News Intel Announces Delay to 7nm Processors, Now One Year Behind Expectations

[bit_user]

Well, if Intel is hurting in any way, shape or form, it isn't showing up in their financial statements' record sales, record profits and many SKUs being perpetually sold-out. Most of the market share Intel is losing is simply from being unable to meet demand.

A rising tide lifts all boats.

This success comes in spite of their failures, but that growth isn't something they can really take credit for, either, as it didn't come as a consequence of corresponding improvements in their products.

 

[techconc]

Intel and AMD are designing cores to clock higher, which naturally makes them less efficient than they could be, at lower clock speeds. Meanwhile, Apple targets lower clockspeeds, but that means their chips won't clock very high. The same has been true for ARM's own cores.

Targeting high clock speeds is exactly what got Intel into trouble with their Netburst architecture in the Pentium 4 series. Designing for efficiency leads to performance. This is especially true in mobile products where thermal considerations can limit real world performance. This is exactly why Apple Silicon will lead to industry leading overall performance. This is what also allows the technology to scale so well.

You’re not the only person to be skeptical about what’s happening here. We’ve all been conditioned to think of ARM as being great for low end devices but not suitable for desktops, etc. Yet, there reality is, Apple is at a point where they meet or exceed Intel’s core performance at a fraction of the power. Part of that is due to being on a much better manufacturing process but a big part of that is due to superior design. Apple has a world class team designing their silicon. You can debate this as you wish, but we’ll see the initial ”proof” with Mac based silicon this fall.

 

[JamesSneed]

but AMD is not using EUV yet. Why?

This one is easy. TSMC's first goal on a new node is low power. Then they take about a year to tweak the process for high power chips. Low power designs will make things like DRAM, cell phone APU's, etc. The high power designs will make desktop/server CPU's and GPU's. So 5nm EUV is being used to make Apple iPhone APU's now. Expect AMD's next chip to use 5nm and that next chip may not be Zen4 it could be APU related.

 

[Kamen Rider Blade]

That's mostly about ISA, and I'm with you on that. It remains to be seen about A14, and we don't yet know how well Apple's cores can compete outside of a power- or thermally- constrained application, like phones, tablets, and notebooks.

Out of the RISC family of ISA's

I think OpenPOWER has been designed to target High End Performance.

ARM was always designed to be "Ultra Low Power".

https://www.phoronix.com/scan.php?page=article&item=rome-power9-arm&num=4

The POWER 9 is the closest to the x86 Duopoly.

ARM hasn't improved nearly as much as POWER when it comes to High Performance Computing.

 

[bit_user]

The slow-initially-optimize-later approach is fundamentally flawed. People aren't benchmark software. We don't average over our experiences. We remember the salient and neglect the unexceptional. If we're annoyed by slow start-up of a program, we aren't going to be somehow unannoyed by decent performance in the hours following.

Your timescale is completely off. Optimization can occur very granularly - down to the level of a basic block - and frequently, if necessary. The results can also be cached on disk, so that a set of optimized codepaths are ready for next use.

If you've used Chrome since 2017, you've already experienced this technique:

the compiled code is additionally optimized (and re-optimized) dynamically at runtime, based on heuristics of the code's execution profile.

https://en.wikipedia.org/wiki/V8_(JavaScript_engine)#Design
​

 

[bit_user]

Targeting high clock speeds is exactly what got Intel into trouble with their Netburst architecture in the Pentium 4 series.

That's prioritizing clock speed, only.

Designing for efficiency leads to performance.

If you're talking about mobile, GPUs, or perhaps massively multi-core server CPUs.

What I'm talking about is that AMD & Intel are prioritizing single-thread perf. Not above all else (at least, AMD seems to be trying to strike a balance with efficiency), but Apple is much more focused on prioritizing perf/W.

You can't get the best single-thread perf at mobile clock speeds, so Intel & AMD design their cores to clock higher. This forces them to design to a shorter critical path, which forces compromises on IPC.

Meanwhile, Apple doesn't need its cores to clock high, so they probably use a longer critical path, meaning their cores couldn't clock high, even if they wanted to.

So, they might be on par - or even slightly ahead, on IPC - but that won't translate directly to single-thread perf, if they can't approach comparable clock speeds.

