a desktop nor does he even consider it. Intel still has the enthusiast PC gamer market, but really - everyone else outside of that buys a mobile product such as a tablet or laptop. So if this news is true I completely understand it - intel is doing what's good for their business. Going for the markets that sell (mobile, enterprise, data centers, etc) as opposed to the one that doesn't grow or sell all that great - for the markets that are growing, efficiency matters more than anything else.

 

[sm625]

If all it offers is lower power then I dont see why anyone would want to pay the premium for it on the desktop. Who cares about an extra 13 watts on a desktop? Especially if you're running a GPU that can pull over a hundred or even two hundred?

 

[USER8000]

People need to consider one important factor - cost.

Remember,Anandtech mentioned at the end of last year over 50% of the 22NM capacity Intel had,was not used.

Fast forward one year and now we have chipsets,Bay Trail,Haswell, Haswell E and MIC ensuring that 22NM is well utilised for the next year or so.

This means Intel can make more money on its expensive 22NM development and new process nodes are costing more and more anyway.

Hence,it also makes sense that Intel is first using 14NM for products where it counts the most and they can make the most money,ie,mobile workstation CPUs,high performance thin and light CPUs and CPUs with large IGPs.

It also means they can ramp up 14NM production in a more cost effective manner instead of moving all CPU lines in a short space to 14NM,which would mean more capacity is needed in a brief period. It also gives them breathing space in case there are issues encounted along the line such as yield problems,etc.

 

[ShintaiDK]

Broadwell adds a few new instructions. Better IGP (Up to 2Tflop shader with GT4.) and lower power consumption (roughly 30% lower).

ADOX/ADCX/MULX for improving performance of arbitrary-precision integer operations
RDSEED to generate 16-, 32- or 64-bit random numbers according to NIST SP 800-90B and 800-90C
PREFETCHW instruction

 

[ShintaiDK]

People need to consider one important factor - cost.

Remember,Anandtech mentioned at the end of last year over 50% of the 22NM capacity Intel had,was not used.

Fast forward one year and now we have chipsets,Bay Trail,Haswell, Haswell E and MIC ensuring that 22NM is well utilised for the next year or so.

This means Intel can make more money on its expensive 22NM development and new process nodes are costing more and more anyway.

Hence,it also makes sense that Intel is first using 14NM for products where it counts the most and they can make the most money,ie,mobile workstation CPUs,high performance thin and light CPUs and CPUs with large IGPs.

It also means they can ramp up 14NM production in a more cost effective manner instead of moving all CPU lines in a short space to 14NM,which would mean more capacity is needed in a brief period. It also gives them breathing space in case there are issues encounted along the line such as yield problems,etc.

There is alot of wrong in your post. First of all Intel didnt have 50% 22nm capacity idle. But overall in Intel, its normal to have a utilization around 50% due to fab conversions.

Chipsets are made on 32nm, not 22nm.

Intel brings 3 new large fabs online for 14nm. D1XM1, D1XM2 and 42. All 450mm ready as well.

 

[tweakboy]

You can get a 20 core dual socket xeon. 40c 80 threads. Not really broadway but Macy's. gl

 

[Pilum]

Well, after the Baytrail-T previews, this shouldn't come as a surprise. It's all about fab capacity management. Silvermont already is a highly competitive architecture and Atoms will now begin to sell in volume. When Airmont comes out next year, it should be a performance leader across a wide range of mobile/ultra-mobile; it will probably sell like hot-cakes and that will require lots of 14nm capacity. And that capacity has to come from somewhere - even Intel can't convert all its fabs to the leading process at once. Skipping the DT tick is the rather obvious answer.

For people interested in better perf/W, there are the BRW Xeon E3s. Normal DT users won't know or care. The enthusiasts won't have to spend so much energy and time on bitching how disappointing the new Intel CPUs are. And everybody can buy Airmont-derived tablets and phones next year.

Win-Win.

 

[frozentundra123456]

Die shrinks will continue for a long foreseeable future. And we will get them every second year from Intel.

I think you misunderstood the statement. CPUs and node shrinks is not irrelevant, nor becoming it. The desktop is becoming irrelevant. Hence laptop, servers, tablets and smartphones will get the latest node shrink before the desktop.

And you can buy more cores with the E series. Nothing is hindering you to get 6 cores today and 8 cores next year. Why do you think AMD is going from 8 cores to 4 again on the desktop? There is no sales in "moar cores" for average joe.

Nothing but the price for anything more than a quad core chip.

I know you think those who want more cores from intel at mainstream prices are expecting something for nothing, but I have to disagree. I dont recall the timeline exactly, but I know quad cores came out very close to duals, maybe within a year. Now we have shrunk from 65nm to 22nm since the initial quads came out, and still no increase in core count in what, 5 or 6 years. (I am referring to mainstream desktop chips).

