How does that help?Microsoft's possible solution would be to force Bitlocker encryption on all laptops.
Then the partition shouldn't be accessible from Linux.How does that help?
You should see the interview with ARM CEO Rene Haas;View attachment 100725
Delusion or do they have an actual plan to invade PC users' homes and offices?
Isn't it like 80% of all PCs (laptop+desktop) don't have dGPUs?That's like saying that 50% Windows PCs have no need for dGPUs.
Since Nvidia's PC CPU/SoC ambitions are going to be ARM based, that also contributes to the 50% marketshare figure claimed by Rene Haas.AMD/Intel will probably not make drivers for their dGPUs available on QC's platform and Nvidia is more likely to pair their dGPU with their own ARM CPU.
Delusional. x86 isn’t going anywhere. I predict no more than 10% by 2030, a bit higher if Intel/AMD totally drop the ball and Microsoft/ARM/Qualcomm heavily invest, and a bit lower for any other scenario.View attachment 100725
Delusion or do they have an actual plan to invade PC users' homes and offices?
That's like saying that 50% Windows PCs have no need for dGPUs. AMD/Intel will probably not make drivers for their dGPUs available on QC's platform and Nvidia is more likely to pair their dGPU with their own ARM CPU. The only other explanation seems to be that QC will have at least a Geforce 7070 class dGPU by 2029.
I'd wager atleast 25% by 2030. You are forgetting Nvidia/Mediatek is also getting on the bandwagon.Delusional. x86 isn’t going anywhere. I predict no more than 10% by 2030, a bit higher if Intel/AMD totally drop the ball and Microsoft/ARM/Qualcomm heavily invest, and a bit lower for any other scenario.
It's showing how much RAM the GPU is allowed to use. If the CPU wants to access data from GPU's RAM pool, it should be instantaneous without any data copying involved since it exists on the same RAM chips. But for any other data manipulation, I don't think the CPU will be allowed to modify GPU's data. Then some data copying will be involved from GPU pool to CPU pool but it should be quicker than having to involve the PCIe bus in a dGPU's case.Does Windows not have a Unified Memory model like MacOS?
I think it's a non-issue: almost 80% of the machines sold in Q2 2023 use iGPU.View attachment 100725
Delusion or do they have an actual plan to invade PC users' homes and offices?
That's like saying that 50% Windows PCs have no need for dGPUs. AMD/Intel will probably not make drivers for their dGPUs available on QC's platform and Nvidia is more likely to pair their dGPU with their own ARM CPU. The only other explanation seems to be that QC will have at least a Geforce 7070 class dGPU by 2029.
There's a reason Intel is the biggest graphics manufacturer in the PC space, yeah.I think it's a non-issue: almost 80% of the machines sold in Q2 2023 use iGPU.
In the case that ARM really takes off on the PC, Intel/AMD not supporting dGPUs would pure suicide. Nvidia would support ARM just to watch them squirm.AMD/Intel will probably not make drivers for their dGPUs available on QC's platform and Nvidia is more likely to pair their dGPU with their own ARM CPU. The only other explanation seems to be that QC will have at least a Geforce 7070 class dGPU by 2029.
Unified memory requires the memory management units in the CPU and the GPU (and NPU) to be coherent. Without that, even sharing physical memory, it has to be sectioned off, so that each controls memory that can’t be touched by the other. Which means that data has to be transferred back-and-forth within the same memory die. The operating system can’t support it if it’s not built-in to the chips.Geekerwan video about Snapdragon X Elite (not review):
Highlights:
View attachment 100726
View attachment 100727
Does Windows not have a Unified Memory model like MacOS?
What a shame.So none of these new consumer Windows SOCs have it.
I'd wager atleast 25% by 2030. You are forgetting Nvidia/Mediatek is also getting on the bandwagon.
Nvidia/Mediatek will have the same problem Qualcomm has. They are too late. AMD has been advancing performance per watt for a while now. Strix and Zen 5 will be another step. The new Intel cores and Lunar Lake appear to be transformative.
You are underestimating the value of the mobile heritage that Apple, Qualcomm and Mediatek have. They have technologies that allow the SoC to operate at sub-1W. These technologies are essential for smartphones, where battery life, and thereby the power efficiency of the chip, is crucial. It is these technologies that they are bringing to the their PC chips, that will enable extraordinary battery life.I’ve been arguing for a long time now the Apple’s top performance and performance per watt achievements are based on micro architecture work that could be implemented with any ISA. I think that’s going to be demonstrated with the 2024 x86 chips.
They have to opensource their Hexagon NPU libraries too for maximum benefit.Linux support (this seems to be WIP).
Qualcomm and Mediatek’s mobile heritage is lower performance. There was nothing transformative about their previous designs.You are underestimating the value of the mobile heritage that Apple, Qualcomm and Mediatek have. They have technologies that allow the SoC to operate at sub-1W. These technologies are essential for smartphones, where battery life, and thereby the power efficiency of the chip, is crucial. It is these technologies that they are bringing to the their PC chips, that will enable extraordinary battery life.
Sure, Intel and AMD can also develop these technologies, but they do not have the mobile heritage that Apple/Qualcomm/Mediatek have. They do not have the experience and knowledge gained from developing smartphone SoCs for more than a decade. This is where the real value of the mobile heritage lies...
Lunar Lake is a step in this direction and a huge advancement, but I can say with certainty it's not enough to catch up to Apple/Qualcomm, and make up for Intel's lack of mobile heritage. It will take them several more generations to close the gap.
You are only talking about the CPU core microarchitecture. Why will you not talk about the fabrics? Low-power interconnects? Power gating? These are all features of the Mobile heritage I was talking about.Qualcomm and Mediatek’s mobile heritage is lower performance. There was nothing transformative about their previous designs.
Starting with the A7, Apple started on a path that no one else was doing, fighting for desktop level performance at very low power. This was straight up microarchitecture. There are no circuit-level advantages. Every company using TSMC has access to the same transistors that Apple has access to.
Intel has top semiconductor engineers. Their high-watt strategy was a corporate choice, not because they don’t know how to do low-watt circuits.
Now the microarchitecture concepts needed for high performance combined with efficiency have been authorized by everyone. From Nuvia Oryon to the Cortex-925 to Zen 5 and 5c to Lion Cove and Skymont the work has been done.
Apple and Qualcomm will probably battle for the battery life crown but it’s just not going to be the gap Apple previously enjoyed. Apple has already staked their share, plus Apple has fans of their OS. Qualcomm and Mediatek can’t leverage the unique OS and features that Apple can.
Cortex X4 made a high-performance-low-watt bid, but it’s not quite there. Will Cortex X925 make a stronger case? Mediatek needs it to.
But x86 is just not going to be far behind. The Skymont strategy I expect will give the heart of the market a reason to not consider Qualcomm. Nowhere near as low power as Apple E cores, but strong enough to keep the P cores off most of the time.
And how are slices different from something like WGPs in AMD RDNA, or SMs in Nvidia GPUs?The problem would be workload distribution on these slices. Not every element in a graphics scene has the same complexity. The synchronization of the slices could be a problem in complex scenes where complex geometry may bog down one or more slices and other slices have to wait for the overburdened slices to catch up.