You’re not the only person to be skeptical about what’s happening here. We’ve all been conditioned to think of ARM as being great for low end devices but not suitable for desktops, etc.

No, what I'm skeptical of is your claim that "Intel was never more than a mediocre chip designer".

I don't doubt ARM's threat to server, laptop, and potentially even the desktop markets, and I don't doubt Apple's competence.

Apple is at a point where they meet or exceed Intel’s core performance at a fraction of the power.

It's early to declare Apple a victor. They have work yet to do, to demonstrate they can really compete outside the highly power- and thermally- constrained applications.

You can debate this as you wish, but we’ll see the initial ”proof” with Mac based silicon this fall.

The problem with trying to debate me on this is that I've debated on the other side, multiple times. I'm just not as far along as you, in that Apple has proven they can do low-power performance better than anyone, but that's distinctly different than taking the absolute performance crown. I'm not saying they can't or won't, just that they haven't.

 

[bit_user]

ARM was always designed to be "Ultra Low Power".

https://www.phoronix.com/scan.php?page=article&item=rome-power9-arm&num=4

The POWER 9 is the closest to the x86 Duopoly.

That's ancient. Both the Thunder X and Ampere eMag he reviewed were old, even by the time that was written.

Also, you're confusing ARM core implementations with the ARM ISA. There's nothing about the ARMv8-A ISA that limits it to low-power applications. Indeed, cores implementing it were exclusively used to form the current #1 supercomputer:

https://www.tomshardware.com/news/japanese-arm-based-supercomputer-fugaku-is-now-world-most-powerful​

ARM hasn't improved nearly as much as POWER when it comes to High Performance Computing.

And just how far down the Top 500 do you need to go, in order to find a POWER-based machine that doesn't rely on GPUs for the main compute muscle?

 

[techconc]

What I'm talking about is that AMD & Intel are prioritizing single-thread perf. Not above all else (at least, AMD seems to be trying to strike a balance with efficiency), but Apple is much more focused on prioritizing perf/W.

Apple's existing A13 is within 6% of Intel's top single core performance. They're there already.... and at a fraction of the power.

You can't get the best single-thread perf at mobile clock speeds, so Intel & AMD design their cores to clock higher. This forces them to design to a shorter critical path, which forces compromises on IPC.

A13 says otherwise. Given a wide enough architecture, clock speed is less of a factor.

Meanwhile, Apple doesn't need its cores to clock high, so they probably use a longer critical path, meaning their cores couldn't clock high, even if they wanted to.

Speculation on your part. Given a higher power envelope suitable for a desktop, I'd disagree.

It's early to declare Apple a victor. They have work yet to do, to demonstrate they can really compete outside the highly power- and thermally- constrained applications.

The problem with trying to debate me on this is that I've debated on the other side, multiple times. I'm just not as far along as you, in that Apple has proven they can do low-power performance better than anyone, but that's distinctly different than taking the absolute performance crown. I'm not saying they can't or won't, just that they haven't.

Overall, I'm not declaring anyone a victor just yet. I just don't think there is as much to prove you that you seem to think. The A13 is very close in single core performance to the best single core performance in an i9 today. In September, we'll see the A14 which will at least match and very like exceed the best Intel has to offer at single core performance. In October, we'll get the first latest of Mac specific Apple Silicon. I should also note that CPU performance is just one measurement of Apple Silicon. Apple adds a ton of other things into their silicon such as ISP, Neural Engine, machine learning accelerators, secure enclave, dedicated video encoding / decoding, dedicated encryption/decryption logic, etc, etc. Most of which, they wouldn't get with an Intel chip.

 

[jeremyj_83]

Apple's existing A13 is within 6% of Intel's top single core performance. They're there already.... and at a fraction of the power.

Where do you get that information? If you are looking at things like Geekbench 4 on an iPad and comparing that to an Intel CPU on Win 10 that is an apples to oranges comparison.

 

[TJ Hooker]

I should also note that CPU performance is just one measurement of Apple Silicon. Apple adds a ton of other things into their silicon such as ISP, Neural Engine, machine learning accelerators, secure enclave, dedicated video encoding / decoding, dedicated encryption/decryption logic, etc, etc. Most of which, they wouldn't get with an Intel chip.