As far as node shrinks, I am not a physicist or engineer, but we are already seeing diminishing returns. Eventually we will reach a point of physical limitation or too high cost to continue. That is what I meant. 14nm will be next year in mobile, then what, 10nm? Not sure it is practical to go lower than that.

 

[ShintaiDK]

Nothimng but the price for anything more than a quad core chip.

I know you think those who want more cores from intel at mainstream prices are expecting something for nothing, but I have to disagree. I dont recall the timeline exactly, but I know quad cores came out very close to duals, maybe within a year. Now we have shrunk from 65nm to 22nm since the initial quads came out, and still no increase in core count in what, 5 or 6 years. (I am referring to mainstream desktop chips).

As far as node shrinks, I am not a physicist or engineer, but we are already seeing diminishing returns. Eventually we will reach a point of physical limitation or too high cost to continue. That is what I meant. 14nm will be next year in mobile, then what, 10nm? Not sure it is practical to go lower than that.

Mark Bohr already stated the road for 5nm is clear. And he is perhaps the best knowing person on the planet in that regard.

Your rationale for more cores is weak. More cores is replaced by IGP and bigger IGP. If you want more than 4 cores, you belong to a tiny, tiny niche. And your niche product is the E series. Without servers there wouldnt even be economics behind to make those niche products.

 

[frozentundra123456]

Mark Bohr already stated the road for 5nm is clear. And he is perhaps the best knowing person on the planet in that regard.

Your rationale for more cores is weak. More cores is replaced by IGP and bigger IGP. If you want more than 4 cores, you belong to a tiny, tiny niche. And your niche product is the E series. Without servers there wouldnt even be economics behind to make those niche products.

I dont really care about anything more than a minimal igp on the desktop. Lack of more cores would be somewhat mitigated if the gains in IPC had continued, but since Sandy Bridge, IPC gains have been minimal to non-existant as well. Hence the only way left to increase performance seems to be more cores or higher clockspeed, which Intel is apparently unwilling to give, or perhaps the gain in clockspeed has topped out as well.

 

[zir_blazer]

I agree that I don't get why Intel mentions both "Broadwell" and "Haswell refresh" like if they were two completely different products. I recall that some years ago, Intel was saying that if 22nm/FinFET wasn't mature enough to do Haswell, they were going to do an stopgap 22nm product, that was to be after Ivy Bridge but before Haswell (A third Bridge, after Sandy and Ivy). It never materialized, but "Haswell refresh" sounds too similar to that.
Also, it is confirmed that Broadwell will not be compatible with current LGA 1150 infrastructure? Considering what was stated about Intel two-products-per-platform cycles, I find hard to believe that even if they were going to release a Xeon E3 Broadwell you would need a new Motherboard for it. If that's the idea, then they may as well go fully embedded.

 

[CakeMonster]

I'm still curious about what the Haswell refresh will bring though. I'm not expecting much, but since it is a refresh... something has to have been changed or updated.

Makes you wonder if they will continue to actually release updated "refreshed" models in the future too. Two years is a long time (or more, usually 1yr + x months for cycles)

 

[ShintaiDK]

I dont really care about anything more than a minimal igp on the desktop. Lack of more cores would be somewhat mitigated if the gains in IPC had continued, but since Sandy Bridge, IPC gains have been minimal to non-existant as well. Hence the only way left to increase performance seems to be more cores or higher clockspeed, which Intel is apparently unwilling to give, or perhaps the gain in clockspeed has topped out as well.

There would be a 1000 different SKUs then. Everyone feeling they are entitled to some specific configuration that fits their exact needs and wishes. You simply need to accept you belong to a niche group. Broadwell gives exactly what the very vast majority wants-

 

[ShintaiDK]

I'm still curious about what the Haswell refresh will bring though. I'm not expecting much, but since it is a refresh... something has to have been changed or updated.

Makes you wonder if they will continue to actually release updated "refreshed" models in the future too. Two years is a long time (or more, usually 1yr + x months for cycles)

I have a feeling its just a new platform with SATA Express. Maybe a speedbin or 2. And Broadwell Xeon support.

 

[beginner99]

Well, after the Baytrail-T previews, this shouldn't come as a surprise. It's all about fab capacity management. Silvermont already is a highly competitive architecture and Atoms will now begin to sell in volume. When Airmont comes out next year, it should be a performance leader across a wide range of mobile/ultra-mobile; it will probably sell like hot-cakes and that will require lots of 14nm capacity. And that capacity has to come from somewhere - even Intel can't convert all its fabs to the leading process at once. Skipping the DT tick is the rather obvious answer.

exactly my thoughts too.

 

[SiliconWars]

There is alot of wrong in your post. First of all Intel didnt have 50% 22nm capacity idle. But overall in Intel, its normal to have a utilization around 50% due to fab conversions.

It's not normal, that's why they took capacity underutilization charges two quarters in a row.

 

[Paul98]

So is the mobile market getting the upgrades every year similar to what the desktop market has been getting? Then the desktop market is getting chips about a year after the die shrink on the mobile market?