https://prog.world/amd-epyc-genoa-on-zen-4-architecture-as-early-as-2022-and-...


NOW LET'S THNK MAINFRAME FOR A BIT, without getting it confused with consumer or gaming in our heads.  This is useful to us as it predicts more exactly what AMD will be doing and when it will be doing it as the tech needs to be built in at the chiplet level to be there for mainframe uses when the time comes.
 
AMD EPYC GENOA on Zen 4 architecture as early as late 2021 and new development rumors about SMT4 multi-treading and the AM5 processor socket

AMD has just announced the custom Ryzen series processors on its new Zen 3 architecture and is still preparing the corresponding server announcements, when information about the next generation of the company’s server processors – AMD EPYC GENOA on Zen 4 – appeared on the network.



We are talking not only about processors with up to 96 cores, but also about new generation sockets AM5, TR5 (HEDT platform), as well as SP5 and SP6 (server platform). It also talks about support for PCI-Express 5.0 and DDR5 memory. While the information circulates on the network at the level of rumors, but given the way AMD’s consumer segment is developing, their attempt to regain ground in the server market is a matter of time. So, EPYC MILAN on Zen 3, according to inside information, will be at least 20% more productive than the previous generation of AMD processors of the Zen 2 ROME architecture.

Now let's carefully shift our minds to a mainframe point of view concerning these things, and REMEMBER please that all of these advancements will have to be built into the same base chiplets that get used everywhere in the AMD product line.

If we talk about AM4 vs AM5 sockets, then it becomes clear from rumors that AMD has squeezed everything it could out of the AM4 platform as such AMD has made a huge gift to consumers without updating the socket for Ryzen 5xxx while ensuring backward compatibility of new 5xxx desktop processors with an existing socket. Here you can recall the endless changes to sockets at Intel, of which at least three have been released over the past four years: LGA 1151, LGA 1151 v2 and LGA 1200.

In 2022, however, A4 will be replaced by the A5 socket and, hopefully, the A5 will live as long as the A4 did.   We know that the large server sockets will get updated: we will switch from Socket SP4 and Socket SP4r2 to SP5 and SP6. Most likely, both sockets will be released at the same time and will be suitable for the same generation of EPYC GENOA with the same difference as the SP4 and SP4r2 sockets: the former are designed for single-threaded processors, and the latter are for dual-threaded EPYC ROME processors. If we assume that AMD still implements SMT4, then SP4 will work, respectively, with single-threaded and dual-threaded processors, and SP4r2 – with four-threaded models.

But it’s worth returning to the topic of SMT.    A possible path to further AMD's leadership is to increase the number of threads per core…...    We are talking about the simultaneous processing of four threads, instead of the standard two threads per core. It should be noted that SMT4 technology is almost guaranteed not to be present in EPYC MILAN processors, but will likely be present in the follow on EPYC GENOA processors in 2022.

You need to understand that in the home use segment of the market SMT4 is a dubious feature that will only tend to increase latency and create minor timing losses out of the blue while the processor decides on which of one out of these many threads to transfer the next task, out of the dozens and hundreds of threads that various applications will create in real time.  Deciding takes time, sometimes more time than is saved by using the extra threads.

However, in the case of the pure mainframe processor segment, there we are dealing with more simplified more monolithic systems, which, more often than not, are sharpened for working with some relatively few heavy software complexes. This is where SMT4 can do very well, especially in mainframe computing.

At the same time, SMT4 and even SMT8 are actually a set of ancient systems that were used starting over twenty years ago. The development of multi-threading peaked in the early 2010s, when IBM still had its own production of servers for business.

The IBM POWER8 Core had variable multithreading, ranging from one core to one thread to eight threads per logical processor core. The official POWER8 Core frequencies per core range from 2.5 GHz to 5 GHz. At the same time, IBM servers on POWER8 also had 16 sockets SMP (symmetric multiprocessing) – which made it possible even then to combine a dozen servers into a computing cluster.

It is worth noting that IBM servers were a very specific and narrow solution for large corporate business and scientific computing. Actually, with the growth of AWS and Azure, they were squeezed out of this segment and IBM Power S822LC became the last product of the company in this line.

