Intel plans to reduce performance of next-generation CPUs

[waltchan]

Keep in mind that Pentium G3258 capable in overclocking up to 4.4 GHz max will be almost 2x times faster than after Skylake. It certainly will help its maximum resale value 5 years later as used.

 

[cytg111]

This...



...and this :thumbsup:

First off, thank you for the collaborated collective sawn off shutgun insult, I am sure we all appreciate it .

I dont market myself as a reading comprehension champion so this may be on me, however, I dont see it as clear cut that the future of desktop performance(single thread) is not in jeopardy here :

The company isn’t abandoning silicon — it envisions islands of additional capability embedded in SoCs or other types of circuits — but its move to purchase Altera and its FPGA business could reflect long-term plans for the future of traditional semiconductor performance. If traditional CPU designs can’t provide additional clock speeds and next-generation technologies are aimed at lower-power computing as opposed to higher performance, than either we’re headed for a revolution in distributed computing (unlikely), or a very, very slow performance ramp.

If they're gonna tool their fabs in a certain manner its gonna hit the whole spectrum of products, correct?
Maybe they're confident that they can stretch the process even broader and thinner than now?
Maybe its the only path destined for all competitors left in 2020+?
Maybe Intel will regress on ST performace, maybe a competing product will take the ST performance crown - and it will matter jack(cause segment is dead).
Change is coming?

 

[witeken]

From what I've read (IIRC), it looks like Intel wants to combine different kinds of transistors. So they will continue to use the highest performance transistors for desktop, but have option to also use lower power transistors like TFETs (on the same die) for less performance intensive parts.

Also, Intel will of course do its best to keep performance at least flat, like they've done with the finFET. But yeah, spintronics and (certainly) TFET likely won't ever meet performance of HP FET, but I think they could coexist together since they do have very appealing characteristics.

It's like Quark/Atom/Core: just like one architecture to rule em all is a bit too optimistic, maybe one transistor to rule em all also isn't going to be the best thing going forward.

 

[waltchan]

I believe the future Intel CPUs are returning back to Sandy Bridge performance level, but with 5 times faster graphics, DDR4, and 35W TDP standard power instead. Who accepts it?

 

[crashtech]

I believe the future Intel CPUs are returning back to Sandy Bridge performance level, but with 5 times faster graphics, DDR4, and 35W TDP standard power instead. Who accepts it?

What segment? Mainstream desktop? Maybe, though I'd think they could do a bit better than SB especially when/if new instruction sets become more widely used. If so, the dividing line between mainstream and HEDT will get a lot wider than it is today.

 

[VirtualLarry]

I believe the future Intel CPUs are returning back to Sandy Bridge performance level, but with 5 times faster graphics, DDR4, and 35W TDP standard power instead. Who accepts it?

Seeing as how AMD was not able to charge a premium in the market for their faster iGPUs, but instead pricing seemed to be dictated in a large part by the ST and less so the MT performance of their CPU cores, then will Intel be force to lower prices for their newer CPUs if they have regressed in CPU performance, regardless of power efficiency or iGPU prowess. (Don't forget how poor Intel's drivers are.)

 

[waltchan]

What segment? Mainstream desktop? Maybe, though I'd think they could do a bit better than SB especially when/if new instruction sets become more widely used. If so, the dividing line between mainstream and HEDT will get a lot wider than it is today.

Sandy Bridge alike + more instructions, lower TDP, faster graphics, DDR4 support, and etc, of course.

 

[ShintaiDK]

Sandy Bridge alike + more instructions, lower TDP, faster graphics, DDR4 support, and etc, of course.

Not going to happen. And its bad business.

The only way your product would materialize is if it was a future version of Atom. With a Core product sitting higher up.

 

[jpiniero]

The impression I had is Intel (and everyone else really) is going to have to move away from Silicon eventually if they are going to continue to get any kind of shrinkage. Intel could in theory continue to design products on the final silicon node, but it doesn't seem to make financial sense since so much of their business would greatly benefit from the huge drop in power draw.

Maybe this will be the kick in the butt needed to get Intel to develop a new architecture and instruction set to work around the issue.

 

[waltchan]

Not going to happen. And its bad business.

The only way your product would materialize is if it was a future version of Atom. With a Core product sitting higher up.

I think I'll be a little clearer here. Since die-shrink continues to get smaller and now is in tablet category, single-thread speed has decreased as well, while total benchmark may go up due to possibility making Pentium with 4 cores standard in future. The more cores you have, the higher the power consumption, so Intel is forced to lower all the clock speeds while putting turbo for all CPUs, while reducing TDP, thus, you have up to 2x lower single-thread speed than Pentium G3258 4.4 GHz overclocked it last peaked.

