News SMT4?!

[NTMBK]

Apparently we're going to be seeing SMT4... from Intel.

Intel Royal Core Project to improve x86 architecture efficency - VideoCardz.com

Intel Royal Core Project rumored to improve X86 architecture to better compete with Apple products Moore’s Law is Dead has a video covering a wide spectrum of Intel’s future consumer and data center products. According to the YouTuber, who has a very good track record on Intel leaks, the...

videocardz.com

At some point I might work up the energy to watch the 20 minute video, which will convey things that I could otherwise read in a 3 minute article. Sigh.

 

[soresu]

I am actually hyper optimistic about RISC-V.

I'm optimistic too, but I don't think it's going to happen any time soon going by the time it is taking for an ARM64 transition on the Windows side.

If Google's new Fuchsia went big time (desktop/laptop/tablet/phone) with RISC-V as a 1st class hardware citizen along with ARM64 then I could see it getting a big boost though.

Another possibility is a nu hardware player like Tenstorrent launching off the bat with an Apple Ax level IPC CPU core in concert with a halfway decent emulator / binary translation engine like FEXemu bringing a somewhat efficient RISC-V backend to bear allowing it to take advantage of abundant x86 legacy code (FEXemu is already getting really good at this for ARM64 in combo with Proton/DXVK).

 

[Geddagod]

I don't see Apple as being much of a competitor of Intel's. Not in any way. Sure at the margins there are some people who want a silent / long battery life laptop with high performance and might be willing to switch between Windows and Mac to get the best choice. Most people will not, they make the decision between Windows and Mac before they even begin looking at hardware.

Apple was already selling Macs that were competitively worse than Intel's offerings pretty much the entire time they were x86. Other than right after the release of a new model, if it got the latest and greatest from Intel, Intel always had something better than what was in a Mac. That didn't stop Apple from selling 20 million Macs a year. So even if Intel's products are superior to Apple Silicon, it isn't going to hurt Apple's sales. People who are buying Mac chose to buy a Mac first before looking at what else is out there. People are buying Windows are mostly unaware of what Apple offers and wouldn't consider them even if they were so clearly better it wasn't possible for anyone to come up with a benchmark saying otherwise.

Honestly this. Most of my friends have apple laptops for three main reasons: battery life, reputation of apple, and ecosystem
Even if Intel catches up to Apple in, say, battery life, the apple ecosystem advantage as well as the social aspect of apple devices essentially just block people from going back to windows IMO. Plus the battery life is nice, but there are ports in essentially every classroom, and you can just charge at home. You essentially just need it to last one school day, which isn't much. Same reason I bring my ROG M16 to school, just need it to last the day.
An example of the apple ecosystem is being able to copy paste a link that you took a picture off on your phone, directly on to your apple laptop. I don't think Windows has anything similar (if there is please do tell me it would be a god send for one of my classes lol), or even if it did, it's not nearly as common knowledge to the simple user (me) as those same features are to Apple users.

 

[Geddagod]

Eh? SMT doesn't help (actually, actively hurts) single thread execution. The idea is you can take advantage of underutilization from one thread to fit in more. Very loose correlation to speculative execution.

Nothing fundamental, but that chart shows the problem. For Gracemont vs Golden Cove, just to use one x86 reference point, a 4 core Gracemont cluster provides ~2x the throughput of a Golden Cove core (SMT included) with approximately the same area and power budget. With the same kind of scaling from that chart, even if you added SMT4 to Golden Cove with zero incremental power or area overhead...you'd still be better off with Gracemont. So what exactly is the point?

You have to wonder then, wouldn't stuff like Sierra Forest and Bergamo cater to MOST of the server market rather than just being side projects to the main Zen 4 and Golden Cove server parts? After all, they should have better throughput and also efficiency, and with HT on, better per thread perf as well. Or am I mischaracterizing the performance and hardware needs of the server market?

 

[Exist50]

You have to wonder then, wouldn't stuff like Sierra Forest and Bergamo cater to MOST of the server market rather than just being side projects to the main Zen 4 and Golden Cove server parts? After all, they should have better throughput and also efficiency, and with HT on, better per thread perf as well. Or am I mischaracterizing the performance and hardware needs of the server market?

Yeah, I don't know what the exact workload affinity breakdown is, but I imagine that long term, the ratio between the two will be much closer to 50-50. Will take a number of gens to get there, but I think that's where we're headed.