Neural engine and machine learning accelerator are the same thing. I think the ISP may also be part of the neural engine block in the most recent Apple SoCs, although an ISP is kind of irrelevant for a PC anyway because your desktop isn't packaged with a high resolution camera. Every modern PC has hardware video encode/decode, whether from iGPU (i.e. any non -F Intel chip) or dGPU. All modern x86 desktop CPUs have crypto acceleration (e.g. AES instruction set). All modern PCs have an implementation of a trusted execution environment (similar to secure enclave).

So the only thing a regular PC is missing out on compared to a typical mobile SoC is a DSP. And I think you could always use your GPU to do whatever you'd use the DSP for (possibly faster, if you have a decent discrete GPU), it just would be a lot less efficient.

 

[bit_user]

Apple's existing A13 is within 6% of Intel's top single core performance. They're there already.... and at a fraction of the power.

A13 says otherwise. Given a wide enough architecture, clock speed is less of a factor.

Citation needed.

Again, think about what you're saying. Are you really claiming that the A13 is within 6% of something like the Comet Lake i9-10900KF? And on a diverse workload, or just one cherry-picked benchmark?

Speculation on your part. Given a higher power envelope suitable for a desktop, I'd disagree.

Yes, it's indeed speculation. I dare to speculate that Apple's not dumb enough to waste critical path on a core they know they're never going to clock above 3.0 GHz.

I should also note that CPU performance is just one measurement of Apple Silicon. Apple adds a ton of other things into their silicon such as ISP, Neural Engine, machine learning accelerators, secure enclave, dedicated video encoding / decoding, dedicated encryption/decryption logic, etc, etc. Most of which, they wouldn't get with an Intel chip.

I don't really want to go to bat for Intel, but they do have things like VNNI extensions and BFloat16 in the pipeline. Their laptop chips also have an image processing hardware block. Their iGPUs have had video encoding & decoding acceleration since Sandybridge (released 2011), and they have instruction set extensions like AES-NI for accelerating encryption and decryption. So, I wouldn't say they're exactly showing up to the fight empty-handed.

That said, I'm as critical as anyone of their decision to use AVX-512 for deep learning. It's definitely not perf/W competitive with GPUs or DSP-based deep learning blocks, like what Apple uses. Speaking of GPUs, I believe their Gen12 iGPUs will pack BFloat16 & int8 support.

 

[bit_user]

I think you could always use your GPU to do whatever you'd use the DSP for (possibly faster, if you have a decent discrete GPU),

IMO, GPUs are what DSPs want to be, when they grow up.

 

[Kamen Rider Blade]

IMO, GPUs are what DSPs want to be, when they grow up.

And here I thought FPGA's are what DSPs want to be, when they grow up.

 

[Kamen Rider Blade]

Also, you're confusing ARM core implementations with the ARM ISA. There's nothing about the ARMv8-A ISA that limits it to low-power applications. Indeed, cores implementing it were exclusively used to form the current #1 supercomputer:

7.2 million 2.0 -2.2 GHz cores.

That sounds like they didn't want to clock the Cores higher for Power Efficiency reasons.

 

[bit_user]

And here I thought FPGA's are what DSPs want to be, when they grow up.

FPGAs can make sense, depending on what you're doing.

However, I meant that if you want a fully-programmable compute engine that prioritizes throughput and power-efficiency, GPUs are really the next step up from DSPs.

 

[bit_user]

7.2 million 2.0 -2.2 GHz cores.

That sounds like they didn't want to clock the Cores higher for Power Efficiency reasons.

No more or less-so than GPUs.

 

[Arbie]

Wow, that was some BS response from Swan (Intel CEO). I guess that was expected from an MBA instead of an engineer.

Wow, that was some BS comment.

Brian Krzanich - an engineer - was CEO from 2013-2018, overseeing five years of non-progress and BTW misleading everyone about the true story. Bob Swan was dumped into the CEO job two years ago, and must have relied heavily on his engineers (e.g. Renduchintala) to bring Intel's silicon back on the rails. The major long-term and continuing failures are theirs. Swan just gets the knee-jerk blame.

 

[techconc]

Where do you get that information? If you are looking at things like Geekbench 4 on an iPad and comparing that to an Intel CPU on Win 10 that is an apples to oranges comparison.

