Processors - Page 97

If you are old enough, you'll remember the days of the Pentium brand being the strongest CPU in the consumer world - but now, it's been pushed into the budget world while even the Core i3 is considered a budget CPU. Intel shifts into the mid-range with its Core i5 and even slower Core i7 processors, while we have higher-end Core i7 and even the new Core i9 in the enthusiast market.

According to the latest report from DigiWorthy, Intel is effectively killing the Pentium G4560 after great reception from consumers by scaling production down that will make it harder to find in the market. This method is something I'm sure Intel is hoping will push people into the arms of the more expensive Core i3-7100 processor, which is nearly double the price of the Pentium G4560.

The only difference between the Pentium G4560 and the Core i3-7100 is 400MHz in the favor of the Core i3, but it costs twice as much... doesn't make sense, really.

AMD surprised the world with EPYC, but we all knew Intel would have a response, and that response will reportedly be in the form of the new Skylake-SP platform.

VideoCardz has the exclusive, without any comments and provided the slides on their own. The new Skylake-SP processors will arrive with up to 28C/56T, up to 38.5MB of last-level cache (LLC), and up to 48 PCIe 3.0 lanes. There will be support for 6-channel RAM with 2 x RDIMMS, LRDIMMs, or 3DS LRDIMMs at up to 2666MHz, while the TDP will float somewhere between 70W and 205W.

The new Intel C620 series chipset will have up to 14 SATA 3 ports, up to 14 x USB 3.0 ports and up to 10 x USB 3.0 ports. There will be some motherboards with up to 4 x 10GbE ports, if you need some serious networking performance, too.

Intel's new Core i9-7900X has been on the market for just a few weeks, but it has already been pushed to world record heights of 6.01GHz on liquid nitrogen, setting a new 10-core CPU and highest ever single processor world record. The hardware used:

SOFOS1990 is no stranger to breaking OC world records, hitting 6016MHz at 1.6V on the 10C/20T processor at a super-cold -110C (-116F). Sofos used GIGABYTE's Aorus X299 SOC Champion motherboard, alongside the enthusiast grade Corsair AX1500i PSU.

1.6V is pretty damn high for voltage, and no processor will withstand that for hours on end let alone daily use. For overclocking purposes, it's fine - but that's where the super-chilly LN2 comes into play.

AMD is advanced towards being the first to market with HBM2 on a consumer graphics card with the upcoming Radeon RX Vega, placing the VRAM onto the interposer, along with the GPU die. It's an incredible feat, and a big change from the spread of DRAM chips around the GPU, using precious PCB space. Well, researchers at Stanford and MIT are doing one better: putting the CPU and RAM onto a single unit.

The researchers have teased a prototype processor that is made from graphene carbon nanotubes, with resistive RAM (RRAM) layered on top. The team says that their prototype chip is "the most complex nanoelectronic system ever made with emerging nanotechnologies", creating a 3D computer architecture. Carbon is the key here, as the use of silicon for the CPU isn't good for the high temperatures required, as the RRAM cells would be damaged.

I'm sure you're not surprised to hear that DARPA and NSF are funding the project, with the carbon nanotube sensors also able to detect gases, because the information can be processed and measured on-chip.

Today is a new day, and that means new Ryzen ThreadRipper leaks: with some news of AMD's upcoming flagship Ryzen ThreadRipper 1950X processor throwing up its fists in a Geekbench database war with Intel's latest and greatest Core and Xeon CPUs.

This time, we're looking at the Ryzen ThreadRipper 1950X which is AMD's flagship 16C/32T processor with a base clock of 3.4GHz. We should expect boost CPU clocks of 3.6GHz, while XFR will take it to 3.8GHz. There'll be 32MB of L3 cache, and 8MB of L2 cache for a total of 40MB. But we're here for the performance numbers, right? In the single-core test on Geekbench, the Ryzen ThreadRipper 1950X scored 4074, but compared to Intel's new Core i9-7900X with 10C/20T of performance - the 1950X gets thrashed, with the 7900X hitting 5403.

But no one is going to be buying AMD's new Ryzen ThreadRipper CPUs for their single-core performance, so moving onto the Geekbench multi-core score on the Ryzen ThreadRipper 1950X we have 26,768 - which is a much better result, and closer to the 33,945 that the Core i9-7900X scores. AMD will surely have many more optimizations to make to its Ryzen ThreadRipper CPUs and the upcoming X399 chipset, but for now Intel is in the lead by a massive margin. Somewhere that AMD can destroy Intel is on price... something Intel seems to be happy keeping a premium on.