So, yes, single-thread speed will be reduced and slot between Sandy Bridge and Ivy Bridge in near future, while total benchmark may go up due to standard quad-cores. I stand behind what I said.

The more I see this, looks like the upcoming $5 used LGA1155 Sandy Bridge Pentiums on eBay are all what I need for rest of my life. I currently have 50 in-order for $5 Sandy Bridge Pentium G620 not fulfilled yet.

 

[frozentundra123456]

Where has single thread speed or ipc decreased? Single thread ipc continues to increase, all be it at a slower rate. Clockspeed has increased as well until haswell, and is stable with skylake, and intel has said kaby lake will bring performance increases. These off hand statements about what may happen at some distant point in the future are being blown way out of proportion.

 

[waltchan]

Where has single thread speed or ipc decreased? Single thread ipc continues to increase, all be it at a slower rate. Clockspeed has increased as well until haswell, and is stable with skylake, and intel has said kaby lake will bring performance increases. These off hand statements about what may happen at some distant point in the future are being blown way out of proportion.

Read the first post, "Intel plans to reduce performance..." Most likely, Intel is saying that future CPUs will have reduced single-thread speed (returning back to Sandy/Ivy Bridge level), while gaining more cores to increase benchmark scores. This is the trend heading next...

 

[ShintaiDK]

Where has single thread speed or ipc decreased? Single thread ipc continues to increase, all be it at a slower rate. Clockspeed has increased as well until haswell, and is stable with skylake, and intel has said kaby lake will bring performance increases. These off hand statements about what may happen at some distant point in the future are being blown way out of proportion.

+1 :thumbsup:

 

[TheELF]

Read the first post, "Intel plans to reduce performance..." Most likely, Intel is saying that future CPUs will have reduced single-thread speed (returning back to Sandy/Ivy Bridge level), while gaining more cores to increase benchmark scores. This is the trend heading next...

More cores draw more power no savings there.

Read the included quote of the OP

Now, Intel has acknowledged that the future of semiconductors may rely on technologies that reduce absolute performance in exchange for improved power consumption.

= Desktop CPUs in mobile/wearable devices ,maybe not the top notch then-gen™ CPU line performance but real desktop CPUs nevertheless,just lower TDP without loosing a lot of performance.
Intel already has the T line of CPUs that does just that.

 

[waltchan]

More cores draw more power no savings there.

All the future 4-cores Pentium will be clocked down a lot to match 35W standard TDP, and have base speed starting at 2.0 GHz with turbo at 3.0 GHz. 2.0 GHz base speed Cannonlake has the same single-thread speed estimated to 2.9 GHz G645 Sandy Bridge, or around 1500.

 

[AtenRa]

Seeing as how AMD was not able to charge a premium in the market for their faster iGPUs, but instead pricing seemed to be dictated in a large part by the ST and less so the MT performance of their CPU cores, then will Intel be force to lower prices for their newer CPUs if they have regressed in CPU performance, regardless of power efficiency or iGPU prowess. (Don't forget how poor Intel's drivers are.)

FP performance needs are getting higher and higher every year. If you take SandyBridge as the beginning, Intel is devoting more and more to iGPU IC space and performance than to CPU Cores every year since 2011.
In a Win 10, OpenCL and DX-12 era, iGPU performance especially in low 15/30W TDP environments like Laptops and even lower at 2-4W TDP Mobile is the primary focus the last 3-4 years and it will continue in the future.
I believe from 2017 things will start to become more clear with AMD ZEN APUs and Intel 10nm CannonLake Laptops.
The only segment that will still need more CPU performance but again at lower power is Servers. And that segment will start to need loads of FP performance very soon as well.

 

[poofyhairguy]

If intel manages to get it's mainstream CPU line,even if it's today's line in terms of performance, into smartphones by 2021 then it's pretty much good night to any other chip maker.

Uhh...no.

There is no way that phone makers will capitulate to Intel when when most of them (Huawei, Samsung Apple, etc.) make their own chips now. Hell that Asus Zenphone 2 has great single core performance and 4GB of RAM but it has to sell at half what flagships (with barely 3GB of RAM) cost for a reason- Intel is not considered a premium mobile brand for CPUs.

Intel has a branding problem with phones, consumers don't care if "Intel is inside." If anything Intel is associated with that bloated, malware ridden Windows PC that always "hated" the consumer and that the consumer was happy to trade for an easy to use iPad.