That specific score is based on Geekbench 5 using the highest recorded score which happened to be in the Core i9.
https://browser.geekbench.com/processors/intel-core-i9-10900k

Anandtech has also showed similar results using SPEC.
https://www.anandtech.com/show/14892/the-apple-iphone-11-pro-and-max-review/4

“Overall, in terms of performance, the A13 and the Lightning cores are extremely fast. In the mobile space, there’s really no competition as the A13 posts almost double the performance of the next best non-Apple SoC. The difference is a little bit less in the floating-point suite, but again we’re not expecting any proper competition for at least another 2-3 years, and Apple isn’t standing still either.
Last year I’ve noted that the A12 was margins off the best desktop CPU cores. This year, the A13 has essentially matched best that AMD and Intel have to offer – in SPECint2006 at least. In SPECfp2006 the A13 is still roughly 15% behind.”

 

[techconc]

Neural engine and machine learning accelerator are the same thing.

Generically, I would think so to, but that’s not the case. Apple has included some form of AMX block that is in addition to the neural engine. This is a superset of the ARM ISA.
https://www.anandtech.com/show/14892/the-apple-iphone-11-pro-and-max-review/2

I think the ISP may also be part of the neural engine block in the most recent Apple SoCs,

Nope.

although an ISP is kind of irrelevant for a PC anyway because your desktop isn't packaged with a high resolution camera.

Nope. Resolution of the camera is irrelevant. The point here is that you’re offloading the CPU with a highly optimized and efficient function in hardware.

Every modern PC has hardware video encode/decode, whether from iGPU (i.e. any non -F Intel chip) or dGPU. All modern x86 desktop CPUs have crypto acceleration (e.g. AES instruction set). All modern PCs have an implementation of a trusted execution environment (similar to secure enclave).

In some form perhaps, but not necessarily to the level needed. As an example, Apple had to include their T2 chip to get this functionality up to their standards. They would not waste money on such things if they didn’t have to.

So the only thing a regular PC is missing out on compared to a typical mobile SoC is a DSP. And I think you could always use your GPU to do whatever you'd use the DSP for (possibly faster, if you have a decent discrete GPU), it just would be a lot less efficient.

Well, that’s just it. All of this ”can” be done in the CPU as well. It’s just not as efficient that way. The point here is that SoCs really are diverging from standard CPUs by including more and more specialty functions. These specialty functions are what’s needed to bring “features” to devices. For example, why do you think Apple hasn’t been able to bring FaceID to Macs yet? With the move to custom Silicon, Apple will be able to bring new features to Macs because they’ll control the hardware in the SoC and optimize for those specific features. When you buy chips from Intel / AMD, you’re getting a more generic product that is not going to be tailored to your specific needs.

 

[jeremyj_83]

in SPECint2006 at least.

That right there is the key statement. SPECint2006 is a 14 year old benchmark. On top of that the A13 is running on a mobile OS that isn't designed for multitasking. What happens when it is thrown onto a heavy OS like Windows? What happens when you are trying to push the CPUs to say have 5 browser tabs open, Excel, Outlook, and music playing? My guess is the A13 and its OS will go on tilt and perform like an old Netbook. Reason is they aren't designed to do thing like that.

 

[bit_user]

Bob Swan was dumped into the CEO job two years ago, and must have relied heavily on his engineers (e.g. Renduchintala) to bring Intel's silicon back on the rails. The major long-term and continuing failures are theirs. Swan just gets the knee-jerk blame.

Two years should be far longer than needed to sort out what's going on and hold people accountable ...for anyone who's actually suited for such a role, at least. Usually, 6 months is more than enough to figure out what's wrong with an organization and start implementing out corrective actions. Two years is longer than the average Fortune 500 CEO's tenure.

I don't understand how you can believe he deserves the power and the compensation, without taking on the responsibility. The buck stops at the top.

 

[bit_user]

That specific score is based on Geekbench 5 using the highest recorded score which happened to be in the Core i9.
https://browser.geekbench.com/processors/intel-core-i9-10900k

I have to agree that it's questionable how comparable Geekbench scores really are across different OS and platforms.

