Processors - Page 62

Tachyum unveiled their new Prodigy CPUs not too long ago, with some bonkers specifications that seem far too crazy as the company is getting closer to the launch of some of its Prodigy CPUs that are now in pre-order status for evaluation kits.

The flagship Tachyum Prodigy T16128-AIX CPU will reportedly pack up to 128 cores, crazy 5.7GHz CPU clocks, just-as-insane 16-channel DDR5-7200 memory support, and once-again-insane 950W TDP. The 128 custom-designed 64-bit cores will be running at 5.7GHz+ according to the company, made on the 5nm process node and support 1-2TB/sec of memory bandwidth depending on the setup.

Tachyum says that it will be sampling the 128-core CPU "end of 2022". There is also hardware coherency that supports both 2 and 4-socket systems for some truly crazy performance... there are a lot of performance claims here, where Tachyum uses NVIDIA's new Hopper-based H100 GPU in its benchmark comparisons.

Apple just announced its new M2 chip, with 20+ billion transistors on TSMC's still fresh 5nm process node, while it is testing its next-gen M3 chip in-house for a new iMac of the future.

In a new article, Bloomberg's Mark Gurman reported that: "the company is already working on an M3 chip destined for a future iMac and other products". There are no other details of what Apple will pack into the M3, but I'd like to see something truly beefy for a next-gen Apple iMac desktop system.

Rumors began swirling of the new Apple M3 chip in April 2022, where Bloomberg began teasing the M3 for a future iMac. Bloomberg reported at the time: "Since then, I've heard that the M2 chips aren't the only one in testing within Apple. And if you're waiting for a new iMac, i'm hearing an M3 version of that desktop is already in the works - though I imagine it won't launch until the end of next year at the earliest. Also, for those asking, I still think an iMac Pro is coming. It just won't be anytime soon".

Apple unveiled its new M2 processor at its recent WWDC 2022 event, with detailed specifications seeing it pack an 8-core CPU and 10-core GPU on TSMC's 5nm process node.

We quickly heard rumors on Apple's beefier M2 Max SoC but all we heard is that it would be made on TSMC's next-gen 3nm node... and now, we have rumors on the M2 Max specs. We're looking at the M2 Max packing a 12-core GPU (up from the 10-core CPU on the plain M2) while M2 Max reportedly packs a beefier 38-core GPU (up from the 10-core GPU on the plain M2).

Bloomberg detailed that the M2 Max chips in the new high-end MacBook Pro laptops will "include 12 main processing cores and up to 38 graphics cores, up from a maximum of 10 processing cores and 32 graphics cores in the current models".

AMD has confirmed a bunch of its next-gen GPU architectures at its recent Financial Analyst Day 2022, with details on RDNA 3 and RDNA 4 architectures... as well as the world's first exascale APU: CDNA 3.

AMD's next-gen CDNA 3 architecture will be the first exascale APU design, packing both CPU and GPU chiplets on the same package. AMD has promised an insane 5x or higher (AI) performance-per-watt uplift for its new CDNA 3 architecture over the current CDNA 2 architecture.

The company said that the expected performance-per-watt uplift for CDNA 3 will be achieved through the new 5nm process and 3D chiplet packaging, the next-gen AMD Infinity Cache, 4th Gen Infinity architecture, unified memory APU architecture, and new math formats. A huge +500% increase... but the chart does note "AI performance/watt uplift".

AMD has been very confident with its rollout of its Zen-based CPU architecture, but now the company has better detailed its CPU core roadmap at its recent AMD Financial Day 2022 event.

AMD has said that it will have both Zen 4 and Zen 5 architectures on the market in 2024, with Zen 4 + Zen 4 with V-Cache + Zen 4c CPUs made on TSMC 5nm and 4nm process nodes this year, and into 2023. But in 2023 and into 2024 and beyond, AMD will introduce its next-gen Zen 5 CPU architecture with Zen 5 + Zen 5 with V-Cache + Zen 5c CPUs made TSMC 4nm and TSMC 3nm process nodes.

Zen 5 is a "new grounds-up microarchitecture" explains AMD, saying that it is "optimized for scale across workloads". AMD's new Zen 5 architecture will have "enhanced performance and efficiency", a "re-pipelined front end and wide issue" and "integrated AI and Machine Learning optimizations".

AMD revealed some more details about its upcoming 4th Gen EPYC "Genoa-X" CPU family at its Financial Day 2022 event, and boy, it packs some beef.

The new AMD EPYC "Genoa-X" processor is based on the new Zen 4 architecture, packing the new AMD 3D V-Cache technology... driving up to over 1GB of L3 cache which is pure cache insanity. Even the kick ass EPYC "Milan-X" CPU has 768MB of L3 cache and that was great, but 1GB+ is a huge step up for AMD.

AMD will be using the monster new SP5 socket (LGA 6096) for its EPYC "Genoa-X" CPU, offering both DDR5 support and new PCIe 5.0 technology. AMD has also now confirmed up to 96 cores based on the Zen 4 architecture, and that it's "optimized for technical computing and databases" and drops "in 2023". We already know that Meta (Facebook) will be using EPYC CPUs for the metaverse.

AMD's new Zen 4-based Ryzen 7000 series CPU has fallen into the hands of an "unnamed overclocker" and been instantly delidded. Check it out:

TechPowerUP reported the image, noting that the "person who shared this picture should most likely not have done so and as such, we won't be posting a link to the source". The photo shows us the IHS that is pretty damn thick, but you've also got the two CCDs and the IOD soldered to it.

The two CCDs in question are part of a two-chiplet design: each chiplet packs up to 8 x Zen 4 CPU cores (16 cores in total for up to 16 cores and 32 threads in the flagship Zen 4 processor). The I/O die is also new, with an integrated RDNA 2 GPU, and made on TSMC's fresh 6nm process node.

Intel's next-gen Xeon "Sapphire Rapids" CPUs have been delayed, where the original plans that Intel had for its new Xeon volume ramp hitting a big bump in the road.

This isn't the first delay, but now the second that Intel is putting Sapphire Rapids through since it was meant to debut in Q1 2021 this year, with Intel shipping initial SKUs to customers a few months ago and they're still being validated. But now, Intel is building in more platform and product validation time, with a delayed production ramp deeper into 2022.

Intel Executive Vice President, General Manager, Datacenter and AI Group, Sandra Rivera explains: "At this point we are building in more platform and product validation time, so we see Sapphire, you know the ramp being later in the year than what we had originally forecasted, but the demand is still very high".

Apple unveiled its next-gen M2 at its recent WWDC 2022 event, offering some upgrades over its M1 predecessor, but now it's time for M2 Pro and M2 Max rumors.

Apple's upgraded M2 Pro and M2 Max SoCs will be upgraded versions of the M2, and act as next-gen replacements to the M1 Pro and M1 Max SoCs, leaving a hole for the M2 Ultra afterwards to replace the current-gen M1 Ultra replacement (which is effectively 2 x M1 Max SoCs glued together).

9to5Mac is reporting that Apple will "continue with TSMC as its Apple Silicon chip supplier. The Taiwanese semiconductor company is expected to start mass production of Apple's new "M2 Pro" chip later this year, which reportedly will be built on the 3-nanometer process". The difference here is that the upcoming M2 is being made on second-gen 5nm at TSMC... while the beefed-up M2 Pro and M2 Max will be made on the next-gen 3nm process node at TSMC.

Apple has just announced its new M2 SoC with some impressive upgrades across the board, and will be powering "completely redesigned" versions of the MacBook Air and 13-inch MacBook Pro laptops.

The new Apple M2 is being made one TSMC's new second-generation 5nm process node, with over 20 billion transistors making up the chip. Inside, Apple has an 8-core CPU, 10-core GPU, up to 24GB of LPDDR5 RAM with 100GB/sec of memory bandwidth, and "industry-leading performance per watt".

Apple also has a "high-performance media engine" that handles up to 8K resolution video and HEVC video, and even supports 6K external display support.

We have just had a fresh rumor from Benchlife, that Intel's next-gen Socket V1 -- aka LGA 1851 -- will be the heart and soul of the next-gen Meteor Lake and Arrow Lake CPUs in 2023, 2024 and beyond.

