#CPU_architecture

[phonemantra-blog]

But only server ones Despite the fact that there is still quite a bit of information on the Internet about the AMD Zen 5 architecture and products based on it, today we have received quite detailed data about CPUs based on Zen 6. [caption id="attachment_85418" align="aligncenter" width="780"] AMD[/caption] The information concerns mainly server CPUs, but it is still important, because some things can be implemented in consumer ones. [caption id="attachment_85419" align="aligncenter" width="780"] AMD[/caption] The author of the Moore's Law Is Dead channel obtained data on Epyc-E server processors, which are intended for telecommunications and peripheral systems. Such CPUs will only have up to 64 cores, but the “larger” Epyc will, of course, have more cores. AMD will pack 32 processor cores into one chiplet. [caption id="attachment_85420" align="aligncenter" width="780"] AMD[/caption] In this case, what interests us most is that Epyc-E will have one chiplet with 32 cores! Let us remember that in almost all AMD processors the chiplet contains eight cores, and only Epyc Bergamo with small Zen 4C cores has chiplets with 16 cores. As you can see, by the time Zen 6 is released, AMD will be able to pack 32 cores per chiplet. This obviously won’t affect consumer Ryzen, but perhaps by that time they will switch to a package of 16 cores per chiplet. Additionally, the 32-chip configuration could allow AMD to easily build 256-core processors. This could already be CPU Venice in SP7 version. https://youtu.be/qpzTmKjaIMU Also, a large number of cores in one chiplet allows AMD to place additional chiplets in the processor without any processor cores at all. FPGAs or other specialized chiplets can be implemented in this way.

[nikolapavlica]

Porting an entire Android operating system

Name of activity: Android kernel and system development on the Redmi 7A

Date(s) of activity: June of 2020. all the way towards 23. 7. 2020.

Approximate # of hours: 100 hours (if not more)

This activity is: Creativity, Service

Learning outcomes: Strength & growth Challenge & skills Initiative & planning Working collaboratively with others Showing perseverance and commitment Global engagement Recognize and consider the ethics of choices and actions

Anyway the porting process came as I way playing around with the android kernel of the device, which I got from the official Xiaomi sources: https://github.com/MiCode/Xiaomi_Kernel_OpenSource

But immediately after that the struggles came. 

Firstly, you cannot JUST compile it using any old compiler, OH NO. You have to use exactly the compiler the orignal devs originally used, not your FOSS linaro toolchain (get rid of that ugliness, ugh), but rather a custom propriatery version of the clang toolchain used by Qualcomm internally. Oh yes, you also need to get a Qualcomm account approved before you start, great...

Enough of the compiler blaming, compile that thing already.

Fine, but wait there’s more. You cannot simply use the defconfig provided by the original developer. That’d be silly, instead what you need to know is the special quirk where you need to include DTBO (device tree files that basically describe the device in full detail so the OS actually works, in short) files into the kernel, otherwise you’re toast and will get nowhere.  - I’ve already bricked my phone here once, if it wasn’t stressful enough. And here I’ll go out and say f*** EDL mode and it’s hard to reach pins. This was required for me to unbrick it and put it into a functioning state.

What caused this brick? Basically avoid flashing the DTBO file (AT ALL COSTS), regardless of what people say. The bootloader will wuss out and won’t work anymore if you do, causing a brick.

After that’s been done you’d hopefully get a kernel that boots. Just without the Wi-Fi drivers, FUN. That requires you to properly include the Wi-Fi driver with the according version, you can’t just add any matching driver, you have to match the version as well. Funsies.

Alright already, enough. We’re not done yet :) You also need to make sure those drivers are signed or the android phone’s gonna think it’s been hacked and won’t launch them. This is thankfully easily overridden by changing a simple policy in the Linux kernel. Thank you, Linus.

Ok, finally we get a booting kernel with working Wi-Fi. But this is just for the same one your device manufacturer shipped. You need to actually port this now.

Before anything, you need to know that your CPU actually supports a 64-bit instruction set. Once you figured that out, you need to find (ideally) the most similar device to your own, which happened to be the phone that succeeded this one in it’s own series - the Redmi 8A.

Since the two devices were in the same kernel tree, I could just “git diff” the two kernels, revealing any changes. I’ve saved those changes to a single file that I could read later. vimdiff becomes your friend here. btw the diff file was 10 gosh darn megabytes

I started off of the 8A kernel, instead of the 7A, since it had a few missing entries regarding drivers and whatnot and I couldn’t been bothered to fix them myself.

