I can't give you correct number lol cause I don't know actual throughout of High Na but few Billions easily
We're not taking into account L2 caches when comparing core sizes. L2 caches are not core private caches. SRAM is easy to add. Lion Cove is objectively worse than every other core out there.
Because... that's fundamental to operation. That's transistor threshold voltage where it needs to be to turn on. We've been at ~0.5-1.3V range since 2001. We went from 5V to 3.3V to 2.5V to 1.8V, then 1.5V and finally now.
Try lowering voltage to 0.3V on your CPU and tell me how it goes. No, let's be realistic and set it at 0.55V. Your frequency even if it can be run will crash. Like 200MHz crash. Yes it'll run at ultra low power, but it's unusable. Even M4 is unusable at 200MHz.
This room for voltage scaling is getting smaller and smaller because they've been doing it since 2001. All the levers are starting to be exhausted. There's no room for a massive jump to a 24-wide, 13mm2 CPU from what it is today. And unless most of the core logic is useless it will suck power like a madman too. It has to be steady and balanced so your execution doesn't fail.
Sorry, I shouldn't have used Apple and Qualcomm as examples there, since they use large shared caches to eliminate the need for an L3...
But L2 caches are usually counted as part of the core. Any core private cache usually is. Hence why the figure of "13mm2" without any context isn't very useful.
Except if you have 2 or 3x the IPC of GLC, you can run at ridiculously low clocks and still have very good performance.
And 1T perf is the entire point of RYC it seems like, even if you go to the Ahead computing website you will see them prioritize ST perf and also perf/watt.
And current cores also run much, much higher in even server than their Vmin. I'm guessing Royal Core would be running their cores much closer to Vmin, and take advantage of shifting the V/F curve up to get much better perf/watt.
I think the people in charge of RYC think that just different physical designs(like AMD), or even outright just a different E-core arch (like Apple and Qualcomm use) to significantly lower Vmin for the markets that need core density, or maybe used as a LP island for mobile.
The 7x was a projection or a design goal. Additional searches tried to find the actual, and that's how Grok came up with 3x to 7x range
I think you are way underestimating the abilities of LLM models. CoPilot inadvertently came up with a different approach (to find the cost ratio between CoWoS and InFO), to go from the other side, from the cost of wafer (but mixed up the units).
But then, I prompted Grok to try this approach (starting from cost of wafer) and then, using the same context it already had, meshed together its first reasoning approach with the 2nd one. And came up with the cost of CoWoS in the same range as intel Foveros and various additional wafer cost estimates.
In either case, Intel took on a whole bunch of extra overhead with its implementation of its advanced packaging, and this albatross is going to be hanging over Intel's ability to maintain market share on the low end of the client segment.
Which brings us back to the subject at hand - Gelsinger and his legacy. This information, on high cost of Intel packaging is another piece of evidence how delusional Gelsinger was - in ignoring the costs. Probably thinking Intel would be so far ahead in performance and technology that Intel can charge whatever price they want, and customer will pay.
It's not a "problem", it is a performance parameter that all the models are addressing. Which, stated differently, is for the model to come up with wrong answer with high confidence. So they are addressing both ends, getting fewer answers wrong, and also assessing (and informing of user) of level of confidence.
I take it as the old "I am feeling lucky" button on Google. There is a possibility that's what I am looking for, so I glance at it. And you don't have to click on anything to take you away from the search results. But overall, I find Google to be quite weak in what it presents as a result of it own LLM (in my experience).
Do you happen to have any idea what happened with Intel's research into near threshold voltage operation? IIRC they were talking about this in the early 2010s, targeting voltages as low as 0.3v.
Something's gonna have to give with AI, they aren't going to be able to source sufficient power to build everything they talk about, nor support the cost of operating it without profits to pay for it at some point. If Intel had been able to get that to work that sure would be an advantage for their foundry right about now...
Intel has done lot of research but sometimes they can(as in disband not coco cola can) the useful stuff and go into stuff that doesn't even makes sense
Which is fine. When Wall Street finally gets bored of AI, all the companies will slash spending.
They just stopped talking about it.
That's why germanium transistors were the rage for a while because it had Vth of 0.3V but it's still not here. Also it's one thing for a single transistor, it's another thing for a whole CPU. Also what frequency will they run at? Low Vth and germanium is good, but not so much at 2GHz.
Advancements were made for sure for CPUs to reach 6GHz. Years ago they were talking about more resilient circuitry techniques for higher frequency operation and they might have been using that now. Those were the good old days of IDF.
Gelsinger was likely betting more than that, such as the lockdown premiums continuing. When that stopped the demand plummetted.
There's a company that offers ebike renting service and they haven't been making much money either. The ones where they are randomly put in spots with only a phone needed to unlock, pay and use them. I think they are not far off from bankruptcy. They appeared around the same time period. So few companies also bet on that continuing. They were also way too expensive. But convenience beats not having debt for some people I guess.
L2's aren't really private. It's easy to add, hence why we don't include them in the figures. Everybody can do them. But can they actually make a good CPU? The smarts are in the uarch, not the higher level caches.
Also unless the "private" cache is in the range of 10MB, vast majority of that 13mm2 is the core, and I doubt it was included.
AMD also beats Lion Cove substantially. Remember Zen 5 is on the older process.
Last edited:
i feel like A.I bubble could be worse then dot com in terms of impact to wall street.
like tech bro CEO's can't keep telling everyone they are going to take their jobs and then at the same time ask for more VC funding at some point the rubber is going to have to meet the road.
NTMBK
Lifer
- Nov 14, 2011
- 10,424
- 5,740
- 136
They can support open source just fine without running support and maintenance for an entire distro.
Io Magnesso
Senior member
- Jun 12, 2025
- 578
- 164
- 71
The CEO actually bought $25 Million worth of stock from his own Money it was in the agreement
Why Microsoft is Intel of software: bloated, inefficient, and pricey.
5-15% frames boost and twice the battery life under load is astonishing to me.
5-15% frames boost and twice the battery life under load is astonishing to me.
Apologize to Intel.
Intel has improved their products over the last decade. Microsoft has focused on making Windows worse and adding alternate revenue streams.
Nailed it:
"If we are unable to secure a significant external customer and meet important customer milestones for Intel 14A, we face the prospect that it will not be economical to develop and manufacture Intel 14A and successor leading-edge nodes on a go-forward basis," the company wrote. "In such event, we may pause or discontinue our pursuit of Intel 14A and successor nodes."
But people told me 18A was getting canned not 14A 18A is a long node they will use it throughout 2030 lol
adroc_thurston
Diamond Member
- Jul 2, 2023
- 6,244
- 8,794
- 106
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 21K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter