CPU, APU & Chipsets News - Page 2
CES 2020 - Intel has come out swinging at CES 2020 this year, revealing its new 10th-gen Comet Lake-H family of CPUs that will continue on the 14nm+++ node, but will pack architectural improvements the company has made since Skylake.
Intel has something truly exciting on offer with its 10th-gen Comet Lake-H family of CPUs, with a new flagship chip that will offer up 8C/16T at over 5GHz. There will be Core i5, Core i7, and Core i9 models that will feature 4C/8T, 6C/12T, and 8C/16T respectively.
The company teased just before CES 2020 kicks off that its new Core i9 series processors will offer the full 8 cores and 16 threads at over 5GHz, all inside of a notebook. This will come down to the thermal design and cooling technology used by gaming notebooks like ASUS, Razer, and others -- but it will be powered by Intel's own Thermal Velocity Boost technology that will offer crazy boost CPU clocks when possible.
Samsung has kicked off the new decade in a big way, with the South Korean giant making major progress in its pursuits to become the #1 semiconductor manufacturer by 2030. Samsung has just made a prototype 3nm process using GAAFET technology.
Gate All Around (GAAFET) technology is different to the widely-used and known FinFET standard, with GAAFET seeing a reduction in the total silicon size by around 35%, while consuming 50% less power. This blend can see a 33% performance increase over the still-yet-released 5nm FinFET process.
The design of GAAFET is much different to Planar FET and FinFET designs, with 4 Gates on channel compared to just 1 Gate and 3 Gates, respectively. This means power leakage is kept to a minimum, with control over the channel improved the shrinking of the node process becomes easier. This paves the way for more efficient transistor design, even at its teeny-tiny size -- ushering in a huge performance-per-watt jump over the 5nm FinFET process.
Oh, how the tables have turned -- Intel has no CPUs that are capable of supporting PCIe 4.0 technology, so they need to use AMD Ryzen 3000 series CPUs along with an X570 motherboard to test their latest storage products.
Intel technical marketing performance engineer, Frank Ober, tweeted (above) that Intel can send PCIe 4.0-capable SSDs to developers, but they'll need PCIe 4.0-capable CPUs and motherboards to test them. The super-fast new Alder Stream SSDs (an updated version of their Optane drive tech, with second-gen 3D XPoint technology).
But in order to ramp up those speeds they will need more PCIe lanes, and since PCIe 3.0 is pretty much tapped out -- the doubling in bandwidth to PCIe 4.0 is a tasty offering. But, Intel has no PCIe 4.0 anything right now -- so they're stuck. Intel won't have PCIe 4.0-capable CPUs until 2021 which means they need to lean on their main competitor in AMD until then.
In a world of custom and AIO coolers in all shapes and sizes, it blows my mind to be writing about a beasty 32C/64T AMD EPYC processor and NVIDIA GeForce RTX 2070 graphics card being passively cooled -- yeah, fanless.
AnandTech has the scoop on TureMetal's new system that packs an AMD EPYC 7551 processor (32C/64T) with its own 180W TDP, joined by GIGABYTE's own GeForce RTX 2070 graphics card with a 175W TDP, meaning the entire system has 355W between the CPU and GPU alone and yet it is passively cooled.
TureMetal uses a Supermicro ATX motherboard, and then builds it inside of its in-house UP10 chassis. The chassis has support for a 140W CPU and 160W GPU for 300W total, but with some adjustments the company was able to cram in the 32C/64T processor and Turing-powered GeForce RTX 2070 graphics card.
AMD has just had its stock reach records highs to kick 2020 off in a big way, and while the company has already officially teased it will push the high-performance computing world at CES 2020 -- the company could unveil its next-gen Zen 3 architecture.
New rumors are surfacing after a Taiwanese newspaper published a report that AMD CEO Lisa Su would "disclose details about Zen 3 architecture at a press conference during CES". We reported back in October 2019 that AMD's new Zen 3 architecture could bring up to and over 8% IPC gains, and another 200MHz as well.
AMD's new Zen 3 architecture will power the next wave of CPUs from AMD in 2020 and 2012, with the new EPYC, Ryzen Threadripper 4000 family, and new Ryzen 4000 series family. The new EPYC is codenamed Milan, while Ryzen Threadripper 4000 goes by Genesis Peak and Ryzen 4000 series with Vermeer. AMD will be making the new Zen 3-based EPYC, Ryzen Threadripper 4000 series and Ryzen 4000 series CPUs on the 7nm+ node.
