Should Intel design a CPU core with a perfomance level in between Core and Atom?

[ShintaiDK]

I don't understand what Xeon-d is missing that you think a new core arch would solve. It already crushes all x86 and arm offerings in its thermal envelope. It appears to be near the ideal perf/watt for broadwell. Why would we expect a new arch to improve on this?

Exactly. But I think he just want a new core for the sake of a new core. not due to any demand. See the previous threads.

 

[cbn]

Another processor line that I think could benefit from a mid sized Intel x86 core (designed for a single thread in between the current Core and Atom CPUs) would be the Xeon-D line (which has eight Broadwell cores with base clock of 2.0 Ghz and 2.8 Ghz turbo on the top SKU).

I don't understand what Xeon-d is missing that you think a new core arch would solve. It already crushes all x86 and arm offerings in its thermal envelope. It appears to be near the ideal perf/watt for broadwell. Why would we expect a new arch to improve on this?

Exactly. But I think he just want a new core for the sake of a new core. not due to any demand. See the previous threads.

The main issue is removing the core count limitations on the (low base clocked) high core count Xeons (E5/E7) due to power.

Reason: An x86 core designed from the ground up for lower single thread should be much more power efficient than using a x86 core designed for a much high single thread and then downclocking.

But with that mentioned, since Xeon-D works at the ~same low clocks as the Haswell Xeons I mentioned here it could also be used for that application.

 

[ShintaiDK]

Would that theoretical core actually be better on a performance/watt basis? I doubt.

Not to mention billions would have to be spend to develop it.

Again you seek to create a product with no buyers. You really need to think on this before making even more threads about odd theoretical products.

 

[cbn]

Would that theoretical core actually be better on a performance/watt basis? I doubt.

I can't see how it wouldn't be better.

 

[Ken g6]

IMO the gap between these cores needs to be smaller.

I don't disagree with this. I think Atom is too slow to be useful, and needs to be redesigned. :sneaky:

Say ~2:1 single thread ratio between Core and whatever you want to call the medium core. And ~2:1 single thread ratio between whatever you want to call the medium core and the small core.

This instead of the 3.5x gap between Core and atom.

If you have a 3.5x gap between Core and Atom, and you want a 2x gap, there's an easy solution. Make a Core chip with 1 core 2 threads.

 

[dealcorn]

I read this thread to say desktop and laptop market acceptance of Goldmont should be strong if it delivers the anticipated improvement in CPU performance. Perhaps, next quarter's release of the Goldmont TV stick gizmo will provide some meaningful insight. The definitive answer will come 2H2016 when the broader Broxton line is released.

Has Intel lost non trivial market share to alien cores that perform better than Atom but worse than Core? It does appear Intel is losing home market share to low end ARM CPU's in affordable NAS units. Is this market large enough that it makes sense for Intel to release some Atom core chip fabricated without a IGD?

 

[Zodiark1593]

I don't disagree with this. I think Atom is too slow to be useful, and needs to be redesigned. :sneaky:



If you have a 3.5x gap between Core and Atom, and you want a 2x gap, there's an easy solution. Make a Core chip with 1 core 2 threads.

I would prefer bringing Atom to current Core levels, and focus on high performance at modest power usage with Future Core. Basically, ramp up performance across the board.

 

[raghu78]

The way Apple is going in the mobile market with their custom ARMv8-A cores (Cyclone/Typhoon/Twister) Intel will be forced to either significantly beef up Atom (3 wide atleast) or kill it totally. Qualcomm and the other ARMv8 licensees are chasing Apple and Intel cannot afford to fall behind or they risk becoming irrelevant. imo Intel should just stick with their big core Cannonlake at 10nm and kill Atom. Intel should have a version for desktop/laptop/tablet/phone and another for server/workstations. Intel could make a tablet/phone optimized SKU which uses higher density low power design with a frequency tradeoff. Most mobile CPUs run at 2-2.5 Ghz. So the high density power efficient design would be perfect for the 2-2.5 Ghz range.

With Skylake Intel has started the trend with their server cores having significant differences (avx-512) compared to desktop. It makes sense to continue that trend and just stick with variations of their big core to address the entire range of computing products.

 

[Thanatosis]

I have started to question whether or not intel even has the ability to design a core with performance analagous to current custom ARM cores. They appear to have tried with core M and failed, and tried with Atom and also failed. Maybe that's due to late entry into the mobile sphere, or maybe it's due to a lack of funding.

Personally, I believe it was pure hubris on the part of intel's board to fund server and datacenter research and basically spin off from that (with smaller cores) on other markets. That strategy worked for years because of the x86 lock-in, but it has left them flat-footed when competition in the form of ARM and Apple Ax appearing in their high-margin mobile market products (the chief example being A9 vs core M). Instead of fully funding a custom, original x86 design for their notebook segment they just took a datacenter core and lopped off an much capability as necessary to bring the TDP down to manageable levels. Now, with ARM offering even superior perf/w to Core they are in desperate need of a new design to compete.

And they haven't changed their strategy, they have the same hubris today with Atom being jammed down everyone's throat. They are going to take Microsoft with them and MS knows it, that's why they are working to port everything to ARM and make Windows 10 capable of running on anything.

Intel would do well to drop their margins substantially and use their assets to make a concerted effort to actually out-design their competitors instead of purely out-competing them. I don't think they care though, they have that fat captive x86 server segment to fall back on.

 

[cbn]

If the above roadmap time frame ends up being correct (and there are no delays), then I would expect the Cannonlake follow-up to the Skylake (Purely Platform) probably Q3 2018. Then perhaps the 10nm Toc Xeon Q4 2019/early 2020.

