If there is one thing we can all agree on in the computer hardware industry, its that the industry moves fast. If you aren't constantly following tech blogs, reading through company roadmaps, and staying abreast of the latest acronyms and product names, it can quickly become difficult to just jump back in, read a review or two, and feel like you know whats going on. This guide is designed to help with that - I've done all the research that I needed to do to understand what all these new things are and why they're cool, and now, I've written it down to share with you.
Whether you're a longtime veteran of the scene who is trying to get back into things after a break, a complete newbie who is getting excited about computer hardware and overclocking for the first time, or an absolute expert who is giddy at the chance to call me out on something you think I got wrong, we hope this guide will be (almost) everything you need to get a great set of background info about the last year in computing, and what the upcoming year holds. Major categories are listed below, along with individual products / technologies in each. I've written a short summary of the most important details for each technology, and provided a few great external links where you can dig into the details more if you so desire.
Enjoy.
(image credit: The SSD Review )
Samsung 3D Vertical NAND Flash (aka V-NAND)
Samsung's Overview of V-NAND
AnandTech's Samsung 850 Series SSD Review
Samsung's new V-NAND technology is a method of arranging the individual NAND flash cells in vertical layers (as opposed to the strictly one-layer planar NAND processes we've seen for years). This approach is in contrast to the tried-and-true semiconductor improvement plan of just making everything smaller (which has indeed happened - over the last 15 year or so, as standard NAND flash elements have gone from 120nm to as small as 19nm). Samsung touts the the many benefits of this arrangement, including higher capacity, longer drive lifetimes, better power efficiency, and of course, improved performance. Sounds great, right? It indeed is. The first product line using this new tech, Samsung's SSD 850 Pro line, launched over the summer in most markets, and is blisteringly fast. In AnandTech's typically excellent review of the new line, their verdict was that if you can justify the cost, the SSD 850 Pro line knocks it out of the park in every way. V-NAND delivers on what it promises, and is absolutely worth paying attention to if you're SSD shopping. The Samsung 850 Pro series is at the top of the SSD market right now.
NVM Express (aka NVMe)
AnandTech's Intel P3700 review, which includes a nice intro to NVMe
SSDs redefined our expectations about how mass storage could and should work, and the continued refinement of these drives has now meant that in many cases they outperform the buses used to connect them to the rest of the computer - even 600 MB/s SATA II sometimes simply isn't enough anymore, and so manufacturers are increasingly jumping to another very common, much higher-bandwidth standard for connecting up their SSDs - PCI Express. Though high end (especially enterprise-oriented) SSDs have used PCI Express for some time now, they usually did so using nonstandard communications protocols. NVM Express (where the "NVM" is "non-volatile memory", aka flash) is a standardization effort designed to reduce the need for PCIe implementations of SSDs to require additional device drivers, or any configuration at all beyond what you'd expect when using an older-style SATA SSD.
At the end of 2013 we saw some of the first controller chipsets supporting NVMe introduced to market, but now at the end of 2014 we're seeing commercially available products incorporating these new chipsets and supporting the standard. NVMe is supported natively in Windows 8.1, with updates in Windows 7, and in Linux kernel versions 3.3 and up. Notably, though Apple has used PCIe-based SSDs in recent hardware like the current gen Macbook Air and Mac Pro, they do not support NVMe at this time. NVMe is clearly the way of the future - SSD performance improvements show no sign of slowing down, and with PCIe 3.0 16x offering up to 128Gbps of throughput, there is plenty of room to grow. With companies like Marvel and Phison already announcing 4GB/s SSD controllers, that room to grow can't come soon enough.
The M.2 Form Factor
Andandtech's M.2 / NGFF Form Factor Intro
Most well known as the replacement for mSATA and PCIe MiniCard, the M.2 form factor was previously called "NGFF" and defines a broad specification for expansion cards of many types, supporting both USB 3.0 and PCIe connectivity. This is a form factor in which we'll see many NVMe devices released (see the NVM Express section of this guide for more details). As our own Sean Webster put it, "M.2 is next big thing since sliced bread", and its a great example of a generic form factor that provides a ton of flexibility across both laptop and desktop platforms.