The first thing you’d want to look at if you want to get a device to boot is the defconfig. It’s located within the arch folder of the linux kernel. There you’d find a file in the cpu_architecture/defconfigs folder that contains the name of your device. When you open that file, you see all the features and drivers that had been enabled in the kernel. You can compare it with the original in order to figure out what you need and what you don’t need. For example, you’d want to keep all of the ARM64 related stuff, but you’d want to disable USB-C since my device doesn’t have it and so on. After the mixing and matching you’d ideally get a kernel that perfectly works with your phone. But not so fast.. There are still issues to contend with.

For starters, bluetooth, screen rotation, audio, brightness control and even the touchscreen stopped to work. Here’s how I fixed each of those things:

Touchscreen

In the absolutely massive diff file I spotted a few binary files being changed. You guessed it, it’s the driver for the touchscreen controller, silly. Swapping it back where it belonged fixed the touchscreen, yupee.

Bluetooth

This involved changing out files within the /vendor partition, but overall was also simple to fix. There are policy files within /vendor/etc, you’d want to check them out and find any discrepancies and fix them.

Bluetooth audio was fixed by the community by doing the same thing, although better than me, I guess.

Audio

In the massive 10 megabyte diff the entire audio kernel driver was stripped. Reason? I don’t know, but just including it again seemed to fix the problem. I also struggled with this a lot since the driver took for ducking ever to build. Like imagine having to wait an hour each time you wanted to test something out, that’s how bad it was.

Brightness control and screen rotation

In the same vain as the bluetooth there are policy files for the brightness control and sensor information. Change this accordingly if you want them to work.

There are other things that should be listed here, such as FM Radio, audio jack problems and LTE calls. However I wasn’t able to fix everything on my own, which is why I left it to the community to finish what I couldn’t.

And the result speaks for itself:

https://nik.ex-booking.com/posts/2020/porting-kernels-between-architectures/ https://forum.xda-developers.com/t/64-bit-for-redmi-7a.4145225/ https://forum.xda-developers.com/t/redmi-7a-arm64-kernel.4134381/ https://github.com/nikp123/Xiaomi_Kernel_OpenSource/tree/pine64-p-oss

Thanks to all the awesome people that helped me along the way. I couldn’t have done this all by myself.

Reflection: Ever since I’ve got my Redmi 7A, I’ve gotten to admire the hardware I was using. The battery, screen, CPU, GPU, RAM... all great. However the software frustrated me and a lot of the people on the internet. By writing a 64-bit operating system I could relieve some of the annoyances, which, thankfully, DID happen.

The process wasn’t easy (like. at. ALL), but I got cheered (virtually) along the way, which was nice. Finally my efforts are paying off it seems. I’ve also gotten to meet a few people: some rude, some nice, some striving to do what I did and at the end I believe it was a fun journey.

Also congrats for having the nerve to read this whole mess from beginning towards the end. Hopefully this becomes less of a struggle in the future.

[phonemantra-blog]

There are still a few years left before such a memory is released. HBM4 memory will appear on the market only in a few years, but it has already become known that its appearance may be associated with a small revolution in the field of semiconductor production. Hynix is ​​reportedly working with Nvidia and other manufacturers on technology that will allow HBM4 memory chips to be placed directly on the computing chip. [caption id="attachment_83411" align="aligncenter" width="600"] CPUs and GPUs[/caption] Now HBM memory is placed on the same substrate as the CPU or GPU, but next to it. Placing memory on the CPU/GPU will not only change the way logic and memory devices are typically connected, it will also change the way they are manufactured. This will change the way CPUs and GPUs are made. In fact, this is approximately the same as AMD's 3D V-Cache, only with a different type of memory. SK Hynix has now begun recruiting specialists to design logic semiconductors such as CPUs and GPUs and is discussing its HBM4 integration design method with several companies without its own manufacturing capabilities, including Nvidia. It is likely that Hynix and Nvidia will initially jointly develop a specific GPU, and then TSMC will produce this chip and immediately place HBM4 memory on top. Considering the type of memory, we are clearly talking about specialized accelerators for AI or data centers. [caption id="attachment_83412" align="aligncenter" width="385"] CPUs and GPUs[/caption] Hynix took on this idea for a reason. The thing is that HBM4 memory will use a 2048-bit interface to connect to host processors, so the latter will be extremely complex and expensive. A direct connection between memory and CPU/GPU will simply be cheaper. However, it is too early to draw any conclusions, since on the way to such implementation there is a very banal problem - heating. The consumption of the Nvidia H100 accelerator is very high, and in a few years such solutions can consume much more. Memory also consumes a lot, and when all this is placed next to each other, there are no problems with cooling. And when a “sandwich” is created, the heating increases significantly with a smaller area of ​​the final crystal. Perhaps Nvidia accelerators in a few years will initially be designed exclusively for liquid cooling.