I must say that now, Intel’s Xeon processors, which have practically captured the server market, also do not work with SMT4 mode. If we are talking about processors for “science” – that is, about monstrous solutions for 32-72 cores of the Phi series, for example, Intel Xeon Phi Processor 7295 with 72 cores and a cost of ~ $ 6200 at the time of release, then we do not have multithreading at all … According to the official specification, this processor has 72 cores and 72 threads.

The more popular Intel Xeon E operate in SMT2 mode – two threads per core. This applies to almost all popular Intel server processors released since 2013, starting with the E5-V2 series. To give a concrete example, two threads have already been in the extremely popular workhorse in the form of the Intel Xeon E5-2680V2 processor, which is still actively used.

If the forecasts and data from insiders are confirmed, then in the next five years we can expect a partial redistribution of the server market. AMD is actively developing in the desktop direction, now it is starting to put pressure on the server segment. Intel plans to announce server processors this spring, but it is not a fact that they will be much more powerful than the new MILAN, and even more so the upcoming GENOA. “Blues” are still experiencing significant difficulties with lithography and cannot really go below the 14 nm process technology, somehow launching the production of 10 nm laptop processors.

It's not hard to figure it out, Samsung builds it for them right now because Intel can't build it at all ......




Why pay attention to these mainframe leakers ????

Because mainframe people don't care about PC gaming or general home use functionality and they simply don't see their mainframe world as being the feature predictor of the entire AMD stack.

Because of their long lead times for decision making, AMD gives future info out to the mainframe folks first, giving it to them far enough out for them to be able to use it in their longer extended planning horizons.

Is this information reliable?   Yes, the info is VERY reliable --- lie to the mainframe elephants just once and they will NEVER forget it, so AMD is careful to only tell them the actual factual.

Also please note that the integration of Xilinx into AMD chipsets would also take place in this same time frame, but this is not being revealed or discussed at all by AMD until the sale has full approval by the FTC and it is actually a done deal.   Being a simple stock swap arrangement inside the USA, involvement by other countries is not required for the ARM Xilinx arrangement.

https://www.tomshardware.com/news/samsung-hbm2-hbm-pim-memory-tflops

https://www.zdnet.com/article/samsung-develops-high-bandwidth-memory-with-int...

As Intel struggles to regain relevance, Samsung and AMD-Xilinx and several others strike out in new directions that will render traditional Intel x86 completely nonsensical.

Today, Samsung announced that its new HBM2-based memory has an integrated AI processor that can push out (up to) 1.2 TFLOPS of embedded computing power per chip, allowing the memory chip itself to perform operations that are usually reserved for CPUs, GPUs, ASICs, or FPGAs.

The new HBM-PIM (processing-in-memory) chips inject an AI engine inside each memory bank, thus offloading processing operations to the HBM itself. The new class of memory is designed to alleviate the burden of moving data between memory and processors, which is often more expensive in terms of power consumption and time than the actual compute operations.

Samsung says that, when applied to its existing HBM2 Aquabolt memory, the tech can deliver twice the system performance while reducing energy consumption by more than 70%. The company also claims that the new memory doesn't require any software or hardware changes (including to the memory controllers), thus enabling a faster time to market for early adopters.

Samsung says the memory is already under trials in AI accelerators with leading AI solutions providers. The company expects all validations to be completed in the first half of this year, marking a speedy path to market.

OK, think of all the flexible processing techniques that are already in use.  Now think about all the additional layers in your 14 layer 5nm processor stacks, plus all the GREATLY increased flat space per layer that will exist at the 20 layer 3nm stacked lithography.

Now partner up Samsung with AMD-Xilinx and use all that flat 3nm 20 layer stack space to put a very strong AI processor actually inside your existing memory lay outs .......

This is sensible stuff that is already in trials right now, it is REAL STUFF that is ready to roll out right now.   Why have rack after rack of Nvidia video cards burning up the watts trying to do your AI stuff when you can do this right inside your systems memory at far far less energy cost?

This changes the game up quite a bit for those innovative and flexible enough to go for it, right now.



This is a 1990 ASCI Red, the very first Intel x86 supercomputer to go over a teraflop of processing power. Each stacked chipset of the new Samsung HBM2-based memory has 1.2 of these babies functionally tucked inside it, except that the new Samsung AI chipset is a whole lot quicker off the mark and comes with software and applications that can actually USE this level of computing power.

Change, she comes ......          