Intel's new 14th Gen Core "Meteor Lake" CPUs were rumored to use a new LGA 2551 socket, which we learned about in Tom @ Moore's Law is Dead's latest video... but now Benchlife is saying that's wrong and that Chipzilla will use LGA 1851 instead.

The new LGA 1851 socket measures in at 45 x 37.5mm making it identical -- in size at least -- to the LGA 1700 that the 12th Gen Core "Alder Lake" uses, and the upcoming 13th Gen Core "Raptor Lake" CPU will use. This would be great for CPU cooling companies, as you wouldn't need to buy a new CPU cooler with the new LGA 1800 socket and new 14th or 15th Gen Core CPU of the future.

Intel's next family of CPUs that drops later this year is the new 13th Gen Core "Raptor Lake" CPUs, but after that we have the 14th Gen Core "Meteor Lake" CPUs that reportedly come on a new socket.

The new Intel 14th Gen Core "Meteor Lake" CPUs will reportedly have a new LGA 2551 socket, measuring in at 38mm x 46mm with the new socket being "only slightly bigger in actual footprint compared to Alder Lake's LGA 1700" says leaker Tom @ Moore's Law is Dead.

Intel's new LGA 2551 socket wouldn't be much bigger than LGA 1700, but it craps all over the LGA 1700 socket in terms of pin count: 50% more pins. MLID also posted a real photo of an alleged LGA 2551 processor, so we can see those 2551 pins in all their glory.

We are being teased with 15-21% increases in IPC performance with the new Redwood Cove architecture over the Raptor Cove architecture, but with MLID apologizing for the huge range (15-21%) but I think that's fine. 15% IPC is better than 5-10% but more than that will be welcomed.

Meteor Lake CPUs will have lower clock speeds, but you'll have more performance for the same clocks with the new CPUs over Alder Lake, hence the 15-21% boost in IPC performance. MLID adds that "Crestmont is the new Little Core architecture, but I can't confirm its performance or core count yet" and to also "expect integrated accelerators like a VPU or Neural Engine, and more".

As for core counts, MLID teases that we'll get "at least 2+8 and 6+8 SKUs are confirmed for U/P mobile" while the "desktop is expected to be 8+16 but that's not 100% confirmed yet".

Intel's new 13th Gen Core "Meteor Lake" CPUs will debut later this year, but have been spotted again... this time, next to an Intel Arc desktop GPU.

The CPU that was tested was discovered by TUM_APISAK with its increased 24 cores and 32 threads of Raptor Lake CPU power, with 32GB of DDR5-4800 memory, and an unreleased Arc Alchemist a770 desktop GPU. You'll notice that the Intel Arc a770 GPU only has 1GB of VRAM, but it's a GPU that doesn't exist right now, so it's not being detected correctly.

Intel's new 13th Gen Core "Raptor Lake" CPU with its 24C/32T of processing power was spotted in engineering sample (ES) form with a 2.4GHz base clock, and 4.6GHz boost clock. We should expect more concrete details on frequency in the coming days and weeks, with Intel's current-gen 12th Gen Core "Alder Lake" CPUs seeing much higher (5GHz+) frequencies, and AMD's upcoming Zen 4-powered Ryzen 7000 series CPUs are going to be pushing past 5.5GHz+ without an issue when they debut later this year.

AMD's new 600-series chipsets will launch later this year, ready to handle the world of Zen 4 processors -- but just as excitingly -- DDR5 and PCIe 5.0 technology.

But with all these new features and technologies, normally it costs more... but it seems we might not be paying too much for B650 and X670 motherboards. In a new tweet, Tom @ Moore's Law is Dead teased: "I can now confirm two Prom21 chipsets cost less than one X570 chipset. AM5 platforms were designed to be cost-effective".

"Now, there are components that CAN make X670E more costly than X570, but there's little reason for B650 boards w/ PCIe 4.0 to cost more than B550". It looks like a new B650-based motherboard is going to be a monster value for money board, especially if you mixed it with a new Zen 4-based Ryzen 5 or Ryzen 7 series CPU.