That leaves Intel around four years (perhaps more) to release a uarch for high core count (low base clocked) Xeon using the 10nm Toc.

 

[cbn]

Regarding Intel Quark (which uses the same instruction sets as the P54C), I do wonder how they will evolve that processor.

Its x86, but missing instruction sets from the last 20 years and lacks SIMD.

So what does Intel do with this one?

Do they evolve it into something else? Perhaps adding a little iGPU (and then using it with new instruction sets not found on the current x86 processors)?

 

[VirtualLarry]

Intel would do well to drop their margins substantially and use their assets to make a concerted effort to actually out-design their competitors instead of purely out-competing them. I don't think they care though, they have that fat captive x86 server segment to fall back on.

I've said it before, and I'll say it again. Unless Intel changes their approach substantially, they're going to regress to only being a server company, like IBM is today. That is, assuming that ARM continues to eat desktop/laptop market-share from below, just like Intel's desktop chips ate RISC workstation market share from below.

 

[ShintaiDK]

I've said it before, and I'll say it again. Unless Intel changes their approach substantially, they're going to regress to only being a server company, like IBM is today. That is, assuming that ARM continues to eat desktop/laptop market-share from below, just like Intel's desktop chips ate RISC workstation market share from below.

ARM didn't take any desktop/laptop yet. x86 however took a good chunk of tablets from ARM.

 

[VirtualLarry]

ARM didn't take any desktop/laptop yet.

ARM-powered phone/phablet/tablet volume up, desktop/laptop volume down, and you really want to make that statement?

x86 however took a good chunk of tablets from ARM.

Only with the un-sustainable contra-revenue. If a big company is paying you to take their chips and use them, you would be a fool not to, right? Well, until the gravy-train comes to an end.

 

[cbn]

ARM didn't take any desktop/laptop yet. x86 however took a good chunk of tablets from ARM.

There have been ARM chromebooks (even a Rockchip Cortex A17 one) and ARM NASes. Not sure about ARM Chromeboxes though. (EDIT: There have also been ARM Android mini boxes and ARM Android compute sticks)

Then there is the Apple iPad pro which is a quasi laptop when used with the Smart Keyboard:

http://www.apple.com/smart-keyboard/

 

[Essence_of_War]

ARM-powered phone/phablet/tablet volume up, desktop/laptop volume down, and you really want to make that statement?

Eh, I saw what you were trying to say, but I think a reasonable person could have read it as saying that ARM was making in-roads into the desktop and laptop market.

 

[ShintaiDK]

ARM-powered phone/phablet/tablet volume up, desktop/laptop volume down, and you really want to make that statement?

Tablet volume is down. And next year everything points to smartphone segment will be shrinking.

Qualcomm also posted a horrible result today.

And as said, you missed the entire context.

 

[Qwertilot]

Well maybe, but its surely very hard to realistically maintain that a big chunk (OK, maybe not all) of the tablet sales aren't effectively displaced laptops.

Not that this is an either/or thing of course - don't think either Arm or Intel are precisely complaining just now

 

[cbn]

Regarding this mid size core (in between Core and the current atom) discussion, one thing I forgot to mention is process technology.

With a SoC process tech, this mid size core could be a wide design but designed for a much lower maximum frequency. (And, in addition to the smaller uarch, reap the benefits of the lower leakage SoC process tech)

P.S. As I mentioned in a previous post, sure the 1C turbo won't be high....but with a high core count Xeon (especially one used for VDI with a lot of OS instances running) most (if not all) of the cores should be loaded up leaving little if any thermal headroom for boosting one or two cores to a high frequency.

 

[raghu78]

Regarding this mid size core (in between Core and the current atom) discussion, one thing I forgot to mention is process technology.

With a SoC process tech, this mid size core could be a wide design but designed for a much lower maximum frequency. (And, in addition to the smaller uarch, reap the benefits of the lower leakage SoC process tech)

P.S. As I mentioned in a previous post, sure the 1C turbo won't be high....but with a high core count Xeon (especially one used for VDI with a lot of OS instances running) most (if not all) of the cores should be loaded up leaving little if any thermal headroom for boosting one or two cores to a high frequency.

At 10nm Intel can use a variant of their Cannonlake core and use a high density low power process and achieve excellent perf and perf/watt in the 4-5W TDP range. Intel has to use Core M for high end tablets and phones and use Atom for the low end of the mobile market. Atom based SKUs should also be less than 80 sq mm while Core M based SKUs should be able to go up to 120 sq mm as the higher margins would allow to do so. Intel has an opportunity to gain profitable market share in the mobile market if they do that. Right now Intel's mobile business is stuck with the weak Atom cores. Cherrytrail is basically hopeless against Apple A9. Snapdragon 820 should be another strong competitor. Intel's Goldmont better aim very high or else they risk being irrelevant at launch.

 

[cbn]

At 10nm Intel can use a variant of their Cannonlake core and use a high density low power process and achieve excellent perf and perf/watt in the 4-5W TDP range.

If we look at the following thread where I posted results of Broadwell core sizes, the Core M die, the 2C GT3 and 4C GT3 dies actually all have cores measuring in the 6.8mm2 to 6.9mm2 range:

http://forums.anandtech.com/showthread.php?t=2433600

So up to that point it appears a "high density low power process" wasn't being used with Core.

For Skylake, I have yet to measure cores (due to the processor floor maps not being released yet).

P.S. If Intel did in fact use "high density low power process" for for 10nm, maybe they could do it with the 10nm Toc processor.......but I do hope the architecture would be redesigned to match the electrical characteristics of the "high density low power low leakage SoC process". This basically would become our medium size core. Either that or they could widen up a design based on atom and make whatever comes out of that the new medium core.