Notably, the paper and slides above refer to the tech as Function-In Memory DRAM (FIMDRAM), but that was an internal codename for the technology that now carries the HBM-PIM brand name. Samsung's examples are based on a 20nm prototype chip that achieves 2.4 Gbps of throughput per pin without increasing power consumption.

The South Korean tech giant explained that this extreme progress was possible as the installation of AI engines inside each memory bank maximizes parallel processing while minimizing data movement. By providing this improved performance, Samsung said it expects the new chip to accelerate large scale processing in data centers, high-performance computing systems, and AI-enabled mobile applications.

The 3nm version of this tech next year will be a lot more capable, of course.   And consume proportionally far far less power as well.

New Chromebook OS will have more power using same chipsets, but 5x more power using special built XPU chipsets that are on the way for second half 2021.

Chromebooks have all the newest Google drivers for the full AI capabilities of the chipsets they are built with.   This includes the built in AI and graphics accelerators in the XPU or in the memory itself.    You can use Steam or Google gaming to get your gaming on with your new more powerful Chromebook .......

Chromebooks can run Chrome apps, Android apps and Linux Apps along with any web based app set out there (Microsoft is included in this mix for free).

Chromebooks always cost less (simple, Chrome OS requires less hardware to run well).

Yep, Eegore Chromebooks run EVERYTHING other than formal Microsoft products.    And it is funny, the newer Chromebooks can run MS products off the web and can run full package installs of the MS Office for Chrome variants,  too, but nobody wants to pay the Mickey Tax for that sort of action.

Chromebook people are generally already really good with Google Docs and with Libre Office Suite, both which are excellent FREE alternatives to Microsoft Word, Excel, etc.

https://liliputing.com/2021/02/samsung-could-launch-a-windows-pc-with-exynos-...

Samsung could launch a Windows PC with Exynos CPU and AMD graphics this year



ZDNet says the processor, which might be known as the Exynos 2200, should be ready in third quarter of 2021, which is when Samsung plans to debut the processor in its new Windows on ARM computer, which will most likely be a thin and light laptop similar to last year’s Samsung Galaxy Book S.

It’s unclear how Samsung’s new chip will stack up against Qualcomm’s Snapdragon 8cx Gen 2 processor, which is currently the most powerful chip used in Windows 10 on ARM computers. But hopefully a bit of competition in this space will drive Qualcomm and other chip makers to step up their game in terms of performance.


===================================================


Intel's Rocket Lake midrange chipset is out and has now been benchmarked against AMD's Ryzen 7 from last year.   Intel was boning hard at getting back at least one gaming crown, and they did so at a simply astronomical cost in energy & efficiency and by using all their multiple AI tricks.

Using all Intel's AI tricks and running at a very very hot very high current draw (hyperclocking) rate, Intel managed to eek out a 2% win over AMD's bog standard no tricks no bullshite processor benchmark rating.

2% isn't much, and AMD could counter this with a very slight processor speed bump of their own lower lithography 7nm chipsets vs the Intel 10nm competition.

But AMD won't do this because the next shift they plan in their own AMD processors (to 5nm) wipes out Intel's tiny 2% lead quietly and naturally as part of AMD's 15%+ IPC improvement.

Plus, there are over 6 AMD chipset overclocking utilities out there now (3 are open source and completely free to use) that allow those users who want to go tit for tat with Intel's hyperclocking to do so freely on their own.

AMD simply isn't going to play Intel's mickey mouse tit for tat games with "over running their voltages and their clock speeds" ........


==================================================


We are looking for the substrate based shortages to last for the rest of this year, possibly into 2022 as well.

You are likely to see more severe AMD shortages showing up as everyone wants the AMD chipsets right now and Intel is always trying to get closer to competitive so they can sell their products too.  Hard fact is, that with both of our USA suppliers running as good as they can right now America is going to continue to fall short on available products ---- and even more Near Eastern ARM based Chromebooks will be sold in the USA to fill in the blanks.

https://videocardz.com/newz/amd-epyc-genoa-zen4-to-feature-96-cores-across-12...

https://videocardz.com/newz/amd-zen4-core-in-epyc-genoa-may-support-avx3-512-...



Confirmations of the latest rumors from more detailed sources about AMD 96 core Genoa.