[phonemantra-blog]

Gelsinger isn't shy about criticizing his company. Intel CEO Pat Gelsinger spoke about three big mistakes and failures of the company. [caption id="attachment_79812" align="aligncenter" width="600"] Intel[/caption] The head of Intel called processors for smartphones the first such failure. Recall that Intel has been releasing Atom platforms for Android smartphones for some time, and these platforms have generally been quite competitive. They never became widespread, but there were even flagships based on Atom. True, the story did not last long. The first smartphone with such a SoC was released in 2012, and the direction was closed in 2016. These are three huge mistakes and failures of Intel. Gelsinger also believes that Intel was wrong to abandon plans to create graphics processors focused on computing. More precisely, we are initially talking about the Larrabee line, which was canceled just after Gelsinger left Intel in 2010. The project later became Xeon Phi, but with much less ambition. Soon it was closed too. Gelsinger notes that Intel would not have had to subsequently buy five different AI companies if the Larrabee project had continued to be developed. The third point concerns the semiconductor business. Now Intel is actively developing it, trying to become a major market player together with TSMC and Samsung, and also promising that it will master five technical processes in four years, but the head of the company believes that it was necessary to start much earlier. However, in this case the company is at least actively working in this direction.

[phonemantra-blog]

This is the Emerald Rapids generation The first leak has appeared on the Internet regarding the new Emerald Rapids generation Intel Xeon Scalable processors, which will be presented on December 14 along with Meteor Lake . [caption id="attachment_79561" align="aligncenter" width="780"] Intel processor[/caption] The flagship Xeon Platinum 8592+ has been revealed, which will offer 64 cores, which does not yet look particularly impressive compared to AMD’s 96-core monsters. And relative to Sapphire Rapids, this is an increase of only 4 cores. All cores, by the way, will be large. Technically, these are Raptor Cove cores, like the current Raptor Lake. The new CPU will receive 320 MB of third-level cache. The total with the L2 cache is already 448 MB. This will be one of the main improvements over the current generation Xeon Scalable. [caption id="attachment_79562" align="aligncenter" width="760"] Intel processor[/caption] The peak power of this Intel processor can reach 922W. 64-core Xeon Platinum 8592+ appeared on the Internet The test sample had frequencies of 1.9-3.9 GHz, and all cores can simultaneously operate at 2.5 GHz. The production CPU is expected to operate at frequencies of 2.0-4.0 GHz. The power limits PL1 and PL2 are 350 and 420 W, respectively. The peak reaches 764 W, but the CPU can be configured so that the value reaches 922 W. [caption id="attachment_79563" align="aligncenter" width="483"] Intel processor[/caption] The test copy in the CPU-Z benchmark scored 610 and 25,741 points in single-threaded and multi-threaded modes, respectively, but the result is not very good, which is apparently due to the unavailability of the software.

[phonemantra-blog]

The rest of the CPUs will be shown at the beginning of 2024 Intel introduced the 14th generation Core processors (Raptor Lake Refresh). As expected, today only older models with an unlocked multiplier were announced. Intel Core i9-14900KKF There were a lot of leaks about these CPUs, so Intel didn’t reveal anything new except the price. The new generation differs minimally from the previous one, except for the Core i7 models, which have four smaller cores compared to their predecessors. Intel Core i9-14900KKF Intel Core i9-14900K/KF, Core i7-14700K/KF, and Core i5-14600K/KF processors are presented. But Intel did not raise prices, which is good, because there were such rumors. However, they were there last year too. So, the characteristics of the new products are as follows: Core i9-14900K/KF – 24 cores (8+16), frequencies 3.2–6.0 GHz, 36 MB L3 cache, TDP 125 W and price 590/565 dollars; Core i7-14700K/KF - 20 cores (8+12), frequencies 3.4–5.6+ GHz, 33 MB L3 cache, TDP 125 W and price $385/$410; Core i5-14600K/KF - 14 cores (6+8), frequencies 3.5–5.3+ GHz, 24 MB L3 cache, TDP 125 W and price 295/320 dollars. To demonstrate the performance gain from the transition to the new generation, Intel, as expected, took the Core i7-14700K because it can demonstrate this gain. The new CPUs will be available for purchase starting tomorrow.

[phonemantra-blog]

But this is hardly good for the CPU Alder Lake and Raptor Lake processors are characterized by very high power consumption and heating. Raptor Lake Refresh, due to increased frequencies and the absence of any changes in the architecture, will consume and heat up even more. And if desired, in some cases it will be possible to allow these CPUs to heat up to 115 degrees. [caption id="attachment_68854" align="aligncenter" width="780"] Intel Core processors[/caption] The new 14th-generation Intel Core processors can be allowed to operate at temperatures up to 115 degrees. MSI will make it possible to increase the processor throttling threshold for new Intel CPUs from the current 107 to 115 degrees. This feature will be available in the BIOS of motherboards based on 700 and 600 series chipsets. The setting will allow the processor to operate stably at temperatures up to 115 degrees. And this will only be possible with the new 14th-generation Intel Core. It's unclear whether these temperatures are acceptable for the new CPUs, but the setting is likely designed for overclocking enthusiasts rather than casual users.

[phonemantra-blog]

And we will see Core Ultra 5/7 on December 14 Intel Meteor Lake processors will be presented this year but at the very end. According to recent data, not all at once. [caption id="attachment_64479" align="aligncenter" width="780"] new Core Ultra processors[/caption] The announcement of new CPUs is expected on December 14, but the Golden Pig Upgrade Pack insider says that not all models will be shown to us on this day. We are talking about the announcement of the Core Ultra 5 and Core Ultra 7 models. But Core Ultra 9 will be released in 2024, and it is not yet clear when exactly. Intel can't or won't announce all the new Core Ultra processors at the same time. Why such a division would occur is unclear. At the same time, if a conditional month passes between announcements, then this in itself is not scary, since the younger Meteor Lake will appear in laptops as early as 2024. But at the same time, Core Ultra 9 can linger much more. However, taking into account all the existing leaks, we know that Core Ultra 7 will still have the same six large and eight small cores as Core Ultra 9, so there probably won’t be much point in chasing older CPUs. Insider Golden Pig Upgrade Pack once shared performance leaks of Intel CPUs of different generations and published Intel slides describing the Meteor Lake CPU.

[phonemantra-blog]

Just an engineering sample The first tests and characteristics of the 60-core Intel Xeon Platinum 8580 processor have appeared on the Internet. We are talking about an engineering sample (ES2) of a processor from the Emerald Rapids line, which will be presented on December 14, and will hit the market next year. [caption id="attachment_58593" align="aligncenter" width="780"] Xeon Platinum 8580[/caption] These processors will replace the current Sapphire Rapids, offering a new architecture (Raptor Cove), but maintaining the Intel 7 process technology. The latter is the reason why the maximum number of cores will remain unchanged. Xeon Platinum 8580 appeared on the Internet [caption id="attachment_58594" align="aligncenter" width="780"] Xeon Platinum 8580[/caption] Each of the 60 processor cores has 2 MB of L2 cache, so the CPU has 120 MB of L2 cache in total. The third-level cache memory reaches an impressive 420 MB. However, Intel's brainchild is very far from the AMD Epyc Genoa-X line with its 1152 MB cache. It’s not worth focusing on tests yet, since this is an engineering sample.

[phonemantra-blog]

In single-threaded mode, the Core i9-14900KF beat everyone The Core i9-14900KF processor appeared in the PassMark benchmark database and immediately became the leader in the single-threaded performance rating. [caption id="attachment_58562" align="aligncenter" width="780"] Intel CPU[/caption] Despite the fact that technically the Core i9-14900K/KF is a complete copy of the Core i9-13900KS, apparently, there are some settings related to power limits. As a result, the new model is still a little faster than its predecessor, although this is unlikely to be noticeable in practice. The results of the new Intel CPU have appeared in PassMark But in multi-threaded mode, which is much more important from a practical point of view, everything is more interesting: the new product differs vanishingly little in performance from the Core i9-13900K and is even inferior to the Core i9-13900KS. And all of them are noticeably behind the Ryzen 9 7950X3D.