Intel must be shaking in its boots over the various threats from AMD, and now there's news that AMD has scooped back a huge 10.4% of the CPU market share from Intel... all in Q2 2017.

PassMark's quarterly market share report has interesting results, from thousands of submissions that flow through their database each quarter. Remember: PassMark's market share data comes from benchmark submissions, and not total systems sold - it also doesn't include any operating systems other than Windows.

AMD now owns 31% of the CPU market, which is a huge freakin' deal compared to the 20% they had just three months ago. This is the largest CPU share dominance AMD has enjoyed over the last 10 years, and a 10% gain in the CPU market in a single quarter is a massive achievement.

AMD has been on the offensive ever since it launched Ryzen earlier this year, with the next few weeks leading into the Ryzen ThreadRipper family of CPUs and their respective X399 chipset, but AMD also teased its mid-range Ryzen 3 CPUs destined for a couple of months from now.

AMD's upcoming Ryzen PRO family of processors were announced in the last 24 hours, with some - but not all, details on Ryzen 3. Ryzen 3 should have the same cache structure as the lowest-end Ryzen 5 processor, the Ryzen 5 1400. We will receive 8MB of L3 cache enabled, versus the 16MB of L3 cache on Ryzen 5 1500X, 1600, and 1600X. Each core will still retain its full 512KB of L2 cache, and lowered TDPs of just 65W.

We don't know the full details on Ryzen 3 just yet, but Ryzen 3 will be 4C/4T processors without SMT enabled like Ryzen 5. Ryzen 3 1300 should be a 4C/4T chip @ 3.5/3.7GHz, while the Ryzen 3 1200 will be a 4C/4T processor at 3.1/3.4GHz. We should expect the Ryzen 3 1300X as well, but we'll have to wait and see about details on that in the near future.

The battle of the top chips continues. Samsung Electronics is reportedly focusing on the next generation of 6nm chip fabrication technology. In order to do so, the Korean tech giant will stop further investments in the 7nm chip manufacturing technology.

Samsung is expected to introduce 8nm chips next year as an improved version of the 10nm technology, which has already been found in some of the best Android smartphones in 2017 so far. The Koreans will produce a limited number of 7nm line and focus its investment on the 6nm market.

Samsung should be installing two new ASML machines, vital for improving chip manufacturing efficiency, this year, and seven more in 2018. The manufacturing of Samsung's 6nm process and chips using it is expected to begin starting in 2019.

UPDATE: The video points fingers at not just Intel and GIGABYTE, but also ASUS and MSI X299 motherboards. The video stated that the X299 Aorus Gaming 3, Prime X299-A, and MSI's X299 Gaming Pro Carbon VRMs throttled when all cores were overclocked to 4.6GHz on a test bench. It was also pointed out that with proper airflow, the VRMs do much better and shouldn't throttle. These motherboards are at the entry-level to mid-range side of each brands' X299 line up.

Intel's new X299 platform is now out, with various issues towed behind it, professional overclocker 'der8auer' says the platform is a complete "VRM disaster".

It's not just Intel that get the blame, but motherboard makers as well with der8auer saying that the blame is "50/50" on both Intel and motherboard companies. Intel's problems lie in the new X299 platform launching two months early, as it was meant to be released in August and came out in June. This forced motherboard makers into overdrive, with der8auer adding the companies had "almost zero time for developing proper products".

At MWC Shanghai Qualcomm announced a bunch of new products including the new mid-range processor, the Snapdragon 450.

This processor is designed to be put in mid-range smartphones, with a heavy focus on the Chinese market and Chinese manufacturers.

The Snapdragon 450 has an improved CPU and GPU with a 25% increase to both over the Snapdragon 435. It will feature an eight-core processor and Adreno 506 GPU which should offer better performance at lower power thanks to also being a 14nm FinFET chip.

Apple is preparing their next-gen iMac Pro systems, which can be configured with Intel's new Purley-based CPUs rocking up to 18C/36T and ECC RAM for the best you can get in an AIO desktop PC.

These are the configurations we should expect for the new iMac Pro:

But, there's now news coming from Pikeralpha that one of the most interesting parts of Apple's new iMac Pro could be the fact that it uses an ARM co-processor. The use of an ARM co-processor would allow the iMac Pro to perfect regulate the power going through the machine, perfectly tweaking the CPU and GPU management to keep power draw down, and thus temperatures. This is incredible important for an all-in-one that rocks a 18C/36T processor, 128GB of ECC RAM, and a bunch of PCIe-based NVMe SSDs, and AMD's new Radeon Vega GPU architecture.