LSI 3700 Family SSD Controllers
AnandTech's LSI 3700 Introduction
LSI's 3700 Series Press Release and Feature Breakdown
The general theme for SSD technologies continues to be better performance across the board, and LSI's much-anticipated new line of 3700 series controllers is no exception. We've known about these for a while, and they're the first controllers to feature both SATA and PCIe support on one chipset, as well as being one of the first devices families for the consumer market which are nearly capable of saturating a PCIe 2.0 x4 connection - in other words, these chipsets will be used in some seriously fast SSDs, and the easy on-chip inclusion of both SATA and PCIe should help to drive adoption of PCIe in high-performance enthusiast drives. LSI is also introducing a new adaptive error-correction technology - called SHEILD - intended to prolong the useful lifetime of SSDs, especially with the higher error rates of smaller manufacturing processes for NAND flash.
DDR4 SDRAM
Micron's DDR3 to DDR4 Comparison Chart
Overclock.net and Corsair's DDR4 Panel Discussion / QA Session
It isn't very often that we see a new generation of DRAM come around, so this one is notable almost by default - the fourth generation of the DDR SDRAM standard includes many of the typical improvements we've come to expect: lower supply voltages, higher frequencies, and larger chip densities. Additional improvements should come in the form of a work-in-progress low power version of the standard, which isn't finalized, but will shave a few tenths of a volt off the already low. Micron has published an excellent chart detailing the differences between DDR3 and DDR4 for those of you who are very familiar with DDR3 already. Q2 2014 saw some initial DDR4 product releases for the server market, while the desktop and personal computing market took another few months to get things together and didn't see DDR4 product releases until Q3 2014. Back in August, we hosted a QA discussion with some R&D and Marketing folks from Corsair, answering many of the questions members had about the real-world impact of DDR4, what it means for overclocking and gaming, and misc other bits of curiosity. Whats the takeaway? For right now, DDR3 and DDR4 performance largely overlaps - which is pretty typical for any new architecture coming to market and competing against a highly refined version of its predecessor. The good news is DDR4 is expected to have a long runway for performance improvements into the future, from both a speed standpoint and a power usage standpoint. There are other immediate benefits (like the new connector shape designed to reduce stress on both RAM DIMMS and the board slots they plug into), but these probably aren't applicable to all but the most extreme subset of users who very frequently insert / remove DIMMS from motherboards, etc.
What are your options if you want to use DDR4 right now? Intel's Haswell-E platform launched at the end of August, and it was the first consumer platform to make use of DDR4. Its high end, high performance, and high price - but you can buy it and use it right now. If you're an AMD fan, at this time we know very little about their next CPU architecture (though it presumably will support DDR4, as well). Curious to know more about Haswell-E or Skylake, Intel's upcoming mainstream platform with DDR4 support? Check out the CPUs and Motherboards section of this guide.
Intel Haswell-E CPUs and X99 Chipsets
Tom's Hardware's Haswell-E Reviews
Overclock.net's Haswell-E Review and Article Roundup
Anandtech's Guide to X99 Motherboards, Holiday 2014
August 29th at PAX Prime in Seattle was a very busy day for Intel - they introduced a new line of i7-5000 series CPUs, codenamed Haswell-E, their new X99 series motherboards using the new LGA2011-3 socket, and those two lines combined also were Intel's first step into consumer-level DDR4 support. (For more on DDR4, check out the Storage and Memory section of this guide). The new i7-5000 series chips keep the same 2011 pin count as their predecessors, but, Intel has added a slightly different physical footprint to the chip to make sure that the "new" 2011 can't be mixed up with the "old" 2011 - these are not compatible platforms. You can use the same cooling hardware between them, but that's it. Sandy Bridge-E and Ivy Bridge-E will not fit into the new X99 motherboards.
Lets talk about X99 on its own, which Intel is calling the "ultimate desktop PC platform for extreme gamers, enthusiasts, and content creators" - positioning it squarely in the HEDT (high end desktop) market segment. X99 based motherboards are designed as enthusiast and prosumer level products; they're expensive, high performance, and full of features. The new chipset is the first to chip with the LGA2011-3 socket, and thus is the first to support Haswell-E processors. At present, that means only i7 chips - there are no lower level i5 or i3 processors available on the platform. The X99 chipset family supports 10 SATA 6 Gbps ports, up to 6 of them in RAID configurations, 14 USB ports, six of them supporting USB 3.0, and the relatively standard package of onboard audio, onboard gigabit Ethernet, and a few other things stuck onto a DMI 2.0 bus. Nothing revolutionary here, but, a nice refresh in Intel's slow-moving desktop chipset line. All of the major motherboard makers are churning out products based on X99 - for a detailed roundup of the individual products, take a look at AnandTech's 2014 Holiday X99 guide, linked above.
