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

[InvalidError]

If they don't come out with something akin to the Nehalem in the Athlon 64 age pretty soon the hole they have dug will be their grave.

Intel hasn't been "doing nothing" for years, its CPU architecture people are two generations ahead of what Intel is currently able to manufacture and with the 7nm delays, it looks like Intel will be held back by process tech for a while longer.

Whenever they sort out their 7nm issue, they should be ready to roll out CPUs based on Golden Cove and likely make up whatever ground they may have lost to Zen 4.

 

[Kamen Rider Blade]

Okay, Dr. Frankenstein.

Or maybe you're working on a Flux Capacitor, like Doc Brown.

My Bad, got the wrong founder.

We want the one that is still alive.

Gordon Moore.

 

[Kamen Rider Blade]

Whenever they sort out their 7nm issue, they should be ready to roll out CPUs based on Golden Cove and likely make up whatever ground they may have lost to Zen 4.

Hopefully, they'll have patched all the security vulnerabilities by then.

 

[Kamen Rider Blade]

Stomping? LOL, more like eeking out a narrow victory at useless 1080P gaming. Like I am going to spend 1k on a video card to game at 1080P. Give it a rest. As if anyone can tell the difference between 240 FPS and 210 FPS, lmao.

Just to back up your statements.

The moment you hit 100 Hz/FPS <- Same Difference.
_80-100 FPS, your (Frame-Time/Image Persistence) drive +20 FPS = 2.5 ms Frame Time improvement
100-125 FPS, your (Frame-Time/Image Persistence) drive +25 FPS = 2.0 ms Frame Time improvement
...
200-250 FPS, your (Frame-Time/Image Persistence) drive +50 FPS = 1.0 ms Frame Time improvement

Your rate of improvement in (Frame-Time/Image Persistence) dramatically dwindles the moment you shift from 100 -125 FPS and beyond. The difficulty in delivering that many FPS also starts getting exponential.

You need a monster of a GPU to push modern games up that high.

All that resource is better spent on Wider Screen and Wider FoV to gather more useful information in FPS and other games.
Going (16:9 -> 21:9 -> 24:9) is more worth while use of the same resource past 100-125 FPS.

How far you want to go, is up to you, but just remember the laws of diminishing returns.

 

[Kamen Rider Blade]

Until recently, ARM hasn't been making performance-oriented cores. So, even the existing server chips don't show the ISA in its best light. However, things aren't looking so good for x86:

RISC-V is the future of RISC IMO

PowerPC/ARM is the NOW of RISC.

If Nvidia insists on buying ARM, I'd honestly rather have IBM buy out ARM and combine their RISC R&D teams along with incorporating and working with Si-Five on RISC-V.

Pushing RISC to the future on Performance & Power Efficiency.

 

[bit_user]

My Bad, got the wrong founder.

We want the one that is still alive.

Gordon Moore.

He's 91 years old. And while there are some very astute 91-year-olds, most are well past their prime.

I don't get the deal with this whole cult of personality around tech. Maybe it's the Steve Jobs-effect, but there are people who can design CPUs besides Jim Keller and design GPUs besides Raja Koduri. Those names just get tossed around because people know them.

There are lots of capable people in the industry, today. The limiting factors are usually the boards of these companies and the culture of their management hierarchy. Corporate boards and management are full of people flocking from one management fad to the next and overly-focused on short-term results. Google has actually made a study of how managers affect performance. And Amazon is famous for their disregard of short-term results.

 

[PCWarrior]

Ice Lake-based laptops have been shipping for nearly a year.

That’s why I used the word meaningfully. We still don’t have Icelake desktop (either mainstream or HEDT) cpus neither server cpus. And we still don’t have performance H-series mobile cpus with a 45W tdp and we still don’t have cpus with more than 4 cores on the 15/28W segment even though we have 14nm ones.

That seems a little hard to believe. So, how many instructions does that have them retiring per cycle?