In the last 48 hours, numerous details on AMD Zen4 in EPYC 7xx4 series codenamed “Genoa” have leaked and 24 hours later have been confirmed. We have learned that the new server processors will be 96 core and support 12 channel DDR5 memory, but we also learned it will have PCI Gen5 support. The internal configuration of the chiplets has also been leaked.

AMD Genoa will require a new SP5 socket. This socket will support CPUs with the LGA6096 package measuring 72×75.4 mm, according to another leak from Patrick Schur. The package is not a perfect square, but not as rectangular as SP3-based EPYC processors. The new package will have a lot more room for future chiplet designs, with the first being 12 chiplets plus 1 I/O die. The first generation SP5 EPYC will be limited to 96 cores though, according to a leak from ExecutableFix.

Today a part of a new slide deck has allegedly leaked on Chiphell. The slide appears to list all features of the new Zen4 cores, confirming it would have more than 64 cores and support 2 threads per core. It was previously rumored that Genoa might have 4 threads per core, but that was quickly debunked. More importantly, though, the slide would suggest that Zen4 supports AVX3-512 and BFLOAT16 instructions. Both are extensively used in deep learning computations. The AVX 512 support on Intel CPUs and accelerators one of the key advantages over the AMD EPYC series. If this leak is true, then this is about to change.

ExecutableFix, who yesterday leaked the Zen4 details, has already confirmed that AVX3-512 support is indeed coming to the EPYC series.

AMD EPYC “Genoa” based on Zen4 core design is now expected to launch in 2022. AMD is yet to unveil its Zen3 based Milan series, a successor to Zen2 based Rome.


We have confirmed that Apple is up and running on 3nm TSMC ---- this means AMD now rolls into the use of the 5nm lines that are now being freed up by Apple.

Thus the sudden waves of AMD 5nm chiplets and new 5nm AMD processors hitting Rumorville ......


      poor Intel, still 2-3 generations behind ........

https://www.notebookcheck.net/Ryzen-7000-rumors-start-flying-as-Phoenix-is-re...

MORE RUMORS FLYING, THIS TIME THEY ARE MOBILE APU RUMORS



The post reveals that Phoenix, which has been vaguely hinted at before in an alleged roadmap that simply stated “P******” for the future of AMD’s mobile parts, has the CPUID A70F00 and more tellingly will use the FP8 socket. This shows that Phoenix is a mobile processor, as the current Cezanne Ryzen 5000 mobile APUs utilize FP6. This is where a lot of rumors start flying, such as Rembrandt, the Ryzen 6000 mobile APU family based on Zen3+ microarchitecture, being the only chip series to use an FP7 socket. Furthermore, it’s also expected that Phoenix will offer DDR5 support, which is also supposedly coming with Rembrandt anyway, and that it will use the 5nm manufacturing process. There is a table below showing the rumored differences between generations.

The rumors about Ryzen 7000 Phoenix don’t stop there. It has also been suggested that PCIe 5 support could also be included, although it has been speculated it will be initially limited with how many lanes can be used. More processing cores may also be on board, with VideoCardz stating that up to 12 cores could feature on Phoenix APUs. As for the iGPU architecture, while Cezanne Ryzen 5000 uses Vega, the above-mentioned roadmap has claimed that AMD will go with RDNA2 (Navi) for Rembrandt Ryzen 6000, and it seems likely that Ryzen 7000 would also feature RDNA2 graphics technology with some additional tweaking. As for a Phoenix release timeframe, we could see it rise from the ashes by the first quarter of 2023, so we still have a whole two years full of rumors, leaks, and confirmations to help build up an accurate picture of what will be Zen 4 mobile.

I find it interesting that AMD early rumors contain relatively more real information than Intel's real product announcements actually have in them ---- something about this says that there is more real change going on with AMD than at Intel.

Realize that what has already been announced in mainframe has to already exist in the base chiplets and remember those same chiplets are used up and down the entire AMD line up ......    between the different line announcements we are actually getting a good bit of various major features being confirmed directly by AMD.