AMD has been making some pretty serious waves with its Zen CPU architecture, but now that the enterprise/server-grade EPYC processors are here, Intel is in for a big fight - even with its spiffy new Xeon Platinum line of CPUs.

AMD's new EPYC 7601 comes in 32C/64T, but has been used in a dual-CPU server with a total of 64C/128T for some serious computational power - with SiSoft Sandra and Cinebench R15 2P benchmarks. We're seeing a competition between Intel's new 28C/56T processors in both single- and dual-CPU configurations, with a dual-CPU rig with Intel's new Xeon Platinum 8180 and a total of 56C/112T of power.

The benchmarks saw an average all-core CPU clock of 2.7GHz on all 64C/128T of AMD's dual EPYC 7601 system, while the dual Intel Xeon Platinum 8180 had 3.4GHz across all of its dual-CPU goodness at 56C/112T. In the Cinebench R15 performance, we're looking at AMD's EPYC 7601 scoring around 6879 points, while Intel's new Xeon Platinum 8180 pushes 8301.

If Intel wasn't already in enough trouble with AMD's constant onslaught of products with Ryzen and now Ryzen ThreadRipper, the company is now going to have a huge storm surrounding it over a newly-discovered flaw in Intel's Skylake and Kaby Lake architectures with Hyper-Threading.

The HT-enabled processors with critical flaws were discovered on the Debian Linux user list, and sent out without a warning notification - but these issues extend to Windows, and other operating systems, too. The errors with HT-enabled Skylake/Kaby Lake CPUs can lead to various issues ranging from your entire system locking up, major data corruption or loss, or even more. As HotHardware points out: "the replication conditions are very specific and are unlikely to be encountered by most users in the wild". Still, it's not a good thing to see a company the size of Intel experiencing major issues like this, especially with AMD now competing in a big way in the CPU market again.

The problems surrounds Intel errata documentation, explained as:

Intel has been reacting to AMD's new Ryzen CPUs with the tease and release of their new Core i9 processors, but the mainstream Coffee Lake CPUs that are set for a reported launch early next year are being teased more and more.

The latest on Intel's upcoming Coffee Lake-based Core i7 processors is that they will come out in 6C/12T flagship models, with similar to performance of AMD's Ryzen 5 1600X, which is also a 6C/12T processor. There's now a listing on the Geekbench database of a few new Intel processors, with a 6C/12T model with 1.5MB of L2 cache and 12MB of L3 cache. We should expect the 6C/12T processor to come with up to 3.5GHz base and 4.2GHz boost clocks, which is hefty for a 6C/12T mainstream CPU, especially from Intel.

With the Geekbench entry, the chip is clocked at 3.2GHz base, and has a single-core score of 4619 and multi-core score of 20,828. If we compare this against a retail AMD Ryzen 5 1600X processor, we have a single-score of 4574, and multi-core score of 20,769. Right on the money against the Ryzen 5 1600X.

Intel just launched its new Core i9 range of processors, but Coffee Lake news continues to roll out, as Coffee Lake is a refresh/optimized slice of Kaby Lake that supports more than 4 CPU cores.

A new engineering sample has turned up on the SiSoft Sandra benchmark database, showing that Intel will have 6C/12T CPUs at 3.5GHz, and over 4.2GHz with Turbo enabled. We should expect 256KB of L2 cache per core, and 9MB of shared L3 cache. There's also a variant of Coffee Lake that will include a larger 12MB of L3 cache, too.

We should expect the Coffee Lake-S series to come in 4C/8T with GT2 graphics, as well as 6C/12T with GT2 graphics. There will be Coffee Lake-S and Coffee Lake-H series processors, but we might see Intel change the socket from LGA 1151 to something else with Coffee Lake, but let's hope they don't - alright?

The NDA has lifted on AMD's next-gen Epyc CPU range of processors, with the company detailing some of the pricing on its Xeon CPU competitors - with some hefty pricing at the flagship end of the scale.

AMD will have a bunch of 32C/64T options, with the Epyc 7601 flagship CPU that is available right now for $4200. Under that, we have the Epyc 7551 without pricing yet, but the Epyc 7501 is priced at $3400 and also available today. Towards the end of July, there will be the Epyc 7401 for $1850, the Epyc 7301 for $825, Epyc 7281 for $650, and Epyc 7251 for $475.