There is already an 18% IPC improvement on average (keyword on average) of Sunny Cove over Skylake. Willow Cove is expected to have at least another 10-12% improvement over Sunny Cove and Golden Cove to have a further 15-17% over Willow Cove. If you do the math it is 1.18x1.11*1.16=1.52x i.e. 52% improvement. We are talking about 5+ years of cumulative architectural innovation. Not dissimilar to what AMD did when they launched Ryzen in 2017 after also a long period of not launching any meaningful new microarchitecture since Piledriver/Excavator. They also achieved huge IPC gains – incidentally also 52%.

Why do you believe that AMD (a company that at the time was on the brink of bankruptcy with miniscule R&D budget) was able to pull off such a feat but not Intel? Intel’s problems are not related to architecture but to the fact that they are unable to manufacture their new architecture designs due to the unprecedented delays in transitioning to a smaller/denser manufacturing node. Do you think that AMD would be able to bring Zen, Zen2 and soon Zen 3 if they were still stack on 28nm?

 

[bit_user]

RISC-V is the future of RISC IMO

I'm no CPU designer, but the analysis I've read of RISC-V suggests it lacks any killer features that would really put it ahead of ARMv8-A.

PowerPC/ARM is the NOW of RISC.

How do you even fit both in the same sentence? PowerPC is 27 years old, and last used in game consoles that stopped selling the better part of a decade ago, not long after ARMv8-A was first announced! Sure, there are some embedded applications, and POWER tried to make a server push that seems to have fizzled, but PowerPC is basically dead and POWER is dying.

And that's a point worth emphasizing: ARM is a company, not an ISA! ARMv8-A is an ISA, and rumor have been circulating about a v9. So far, most AArch64 CPUs include support for AArch32, but it needn't be so (and I think we've seen a couple HPC chips that don't).

 

[Chung Leong]

Yeah Apple is always right on the edge of technology.
I still can't believe those amazing mac pro wheels they launched some months ago (only U$699... and they give you 4 of them!!!! My God!!!!!!), I mean WHEELS!, Can you believe it?!

These wheels are insanely great! Seven hundred dollars are a small price to pay for freedom.

 

[bit_user]

That’s why I used the word meaningfully. We still don’t have Icelake desktop (either mainstream or HEDT) cpus neither server cpus. And we still don’t have performance H-series mobile cpus with a 45W tdp and we still don’t have cpus with more than 4 cores on the 15/28W segment even though we have 14nm ones.

I get what you're saying, but I still take issue with the word "meaningfully". I think your point is that 10 nm isn't in full production.

There is already an 18% IPC improvement on average (keyword on average) of Sunny Cove over Skylake. Willow Cove is expected to have at least another 10-12% improvement over Sunny Cove and Golden Cove to have a further 15-17% over Willow Cove. If you do the math it is 1.18x1.11*1.16=1.52x i.e. 52% improvement. We are talking about 5+ years of cumulative architectural innovation.

Yeah, I'm still skeptical. Perhaps there's some carefully-chosen set of test which can show such improvement, especially if you're counting AVX vector lane operations as individual instructions, so that AVX-512 works out to be a huge win.

Not dissimilar to what AMD did when they launched Ryzen in 2017

It is dissimilar, because AMD was starting from a lower base. That makes it easy to achieve big improvements.

Why do you believe that AMD (a company that at the time was on the brink of bankruptcy with miniscule R&D budget) was able to pull off such a feat

Because Bulldozer was bad, and I think they made back-end improvements as well. I heard that Bulldozer was built on an ASIC flow, I think to make it easier to integrate their GPUs to form APUs.

but not Intel?

Because an infinitely complex, infinitely hot CPU cannot have infinite IPC. We're past the point of diminishing returns.

 

[bit_user]

These wheels are insanely great! Seven hundred dollars are a small price to pay for freedom.

I thought I read they can't even lock!