[phonemantra-blog]

They will be released at the end of the year Intel has announced that Meteor Lake processors will be released at the end of December. At the same time, Intel nominally announced these CPUs. We have not yet been told about the parameters and models, but we have been given a more complete overview of the architectures, features, and technologies. [caption id="attachment_57388" align="aligncenter" width="701"] Intel Core processors[/caption] Intel says the release of Meteor Lake marks the start of a new era for the company as it moves away from the Intel 7 node and begins deploying its 3D EUV lithograph Foveros packaging for the client mobile platform. Meteor Lake is an extremely important product for Intel, and at the same time, it is new in literally everything. This is the new Intel 4 technical process, the chipset layout used by the company for the first time in this segment, and a disaggregated configuration, which means the use of chipsets created using different technical processes. Let us remember that the Intel 4 process technology will be used for the processor chipset, and the GPU is generally produced by TSMC. 15th generation Intel Core processors introduced. Meteor Lake, like the previous several generations of Intel CPUs, contains in the configuration large P-cores, in this case based on the Redwood Cove architecture, and small E-cores on the Crestmont architecture. An unusual feature is the two small cores integrated into the SoC chipset, which Intel calls a Low Power Island. These new LP E-cores are designed to allow light workloads and processes to be offloaded from a more power-hungry compute cluster and moved entirely to a more power-efficient cluster in the SoC. That is these two cores will be used only separately for non-resource-intensive tasks, although this remains to be verified. Other important additions include Intel's first Neural Processing Unit (NPU). Another interesting feature is that each chipset, or tile as Intel itself calls it, has its power controller, which should make the CPU more energy efficient. There are four tiles in total: CPU, GPU, SoC, and I/O tile. However, Intel only produces the CPU tile and the I/O tile. The GPU is manufactured at TSMC facilities using the N5 process technology, and the SoC tile is produced using the N6 process technology. The company also says that the new CPUs have better instructions executed per clock than the current line, but there are no details. Also, Intel has not yet said anything about cache memory sizes. Let's remember that Intel says that Meteor Lake processors will be released on December 14. But these are mobile CPUs, so it’s not very clear in what form they will hit the market that day. It may be possible to buy a laptop with such a processor only at the beginning of 2024.

[phonemantra-blog]

Initially, only K and KF models will appear The Videocardz resource reports that its sources have revealed the launch date of the new Intel Raptor Lake Refresh processors. The advertising embargo will be lifted on October 16, meaning on this day Intel partners will be able to disclose various details about their products related to the new CPUs. On October 17, at 16:00 Moscow time, the embargo on reviews will be lifted. Sales start on the same day. [caption id="attachment_54848" align="alignnone" width="780"] Intel processors[/caption] New and old Intel processors will be released on October 17, but not all It is worth noting that these dates only apply to the top-end Core i9-14900K, Core 7-14700K, and Core i5-14600K models, as well as their F versions with the iGPU disabled. That is, as almost always in recent years, the top models of the new generation will hit the market first, and everyone else will have to wait. However, let us remind you that in the Raptor Lake Refresh line, only Core i7 models will offer something conditionally new, which will have more cores than the 13th generation Core i7. In all other cases, we will only talk about increasing frequencies.

[phonemantra-blog]

Frequency reaches 6GHz The Core i9-14900K processor lit up in the CPU-Z test, which confirms its characteristics, and at the same time allows you to compare the new product with its predecessor. [caption id="attachment_51251" align="aligncenter" width="780"] new Intel CPU[/caption] The new CPU will indeed be an almost complete copy of the Core i9-13900K. The same 24 cores, the same amount of cache memory, but the frequency will increase to 6 GHz, that is, by 200 MHz.  Not 3%, but as much as 8-10% of the increase in the new Intel CPU. In the benchmark, the CPU scores 978 and 18,118 points in single-threaded and multi-threaded modes, respectively. Thus, the new product is 8-10% faster than its predecessor. This is not bad, considering the almost identical parameters, but you need to understand that in real applications the gain may be less. But relative to the Core i9-12900K, the increase reaches a much more impressive 19% and 59%, depending on the mode. Let us remind you that the Raptor Lake Refresh line is expected in October.   [caption id="attachment_51254" align="aligncenter" width="780"] new Intel CPU[/caption]