 

[iiiankiii]

I kinda agree. Unless there's another paradigm shift in software that will demand higher performance CPUs, the next best thing to do is lower power consumption so things can be made smaller and/or last longer.

We're at a point where the majority of the people out there can get away with doing most things on a mobile device. Those lower power/lower performance devices are fast enough for their needs. High performance devices are increasingly becoming a niche.

It's pretty clear the priority has shifted away from absolute performance towards lower, more efficient power consumption.

 

[frozentundra123456]

Read the first post, "Intel plans to reduce performance..." Most likely, Intel is saying that future CPUs will have reduced single-thread speed (returning back to Sandy/Ivy Bridge level), while gaining more cores to increase benchmark scores. This is the trend heading next...

Read your own post. You said specifically single thread performance has decreased. It has not. Yes, there are lower power, lower performance chips available, but performance has not decreased on top end chips. In fact, it continues to increase.

 

[Arachnotronic]

Uhh...no.

There is no way that phone makers will capitulate to Intel when when most of them (Huawei, Samsung Apple, etc.) make their own chips now. Hell that Asus Zenphone 2 has great single core performance and 4GB of RAM but it has to sell at half what flagships (with barely 3GB of RAM) cost for a reason- Intel is not considered a premium mobile brand for CPUs.

Intel has a branding problem with phones, consumers don't care if "Intel is inside." If anything Intel is associated with that bloated, malware ridden Windows PC that always "hated" the consumer and that the consumer was happy to trade for an easy to use iPad.

Intel's problem in phones is that its chips are complete and total garbage.

 

[superstition]

Unless there's another paradigm shift in software that will demand higher performance CPUs

VR

If Intel doesn't go to arsenide transistors with terahertz speeds then China will, eventually.

Any "more focus on just low power applications" is going to be temporary.

You don't power a Holodeck with Sandy Bridge.

 

[krumme]

Qcom new midrange 625 is made on 14nm using a53 cores.
The cpu core part footprint of the soc is next to zero.

End 2016 we will see the first devices using a35. Its intended for wearables and the move from 28nm a7 to 14nm a35 will bring a huge leap in efficiency and aprox 90% a53 perf levels at same freq.

My guess is we will see it not only in wearables and in big little configurations but also as stand alone cpu all the way to midrange.

The basic presumtion here is the performance of octocore a53 aparently is okey as it is and people want even better battery life. A35 14nm will bring that in spades.

Its happening all over. Mobile cloud...
When Intel is changing focus its because there is a good reason. And its probably easy to tell by the numbers already.

 

[IntelUser2000]

Read your own post. You said specifically single thread performance has decreased. It has not. Yes, there are lower power, lower performance chips available, but performance has not decreased on top end chips. In fact, it continues to increase.

It may well be.

Why do you think otherwise when transistor performance is one of the major factors of CPU performance? Intel's "performance-focus" policy with process changed starting in 22nm and immediately we got sluggish performance improvements. Now they are saying that they'll focus more on low power but that they will *reduce* it.

And some are saying "we needed to have post-silicon yesterday". And yes, their response is that it'll be *slower*.

What options do you have as Intel for the distant future?

1. Stick with the ever-diminishing PC sales, and ever-diminishing performance gains and ever-increasing difficulty of manufacturing them
2. Follow the natural, industry trend of lower power chips so hopefully they can find success in an area outside of PCs.

#1 being rendered irrevalent is inevitable, rendering in a very slow painful death. #2, if you die you'll die much faster but there's a chance for turning around the company if they do things right.

VR

If Intel doesn't go to arsenide transistors with terahertz speeds then China will, eventually.

Any "more focus on just low power applications" is going to be temporary.

You don't power a Holodeck with Sandy Bridge.

Haven't you thought of maybe, just maybe there's a technological limit of how fast they can go? If VR does take off, they'll probably find out using GPU to offload more stuff was a much better idea anyway.

Holodeck is a fantasy!

 

[crashtech]

Don't forget server and HPC, small as a percentage but lucrative. I don't see Intel completely abandoning a market in which they are without peer, even though it's obvious the emphasis has to shift to low power devices.

 

[jpiniero]

Don't forget server and HPC, small as a percentage but lucrative. I don't see Intel completely abandoning a market in which they are without peer, even though it's obvious the emphasis has to shift to low power devices.

It's not low power, it's lower clock speed. And those markets would be fine with lower clock speed if it meant much better density and far better perf/w.

I should point out that we don't really know how low of clock speeds we are talking about.