Intel is preparing its new range of 10th Gen CPUs for 2020, with VideoCardz tweeting out some juicy information on the upcoming Intel Core i9-10900K processor and its clock speeds.
According to this tweet from VideoCardz the Intel Core i9-10900K will have a base CPU clock of 3.7GHz, single turbo of 5.1GHz, max turbo boost of 5.2GHz and a mysterious new 'velocity boost' of 5.3GHz. The new velocity boost is something that isn't quite that new -- but it will be for most people.
Intel has used Thermal Velocity Boost (TVB) previously in notebook CPU designs, but it seems it could be skipping the silicon pond into the desktop CPU realm with its new Core i9-10900K (and I'm sure, other 10th Gen CPU designs). If the tweet from VideoCardz is correct, the Thermal Velocity Boost on the Core i9-10900K will be 5.3GHz.
Intel has its Xe GPU still cooking in the oven, as it does with its upcoming Tiger Lake CPUs -- but now we have some more details on the Xe DG1 GPU itself.
VideoCardz is reporting the latest leak on Intel's new Xe DG1 GPU from a leak from the EEC, which shows us that DG1 could sport 96 Execution Units, and up to 768 shading units. This would mean Xe DG1 is a very entry-level graphics solution, something along the lines of integrated graphics now -- or up to a GeForce GT 1030 (yeah, the one without the X).
We are to expect Intel's Tiger Lake to roll out with 96 EUs based on Gen12 Xe graphics, compared to the 72 EUs on Skylake, 48 EUs on Broadwell, and 40 EUs on Haswell. We should expect more news on Xe and DG1 at CES 2020, which is right around the corner.
AMD has been on the fresh new 7nm node for a few months now, pumping out a bunch of Zen 2-based EPYC, Ryzen Threadripper, and Ryzen CPUs -- as well as a slew of new Navi-based Radeon 5000 series cards. Intel has struggled, to say the least, to get to 10nm let alone 7nm.
But now it seems there will be another competitor in 7nm space with Chinese company Zhaoxin its new KX-7000 processor coming up next, using an unannounced CPU architecture and built on the 7nm node with an integrated GPU. The new Zhaoxin CPU will have a GPU that has DX12 support, and features PCIe 4.0 and DDR5 technology support.
Zhaoxin has its KaiSheng series CPU which rocks up to 32 cores, and is made on the older 16nm node... but the company has plans to offer up to 32C/64T on its new server CPU offerings. This will bring Zhaoxin up to the same level as AMD's current Ryzen Threadripper CPUs, leaving AMD to dominate the server CPU game with its Zen 2-based EPYC range of processors.
Intel is years behind in the CPU node game allowing AMD to rip right past and take over with its Zen CPU architecture firmly placed into the leadership position on 7nm -- but it doesn't mean Intel will be taking this lying down.
Dr. Ian Cutress of AnandTech was on the ground at the IEEE International Electron Devices Meeting (IEDM) where it was revealed what Intel plans over the next 10 years. In the slide you can see above, Intel has mapped out the next 10 years that starts with its (continued struggles) on 10nm while seeing 7nm CPUs drop in 2021.
But we can also see that Intel is aiming for 5nm in 2023, 3nm in 2025, 2nm in 2027, and finally -- if you can even believe it, 1.4nm CPUs in 2029. Intel is expecting a two year flow between CPUs, even though it has been stuck on 14nm for what feels like forever now. Just how small is 1.4nm though? Cutress writes that it is the "equivalent of 12 silicon atoms across". Incredible stuff.
With the heating up of the CPU race, one thing has been touted among PC enthusiasts forum and social media users far and wide. "The death of Moores Law is coming and will limit how far we can go now" or some derivative of this.
For those uninitiated, here is a simple explanation of what Moores law is, from Intel's co-founder, engineer, and the namesake of Moores Law itself, Gordon Moore. "the number of transistors incorporated in a chip will approximately double every 24 months."
So, basically, the density of a chip as far as transistor count should double roughly every two years along with refining and shrinking the package to make these substantially more compact packages smaller while offering up to double the amount of processing muscle.