Is that true? If so, it means people with floors that aren't flat might have to ruin the aesthetic by putting them in those ugly furniture cups.

Anyway, as bad as $700 for a wheels kit is, if you got a Mac Pro with the wheels pre-installed, you have to pay $300 if you later decide you'd rather have the feet!

https://www.apple.com/shop/product/MXNM2ZM/A/apple-mac-pro-feet-kit​

 

[PCWarrior]

I get what you're saying, but I still take issue with the word "meaningfully". I think your point is that 10 nm isn't in full production.

Intel hasn’t even paper-launched 90% of their 10nm cpu portfolio. They had no problem paper-launching (or producing in very low volume) other 14nm cpus (e.g. the 56-core MCM xeons) or 10nm cpus (remember Canonlake) in the recent past. As far as I am concerned ‘not in-full production’ would mean that Intel had launched pretty much their full portfolio of cpus but the production was limited/low volume. The context that this was said was that people haven't really seen Intel's post-Skylake architectures in action - the single instance they did it was hard to test for ipc as it was in an ultrabook/laptop format with no control over frequency (core or memory) in order to compare apples to apples.

Because an infinitely complex, infinitely hot CPU cannot have infinite IPC. We're past the point of diminishing returns.

I don’t think this point has been reached yet and neither it seems the likes of cpu architects like Jim Keller. Start watching his presentation from 27:05.

View: https://www.youtube.com/watch?v=oIG9ztQw2Gc&feature=emb_title

 

[Kamen Rider Blade]

How do you even fit both in the same sentence? PowerPC is 27 years old, and last used in game consoles that stopped selling the better part of a decade ago, not long after ARMv8-A was first announced! Sure, there are some embedded applications, and POWER tried to make a server push that seems to have fizzled, but PowerPC is basically dead and POWER is dying.

Sorry, confused PowerPC for OpenPower.

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

POWER needs to have significant performance wins, it's not there yet, but getting close with Power 9
I'm looking forward to see how Power 10 compared to the best that XEON / EPYC can deliver.
I'm glad IBM is pushing on the OpenPOWER side.

Somebody from the RISC camp's gotta deliver some threatening performance to the CISC Duopoly.

The ARM chips aren't delivering enough performance to be truly threatening on the high end yet.

 

[Kamen Rider Blade]

I'm no CPU designer, but the analysis I've read of RISC-V suggests it lacks any killer features that would really put it ahead of ARMv8-A.

Give it time.

 

[bit_user]

Intel hasn’t even paper-launched 90% of their 10nm cpu portfolio.

I'm unconvinced, but also unwilling to waste more time debating the semantics of "meaningful". Like I said, I see your point, which I think is that they have yet to deliver the volumes and yields that would show 10 nm is mature.

I don’t think this point has been reached yet and neither it seems the likes of cpu architects like Jim Keller. Start watching his presentation from 27:05.

View: https://www.youtube.com/watch?v=oIG9ztQw2Gc&feature=emb_title

Interesting, but mostly beside the point. I clicked through the slides but noted only how enormous Sunny Cove had to be, to net a mere 3.34 IPC. Also, interesting to see how he had to compare it with Athlon64. There were quite a lot of really old architectures in his comparisons.

I'm still reserving judgement on future IPC until it's delivered, and would note that Jim already left Intel.

 

[bit_user]

POWER needs to have significant performance wins, it's not there yet, but getting close with Power 9

Power 9 will be hampered by a bad process node, sadly.

I think OpenPOWER came too late. If they'd done it a few years earlier, they could've diverted some of the energy that's now going into RISC-V designs.

The ARM chips aren't delivering enough performance to be truly threatening on the high end yet.

I can't make you see it, but I already posted why I think that's wrong, in #111.

Give it time.

I'm not saying RISC-V won't happen, just that it's not inherently better than ARMv8-A. We also don't know what v9 will look like, if that's a real thing - could be a game changer.

