Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Qualcomm's Oryon core: A long time in the making

chipsandcheese.com

Post reply on HN
Print view View original

161–170 of 182 posts

Re: Qualcomm's Oryon core: A long time in the making

#161

Post by ben_w »

TylerE wrote: post #7
Sakos wrote: post #6

Earlier quoted context omitted.

I think people generally expected larger improvements between iterations. Intel and AMD continue to deliver sizeable performance and efficiency gains every 1-2 years while it feels like the Apple M-series isn't getting comparable gains. It definitely seems like Apple has suffered significant enough brain drain in recent years that they're finding it difficult to iterate on the M1.

Hasn't each gen been like 20-30% performance? Isn't that better than what AMD and Intel have managed?

20-30% per year may (I'm not sure) be better than what AMD and Intel have managed since the M1 came out, but it's definitely not as good as the 40-60% per year improvements that I remember when I was a kid.

I'm anchored on those historical performance gains, so I'm naturally (and, I recognise, unreasonably) disappointed by "only" 20-30% improvements from Apple.

Re: Qualcomm's Oryon core: A long time in the making

#162

Post by qwytw »

throwaway55110 wrote: post #159
effie wrote: post #146

Earlier quoted context omitted.

Upgradability of RAM isn't the issue, it's the price of getting access to that RAM. On PC you get it for cheap, you just pay for the RAM modules. On Apple, you get it for arm and a leg, as you don't pay for the modules, but for the privilege to use their higher end model with enough RAM.

Another point of view: The low end models are so cheap that Apple is definitely subsidizing them with revenue from the higher end models. If RAM upgrades were priced based on RAM stick costs, the base model would have to be much more expensive and less people would have access to it. Professional workstations have always cost an arm and leg - both arms and both legs usually, for example a SGI workstation used to cost…

> The low end models are so cheap that Apple is definitely subsidizing

Depends what do you mean by "subsidizing".. Obviously their margins on entry level models are lower but I would still bet than they are much higher if not several times higher than the industry standard. They are certainly not selling them at or below cost, they can probably upgrade the base config to 16GB and still make more per unit sold than Dell/HP/etc.

Even MS get rid of 8GB with the new Surface: 13.8, Snapdragon, 16GB RAM, 256GB SSD is $1000, the equivalent 13.6 M3 Air is $1,299 (8GB is $1100). So if Apple is subsidizing their lower-end Macs what on earth is MS doing? Surface is already a premium(ish) tier device and I'd assume their are paying more to Qualcomm than it costs Apple to make an M3 chip.

> the base model would have to be much more expensive

Why? I don't really get this logic. They are pricing their base models at what the market will bear, it wouldn't make sense to sell them at a discount just because they have more expensive models, they'd hike their prices even more if they could.

What they are making from memory/storage upgrades is basically free money on top of already presumably very reasonable margins (by industry standards).

Re: Qualcomm's Oryon core: A long time in the making

#163

Post by qwytw »

ben_w wrote: post #160
throwaway55110 wrote: post #159

Earlier quoted context omitted.

Another point of view: The low end models are so cheap that Apple is definitely subsidizing them with revenue from the higher end models. If RAM upgrades were priced based on RAM stick costs, the base model would have to be much more expensive and less people would have access to it. Professional workstations have always cost an arm and leg - both arms and both legs usually, for example a SGI workstation used to cost…

They might be subsidising low-end models like that with the M-series, but as they were definitely overcharging for RAM way back when it was user-installable sticks… my gut feeling is RAM* is mostly a differential pricing strategy * for laptops and desktops, storage pricing tiers also give me this feeling; however in the case of tablets and phones, the way they're used — for most people they are the primary computing…

> They might be subsidising

Any (even the slimmest) evidence that this might be the case? Because it seems like an extremely far fetched claim... And year RAM/Storage upgrades seem like a clear example of market segmentation.

Re: Qualcomm's Oryon core: A long time in the making

#164

Post by Rinzler89 »

pjmlp wrote: post #157
Rinzler89 wrote: post #108

Earlier quoted context omitted.

Eh, yes and no. Windows kernel plus it's backend features for security, emulation and virtualization, which enable things like WSL2 and backwards compatibility to work are great, but they're hampered by a crap front end with news, ads, web search in start menu, and dark patterns everywhere being forced down your throat like OneDrive holding your files ransom in the cloud if you don't pay attention or the failed push…

I use Linux since 1995, and have used multiple UNIXes, most of the well known commercial ones since 1993. Unless we are talking about installing it on a classical desktop PC tower, or some embedded device like the Raspeberry like boards with the OEM custom distribution, I am not bothering with it, it goes into a desktop VM and that is it. Too many wasted weekends throughout the years, always sorting out the same issu…

Which issues? Have you considered that maybe those issues from 1995 have been fixed in 2024?