 

[soresu]

An example of the apple ecosystem is being able to copy paste a link that you took a picture off on your phone, directly on to your apple laptop. I don't think Windows has anything similar

I think something similar does exist to link Android phones and Windows together.

Here's a couple of links I foudn with a quick Googling:

How to Connect Your Android Phone to Your Windows 11 PC

Connecting the two lets you send text messages, see notifications, instantly grab photos, and even run apps from your Android on your Windows 11 PC. We show you how to get started.

uk.pcmag.com

Evolving the connection between your phone and your Windows PC

Phone Link, formerly known as Your Phone, brings your Android phone and your Windows PC closer together. Today we are thrilled to announce the evolution of Your Phone app as Phone Link. We introduced the Your Phone experie

blogs.windows.com

 

[Panino Manino]

I think something similar does exist to link Android phones and Windows together.

You can also do this on Linux very easily.

 

[Doug S]

I think people are a little more willing to switch that you make them out to be. Certainly Apple's Mac sales with the ARM transition have been quite favorable. There's only so long this kind of gap can remain.

M1 sales benefited from people knowing they were coming and holding off a purchase to wait for them to come out. It remains to be seen if Mac sales benefit from the Apple Silicon transition in the long term but I doubt it. The offerings are better in some ways which may drive a bit of additional adoption but that's probably offset by no longer being able to easily run Windows software in a VM for those who can't erase Windows from their life completely (i.e. particularly in the corporate world)

 

[Doug S]

That doesn't make it better for Intel. By losing Apple as customer Intel lost a significant chunk of the market. They need to recover that loss somehow somewhere.

I'm not arguing against that, I'm arguing against the idea that Intel sees Apple as a major competitor they have to "respond" to. They lost Apple as a customer, that can't be undone. It really isn't even their fault, it was going to happen though Intel's stumbles made it really easy for Apple by allowing Apple to sell Apple Silicon Macs that were superior to the Intel Macs before them - even beating them under Rosetta 2 emulation in some cases.

If Intel wants to increase sales they have to do so from AMD's share - thus they are the only one Intel sees as competition.

 

[moinmoin]

If Intel wants to increase sales they have to do so from AMD's share - thus they are the only one Intel sees as competition.

That would be incredibly shortsighted of them to do. Intel needs to worry about the TAM. AMD's market share in x86 CPUs is easy to reach but also very limited still. By losing Apple as a customer Intel not only lost a big customer, that also reduced x86's TAM by losing Apple's ecosystem. And that's what Intel has to respond to, preventing x86 from losing further ground to competing ISAs Intel can't profit of in any way, ideally finding ways of increasing its share again.

 

[Doug S]

That would be incredibly shortsighted of them to do. Intel needs to worry about the TAM. AMD's market share in x86 CPUs is easy to reach but also very limited still. By losing Apple as a customer Intel not only lost a big customer, that also reduced x86's TAM by losing Apple's ecosystem. And that's what Intel has to respond to, preventing x86 from losing further ground to competing ISAs Intel can't profit of in any way, ideally finding ways of increasing its share again.

True Intel will have to watch out for Qualcomm, assuming legal issues don't prevent them from releasing their Nuvia core and assuming Intel & AMD don't do enough arm twisting (sorry) on OEMs to prevent proper Qualcomm based Windows PCs from coming to market.

What they really have to do is not stick their head in the sand and ignore future giant market opportunities, like the mobile market they first ignored then shipped crappy cores on N+2 processes, relying on bribery to get OEMs to use them. It is a lot easier to increase revenue by becoming a big player in a new and fast growing market than by stealing share in a stagnant market like the PC market (which had been slowly shrinking for about a decade until the pandemic bubble)

 

[DrMrLordX]

True Intel will have to watch out for Qualcomm, assuming legal issues don't prevent them from releasing their Nuvia core and assuming Intel & AMD don't do enough arm twisting (sorry) on OEMs to prevent proper Qualcomm based Windows PCs from coming to market.

MS will continue to release Qualcomm-based mobile PCs at the very least. They're invested in the WARM ecosystem, and they'll continue on that course until Qualcomm decides to give up (if ever).

 

[soresu]

MS will continue to release Qualcomm-based mobile PCs at the very least. They're invested in the WARM ecosystem, and they'll continue on that course until Qualcomm decides to give up (if ever).

With this whole Nuvia ARM license thing still going Qualcomm are completely up to their necks in it now.

They have to exploit Oryon to the utmost advantage and they will ply MS to the moon and back for that purpose.