 

[InvalidError]

I'm still reserving judgement on future IPC until it's delivered, and would note that Jim already left Intel.

Single-threaded IPC is running against an increasingly steep brick wall - you can predict and speculatively execute only so many instructions ahead before tripping over a misprediction or external dependency. Per-core IPC should still have a fair amount of room to grow, albeit at the expense of single-threaded IPC by moving resources from brute-forcing a single thread as far as it can possibly go to more efficiently and effectively feeding multiple concurrent threads.

I'm not saying RISC-V won't happen, just that it's not inherently better than ARMv8-A. We also don't know what v9 will look like, if that's a real thing - could be a game changer.

RISC-V's biggest benefit is that it is open-source, so no worries about Softbank/ARM deciding to pull the rug from under your feet. I can imagine companies having token RISC-V projects simply to keep ARM's greed in check - screw us over with unreasonable licensing fees or withholding stuff and we'll move RISC-V up our product stack.

 

[Ncogneto]

You may not but there are competitive gamers out there seeking 240+Hz monitors.

FPS and refresh rate are two separate things entirely.

 

[NightHawkRMX]

FPS and frame rate are two separate things entirely.

What?

 

[Ncogneto]

Agreed. The attack was unnecessary to make the point.

FWIW, I don't believe the $700 price tag was a publicity stunt, either. It's just typical Apple profit-maximizing behavior. Anyone seriously in the market for a Mac Pro already knows it, and Apple already got lots of free publicity for it, when they announced it.

The way to think about the $700 wheels is more like a $10k upholstery option, in a high-end luxury car. The manufacturer knows you can probably afford it, since you're already buying the car. They just don't want to miss any chance to make a little more money off you.

View: https://www.youtube.com/watch?v=6TfLVL5GeE4

 

[Kamen Rider Blade]

I'm not saying RISC-V won't happen, just that it's not inherently better than ARMv8-A. We also don't know what v9 will look like, if that's a real thing - could be a game changer.

As fans of RISC, aren't you happy to see innovation?

Be it RISC-V or ARM v9 or the Power ISA v3.1?

 

[Ncogneto]

What?

Sorry, mis-typed, fixed now.

 

[Kamen Rider Blade]

What?

I think he's trying to talk about FPS & the difference between "Frame-Time / Image Persistence"
I have a entire table to illustrate my point in Post #129

 

[bit_user]

As fans of RISC, aren't you happy to see innovation?

I'm a fan of innovation, period.

But I've read there's just nothing very innovative about RISC-V. And that's okay. Just keep your expectations in check. There will presumably be a RISC-VI. So, it's not necessary for RISC-V to be the best and final ISA ever created.

Be it RISC-V or ARM v9 or the Power ISA v3.1?

I think the next level of CPU performance and efficiency will need a much deeper re-think. The whole concept of current ISA's is based on a memory model that basically pretends cache doesn't exist, yet is deeply dependent on it for decent performance. That's got to change, since cache makes life easy for software, by placing nearly all the burden on hardware - very energy-intensive.

For some workloads, I could also see a move towards simpler hardware scheduling, instead relying more on dynamic, runtime optimizations performed by software. Perhaps not entirely unlike what these guys are doing:

https://www.anandtech.com/show/15823/russias-elbrus-8cb-microarchitecture-8core-vliw-on-tsmc-28nm​

 

[bit_user]

View: https://www.youtube.com/watch?v=6TfLVL5GeE4

Just because LTT agrees with you doesn't make it right. It's an opinion and nothing else. In spite of fancy graphs, he has absolutely zero proof. Being an influencer, it makes sense that Linus would tend to view things through that lens.

Apple didn't become a trillion-dollar company by selling hardware at reasonable prices, to its mac addicts. Like any good drug dealer, Apple knows they're hooked on its product. Furthermore, mac users almost seem to enjoy complaining about how much money Apple is squeezing out of them. Once Apple figured this out, they realized it was like a license to print money.