Russia has now been locked out of buying any electronic components from Taiwanese companies that are over 25MHz, which is pretty much all modern electronics.

The United States, United Kingtom, and the European Union slapped Russia over the last few months with restrictions that saw the largest companies in Taiwan being the first to cut ties with Russia. Now... well now we have Taiwan's Ministry of Economic Affairs (MOEA) formally publishing a list of products banned from being exported to Russia and Belarus.

The new list includes virtually all Taiwan-produced high-tech devices -- think anything fabbed by TSMC or any Taiwan-based company and inside of any Apple, AMD, NVIDIA, or Qualcomm device -- as well as the tools used to make the chips themselves, from being exported to Russia or Belarus. Here are the restrictions from Taiwan:

Intel's new scalable chip design that uses x86 CPU cores and Xe GPU cores for supercomputing workloads, the new Falcon Shores XPU, has been detailed a little more at the International Supercomputing Conference (ISC 2022) in Germany.

The new Intel Falcon Shores XPU is described as a "New Tile-Based Flexible & Scalable Architecture" by the company, with three different configurations shown off: a completely x86 Tiled solution, an Xe GPU Tiled solution, and another that glues both x86 CPU cores and Xe GPU cores together on a single socket.

Intel is going with at least four Tiles on all of its Falcon Shores XPUs which matches the Tile layout of the Intel Sapphire Rapids Xeon CPU family, but then Intel is making some huge performance claims here. Falcon Shores XPUs will be delivering 5x the performance/watt, 5x the Compute density in x86 socket, and 5x memory capacity and bandwidth.

Apple's next-gen iPhone 14 will be released later this year, and inside we're looking at the new A16 SoC that will be made on TSMC's new 5nm process node according to industry analyst Ming-Chi Kuo.

In a flurry of new tweets, Kuo says that Apple's new A16 SoC will continue using TSMC's N5P process node -- which is the same as the current-gen A15 SoC. The reason? TSMC can't make enough chips on its next-gen N3 node this year in order to meet Apple's insatiable demand, while N4P is an improved variant of N5, going into mass production later this year.

Kuo states that the "significantly better N3 & N4P won't start mass production until 2023" and that "N5P & N4 are the latest technologies Apple can use" for their products in the second half of 2022, finishing off the first tweet with "N4 has no advantages vs. N5P".

AMD's new Mendocino APU specs have been leaked out, which will be the basis of future entry-level Athlon + Ryzen 5-powered laptops later this year.

The upcoming AMD Mendocino APU will reportedly pack a 4-core, 8-thread Zen 2-based CPU with RDNA 2 GPU cores, with the APU made on TSMC's fresh 6nm process node. AMD's new Mendocino APU is already powering AYANEO's new Air Plus handheld console, which launches later this year.

AMD is updating its Zen 2 CPU and RDNA 2 GPU cores in the Mendocino APU, optimized on the TSMC 6nm node, and adding 4MB of L3 cache. We're expecting up to 10 hours of battery life, with active cooling inside of future laptops expected as passive designs on Mendocino requires more engineering... and would increase the cost of the laptop to the consumer.

AMD has been out in a not-so-panicked damage control post-Computex 2022, where expectations of Zen 4 processors are kind of all over the place... until now. AMD is clearing things up, and now one of the biggest leakers of Zen 4 content is out with a monster new 40-minute video.

We've been hearing all sorts of crazy rumors about IPC performance on Zen 4, where all the way back in February 2021 we were fielding "around 40% faster" IPC performance over Zen 3 for the new Zen 4 architecture. AMD has said it'll be over 15%, but we know it'll be higher than that.

In his latest video, Tom @ Moore's Law is Dead says (beautifully): "other leakers have apparently been saying that it's (IPC) gonna be above 25%, 30%, 40% IPC and it's just fucking NOT gonna be that high, and so if AMD looks around and sees some of these fake leakers still get a lot of time on websites. They do not want people to be disappointed by say a 20-30% higher single-threading increase on high-end SKUs, a 15% or higher or higher single-threading increase on lower-end SKUs... they want you to be excited by that because that is a big gen-over-gen increase".