In terms of efficiency, quite a bit of that page has some alarming statements you glossed over:

... the power and efficiency figures on the other hand are extremely unexpected. In virtually all of the SPECint2006 tests, Apple has gone and increased the peak power draw of the A13 SoC; and so in many cases we’re almost 1W above the A12. Here at peak performance it seems the power increase was greater than the performance increase, and that’s why in almost all workloads the A13 ends up as less efficient than the A12.

...
The total power use is quite alarming here, as we’re exceeding 5W for many workloads. In 470.lbm the chip went even higher, averaging 6.27W. If I had not been actively cooling the phone and purposefully attempting it not to throttle, it would be impossible for the chip to maintain this performance for prolonged periods.

Here we saw a few workloads that were more kind in terms of efficiency, so while power consumption is still notably increased, it’s more linear with performance. However in others, we’re still seeing an efficiency regression.

In terms of power and efficiency, the A13 seemingly wasn’t a very successful iteration for Apple, at least when it comes to the efficiency at the chip’s peak performance state. The higher power draw should mean that the SoC and phone will be more prone to throttling and sensitive to temperatures.

This suggests that Apple still has some work cut out for it, in order to further scale up performance.

What's weird is that this whole tangent started with you claiming "Intel was never more than a mediocre chip designer", yet you haven't shown a like-for-like comparison to back this up - only Apples and oranges (or ARM and x86, in this case).

ARMv8-A is a fundamentally more efficient ISA. That's what gives it an advantage in mobile and server applications. So, it's hardly surprising that a competent implementation is going to be challenging x86. That really says more about x86 than about Intel's abilities as a chip designer. If anything, it's amazing Intel & AMD managed to wring so much performance out of their chips, in spite of the ISA!

 

[bit_user]

why do you think Apple hasn’t been able to bring FaceID to Macs yet?

For the lack of cameras with sufficient capabilities.

You're stretching to make your points, instead of logically thinking about how much compute their SoCs have vs. a desktop PC. If you include their GPUs, even Apple's lowest-end iMac or mac mini has more than enough compute for Face ID.

What's missing is the camera needed to cope with challenging lighting conditions and to defeat exploits involving static photos. You can't properly do all of that stuff with a generic webcam.

When you buy chips from Intel / AMD, you’re getting a more generic product that is not going to be tailored to your specific needs.

That's not even true. Intel and AMD both make custom models of their CPUs and GPUs specifically for Apple.

Rumor has it, AVX even had its roots in Steve Jobs' talks with Intel to transition Mac to it. Apple reportedly wanted something that was more like AltiVec, and the result was AVX.

 

[bit_user]

On top of that the A13 is running on a mobile OS that isn't designed for multitasking.

I believe iOS and Mac OS now use the same kernel, FWIW. With mobile CPUs having multiple cores for more than a decade, I don't believe iOS is deficient at multitasking.

 

[techconc]

I have to agree that it's questionable how comparable Geekbench scores really are across different OS and platforms.

What you agree to is irrelevant. For starters, I offered both Geekbench and SPEC. If you want to go down the path of explaining why these are not valid cross platform benchmarks, then you are welcome to attempt to make that case. Moreover, unless you can provide something which directly contradicts this comparison, you have no basis for your disagreement other than for the sake of being argumentative.

In terms of efficiency, quite a bit of that page has some alarming statements you glossed over:

This suggests that Apple still has some work cut out for it, in order to further scale up performance.

You make blanket statements which you are apparently not able to support.

What's weird is that this whole tangent started with you claiming "Intel was never more than a mediocre chip designer", yet you haven't shown a like-for-like comparison to back this up - only Apples and oranges (or ARM and x86, in this case).

My original claim is made evident by such examples of 680x0 vs 80x86 comparisons over multiple generations. The same goes with PowerPC, etc. Intel has never had the superior architecture or chip design. Rather, Intel has always had the advantage on chip manufacturing process.

ARMv8-A is a fundamentally more efficient ISA. That's what gives it an advantage in mobile and server applications. So, it's hardly surprising that a competent implementation is going to be challenging x86. That really says more about x86 than about Intel's abilities as a chip designer. If anything, it's amazing Intel & AMD managed to wring so much performance out of their chips, in spite of the ISA!

Yes, Kudos to Intel to drag x86 so far forward. Yet, at the end of the day, it’s not the most efficient design.. by a considerable margin. Worse, it’s losing it’s advantage in terms of performance as well.