===================================================


Production crunch at TSMC seems to be easing a bit.  Some AMD products are in stock again at AMAZON and scalper prices have abated on those items that are "in stock" at AMAZON.

https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+9+5900X&id=3870

click on it to see it    https://www.cpubenchmark.net/singleThread.html    click on it to see it

Intel's new Rocket Lake supposedly edges out Ryzen on single core performance (but only on a single core test and only on Intel's favorite test of the moment (a special "single core best of the lot" test that is heavily tweeked for AI usage) and only for a short thermal duration and only on two very specific Intel sku units that were cherry picked for this very purpose).    
New Intel BS trickery is suspected (see follow on posts for analysis).

What is this data good for?   Well, it challenges AMD to go beat it, and you know they will, plus an extra 10-15% for good measure.

As will Apple's M-2 chipsets coming out later this quarter .......

But, by then Intel will have printed these "new results" on 10's of thousands of product boxes so Joe and Rita Sixpack can continue to read that Intel is still in the lead, still, whenever they go computer shopping.

So Intel's whopper marketing lies actually helps motivate AMD and Apple to build "the processor staircase to Heaven" one step height at a time.

Also, Intel likes to push out yak yak that they plan to do this and plan to do that, but the generations of stuff Intel promises never gets built as it was never real in the first place.

Meanwhile, Joe and Rita never see this stuff ........            click on each item, it is a full page on all the processors that are for sale now

https://www.cpubenchmark.net/desktop.html

https://www.cpubenchmark.net/laptop.html



===================================================



https://liliputing.com/2021/03/lilbits-intels-adaptive-boost-for-rocket-lake-...

https://www.anandtech.com/show/16564/intels-new-adaptive-boost-technology-flo...

And this, boys and girls, is why AMD provides a very good fan and fin stock cooler system for all their chipsets, and insists that all published testing is done using the stock cooler.   AMD only claims what a stock AMD rig can do ...... unlike Intel which lies out their ass all the time using a variety of tricks which include non-standard settings and custom cooling.


Lilbits: Intel’s Adaptive Boost for Rocket Lake chips


This week Intel introduced its new 11th-gen 14nm desktop processors, code-named Rocket Lake-S. Set to hit the streets at the end of the month, the new processors are manufactured using a 14nm process rather than the 10nm process the company uses for most of its recent laptop chips. But Intel is still promising a modest boost in CPU performance and bigger boosts in integrated graphics.

Intel is also promising even better performance for its most powerful Rocket Lake-S chips. The upcoming Intel Core i9-11900K and Intel Core i9-11900KF will support a new feature called Adaptive Turbo Boost, which allows some of the eight CPU cores to run at up to 5.1 GHz simultaneously, assuming they processor is running cool enough to support the feature.

AnandTech has a more detailed explanation, including some useful graphics that show how Adaptive Turbo Boost differs from regular Turbo Boost and Thermal Velocity Boost. The tech news and analysis site also already has a review of one of Intel’s less powerful Rocket Lake-S chips, the Core i7-11700K, and it seems like the processor only kind of delivers on Intel’s promises of performance improvements. It all depends on what you ask the chip to do.


Intel’s New Adaptive Boost Technology: Floating Turbo Comes to Rocket Lake [AnandTech]

Intel’s new Adaptive Boost technology will bring better multi-core performance to the highest-performance Rocket Lake-S desktop processors by enabling selected CPU cores to hit speeds as high as 5.1 GHz simultaneously under the right conditions.

Question:  how many seconds of this action before the BIOS thermal sensor built into the chipset itself shuts it down for thermal throttling (self preservation) ????





So, it is a another hyperclocking gimmick that only operates in certain circumstances (such as a carefully set up CPU benchmark test) to give out a false impression of real processor performance under those certain specific circumstances.

An Intel PC equipped with chilled water or freon cooling might be able to get all 8 cores actually up and running on this trick (aside from the fact that Intel rates the processor off the single one best performing core and the rest are always catch as catch can as far as real performance goes.

     Using them Intel "Magic Minute" tricks again I see ........   however this time they are only a few magic seconds in duration.

USING FAN AND FIN COOLING, THERE ARE CURRENTLY NO REAL WORLD TASKS THAT CAN ACTUALLY USE INTEL'S ADAPTIVE BOOST TECHNIQUES INSIDE THE VERY FEW SECONDS THAT BIOS CONTROLLED THERMAL THROTTLING WOULD PERMIT FOR THEM TO BE EFFECTIVE.

Intel Marketing BS, smoke and mirrors, lies  you pick the term you would prefer to use.