Re: Qualcomm's Oryon core: A long time in the making

#165

Post by ben_w »

qwytw wrote: post #163
ben_w wrote: post #160

Earlier quoted context omitted.

They might be subsidising low-end models like that with the M-series, but as they were definitely overcharging for RAM way back when it was user-installable sticks… my gut feeling is RAM* is mostly a differential pricing strategy * for laptops and desktops, storage pricing tiers also give me this feeling; however in the case of tablets and phones, the way they're used — for most people they are the primary computing…

> They might be subsidising Any (even the slimmest) evidence that this might be the case? Because it seems like an extremely far fetched claim... And year RAM/Storage upgrades seem like a clear example of market segmentation.

When I chose "might be" rather than "are indeed", that was to say it's not impossible rather than to outright agree. Other companies do have loss-leaders, I cannot rule out the possibility that Apple also does exactly what was asserted in the comment I was replying to.

Re: Qualcomm's Oryon core: A long time in the making

#166

Post by pjmlp »

Rinzler89 wrote: post #164
pjmlp wrote: post #157

Earlier quoted context omitted.

I use Linux since 1995, and have used multiple UNIXes, most of the well known commercial ones since 1993. Unless we are talking about installing it on a classical desktop PC tower, or some embedded device like the Raspeberry like boards with the OEM custom distribution, I am not bothering with it, it goes into a desktop VM and that is it. Too many wasted weekends throughout the years, always sorting out the same issu…

Which issues? Have you considered that maybe those issues from 1995 have been fixed in 2024?

Have you considered that maybe I have been using various flavours of Linux for the last 29 years?

Re: Qualcomm's Oryon core: A long time in the making

#167

Post by jodleif »

qwytw wrote: post #78
pantalaimon wrote: post #30

Earlier quoted context omitted.

> The world's most powerful CPU maybe the most powerful / watt - desktop CPUs like the 9950X are still way more powerful.

M4 still has a 10% higher Geekbench single core score than the 9950X.

Geekbench is hardly the definitive performance measure…

Re: Qualcomm's Oryon core: A long time in the making

#168

Post by qwytw »

ben_w wrote: post #165
qwytw wrote: post #163

Earlier quoted context omitted.

> They might be subsidising Any (even the slimmest) evidence that this might be the case? Because it seems like an extremely far fetched claim... And year RAM/Storage upgrades seem like a clear example of market segmentation.

When I chose "might be" rather than "are indeed", that was to say it's not impossible rather than to outright agree. Other companies do have loss-leaders, I cannot rule out the possibility that Apple also does exactly what was asserted in the comment I was replying to.

I'm not sure even if that was the case it would fit the definition of a "loss-leader" unless we assume that Apple makes back the loss and more through the App Store and other services which seems extremely unlikely.

Otherwise who would selling laptops at a loss increase the sales of higher-end laptops, most people don't buy both.

> I cannot rule out the possibility

That seems like an extremely high-bar.

Re: Qualcomm's Oryon core: A long time in the making

#169

Post by qwytw »

jodleif wrote: post #167
qwytw wrote: post #78

Earlier quoted context omitted.

M4 still has a 10% higher Geekbench single core score than the 9950X.

Geekbench is hardly the definitive performance measure…

Well.. it's rather hard to run most other benchmarks on any M4 device. Anyway I don't think the exact % matters too much to get the point across .

But if we compare the difference between M3 and the 9950X in Cinebench and Geekbench and assume the ratio stays comparable it would still be slightly faster

Re: Qualcomm's Oryon core: A long time in the making

#170

Post by oasisaimlessly »

hajile wrote: post #61
monocasa wrote: post #48

Earlier quoted context omitted.

To a degree, that makes sense. Due to the end of dennard scaling, most high end raw compute makes more sense as more but simpler cores, and has for a long time. For instance the blue gene supercomputers made of tons of pretty individually anemic PowerPC 4xx cores. For battery powered devices, race to sleep is the current meta, where once you have some bit of heavy compute work, you power up a big core, run it fast to…

Because clockspeed ramps power exponentially, there’s a limit to how high you can clock before the cost of racing outweighs the savings of running a short time. I believe I’ve read that Apple’s chips run under 3GHz unless their job runs longer than 100-150ms. I suspect that’s their peak race to sleep range.

> Because clockspeed ramps power exponentially

It's not exponential; P = C V^2 f, so power is something between O(f) and O(f^3), depending on how voltage scaling works. It follows that efficiency (energy per instruction) is something between O(1) and O(f^2).

Post reply on HN