Our resident CPU editor Steven Bassiri, has detailed AMD's new Epyc CPUs here - with a more detailed piece on the way very soon.

The massive multi-threaded consumer CPU war is now starting thanks to AMD hitting Intel where it hurts with Ryzen and hitting 8C/16T @ $499... but Ryzen ThreadRipper with up to 16C/32T is on the way, so the preemptive strike from Intel with Core i9 happened.

Now we're seeing the first review of Intel's upcoming Core i9-7900X, their new X299-based Core-X series processor with 10C/20T of CPU performance that seems to perform well, overclock well, but is priced pretty high. Hexus has published their review from an Intel CPU sample that was not provided to them by Intel - so they could throw the review online and not worry about Intel's legal team emailing them saying they've broken NDA.

We can see that the fresh X299 chipset needs some more time to mature, but we have some gaming and overclocking charts for you to gander over - and consider buying the 10C/20T chip in the form of the Core i9-7900X over the upcoming Ryzen ThreadRipper range from AMD. My tip: wait.

AMD is killing it with their new Zen CPU cores, with Ryzen already a success and Ryzen ThreadRipper going for Intel's throat - but what about the server market? This is where Epyc comes into play, with the upcoming Epyc 7000 series from AMD. VideoCardz picked up the news, with some details on the Epyc 7000 series:

No compromise 1-Socket

There's some huge performance increases over Intel's Xeon E5-2699A v4 processor, which is a 22C/44T @ 3.6GHz CPU, with 55MB of cache, and a price of $4000+. We have 47% more performance over the single socket CPU, but given there are 18 more CPU threads (albeit at 400MHz slower clocks), AMD is going to send some massive shocks throughout the server market with Epyc.

Remember the fully integrated voltage regulator (FIVR) introduced with Haswell CPUs? Turns out that news of Intel axing the technology was not completely true. Intel's previous FIVR was implemented to reduce motherboard voltage regulator (VR, VRM, or MBVR) complexity, as it fed the CPU a single voltage and the CPU then internally branched off, reduced, and controlled the other voltages needed for other CPU domains (graphics, system agent, cache, IO, etc.).

Traditionally, CPU voltage regulators are fed 12v from your main system power supply, and then they reduce it down to voltages below 2v for the CPU and its different domains. However, as CPUs became more complex they required multiple separate VRs, each with their own controllers (PWM Controller). Adding more external VRs not only increases motherboard costs because of the need for each VR to be individually controlled, but also more importantly, it takes up motherboard real estate which is scarce on a motherboard with eight memory DIMMs surrounding the CPU. Over the past week, Intel quietly made public volume 1 of the Skylake-X datasheet, and you can find it here: Intel Skylake-X Datasheet Volume 1 of 2.

In table 1-1 on page 14 we find that "IVR" will be referenced in the datasheet, and it stands for "Integrated Voltage Regulation (IVR): The processor supports several integrated voltage regulators." Later in the datasheet we find reference to the fact that the IVR is related to the previous FIVR through a signal name called "FIVR_Fault", which "Indicates an internal error has occurred with the integrated voltage regulator. The FIVR_FAULT signal can be sampled any time after 1.5 ms..." We then learn further down about the VCCIN signal, "1.8 V - 1.55 V input to the Integrated Voltage Regulator (IVR) for the processor cores, lowest level caches (LLC), ring interface, PLL, IO, and home agent. It is provided by a VR 13.0 compliant motherboard voltage regulator (MBVR) for each CPU socket.

AMD really has the high-end desktop market in a twist with their upcoming Ryzen ThreadRipper CPUs, with a new tease of the Ryzen ThreadRipper 1950X processor, a 16C/32T part with a purported 3.4GHz base clock.

The latest news is comign from a purported engineering sample on the ThreadRipper 1950X processor, which packs a base CPU clock of 3.4GHz and XFR boost clock of up to 3.7/3.8GHz. We should expect a huge 32MB of L3 cache, 8MB of L2 cache, bringing us to a total of 40MB of cache. In some benchmarks, it has been compared to Intel's higher-end Xeon E5-2697A V4 processor that features 16C/32T @ 2.6GHz base and 3.6GHz boost.

The multi-core score of the Xeon processor is 30,450 while the ThreadRipper 1950X hits 24,539... but I'm sure there are plenty of optimizations that AMD will be making just like the normal Ryzen CPUs, and then the motherboard makers will surely optimizations for ThreadRipper and its multi-core performance.