On to Haswell-E, now more properly known as the i7-5000 series line of processors: These are high-end enthusiast CPUs, including Intel's first 8-core processor in a consumer socket, the Core i7-5960X (8 cores at 3.5Ghz, with 8 cores at 3.0GHz). Others in the line include the Core i7-5930K (6 cores at 3.5GHz), and the Core i7-5820 (6 cores at 3.3GHz, but, with only 28 PCIe 3.0 lanes instead of the two higher-end models' 40 PCIe 3.0 lanes). All three models operate with a 140W TDP, and all are either Extreme Editions or K models, so they're very overclockable. So whats the verdict? The Core i7-5960X does exactly what it is supposed to do - perform fantastically well on parallelized workloads like rendering, software builds, multimedia encoding, etc - but despite being the cheapest 8-core chip in Intel's entire product lineup, its still very expensive. For many tasks like gaming, the benefits of the extra cores are overshadowed by the higher clock speeds of the two other i7-5000 series chips (especially the Core i7-5820, which manages to offer a very compelling value for the gamer or overclocker, which its only main shortcoming being its relative lack of PCIe 3.0 lanes). If you want a DDR4 rig, and you want it now, this is the product line you should look to - and despite a lack of anything groundbreaking on either the chipset or CPU front, overall great performance and a nice stepping stone into DDR4 (with rumored forward-compatibility with future Broadwell chips on the same socket) make the X99 and Core i7-5000 combo very appealing to the enthusiast, gamer, or overclocker building a new high-end Intel rig right now with an eye towards the future. For detailed specs, see our compilation of many Core i7-5000 series reviews linked above.
Intel Broadwell CPUs
Anandtech's In-Depth Look at Intel's 14nm Process
PC World's "Broadwell or Skylake" Opinion
"Broadwell" is Intel's codename for its 14nm successor to the Haswell family of chips, and Intel's first foray into 14nm production processes in general. As we expect from a process shrink like this, Broadwell delivers significant power savings (early numbers from Intel cited 30% improvements on mobile device battery life) and other incremental improvements to instruction sets, onboard graphics hardware, and modest bumps to raw performance numbers. The Broadwell chips which are initially available on the market target mobile devices. Intel has not only shrunk the power demands, but also the physical footprint of the chip, and even advertises fanless configurations - but though the emphasis is on mobile, these are still high-performance processors destined for high-end devices. Broadwell is not designed to replace the Atom family or compete directly with ARM's offerings in the mobile space. Broadwell chips can currently be purchased in mobile devices advertising the Intel Core M label.
Though the focus is on mobile, Intel has not forgotten about the desktop. Early reports in 2013 prompted fear that there would be no socketed chips in the Broadwell family for use in desktop configurations, but, Intel eventually announced the Broadwell Unlocked line. Broadwell Unlocked targets the enthusiast market, is pin-compatible with existing Haswell motherboards, and features unlocked multipliers (much to the delight of overclockers everywhere). Expected to launch in mid 2015, Broadwell Unlocked is also likely to be the last enthusiast product release on the LGA 1150 socket, with a launch that closely coincides with Skylake's general availability.
Intel Skylake CPUs
ExtremeTech's Intel Roadmap Update (Skylake Focus)
Anandtech's Coverage of Intel's Skylake Demo
Skylake is positioned to be the "tock" to Broadwell's "tick" in the standard Intel major-minor tick-tock product release cycle. Shipping on a new socket (LGA 1151), Skylake will bring a number of updated technologies to mainstream platforms - DDR4 (with simultaneous DDR3 support available through the UniDIMM socket), Thuderbolt 3.0, SATA Express, and the Rezence wireless charging standard. Skylake is expected to come in 5 major variants ("S", "H", "U", "Y", and "K"), with the first 4 available at launch and the unlocked, overclocker-targeted "K" version arriving later in the product lifecycle. Only the "S" and "K" variants will use an LGA socket, the other variants will be soldered BGA form factors for OEM use. The BGA variants also move the Platform Controller Hub (PCH) onto the CPU die, creating a true SoC configuration. Socketed versions leave the PCH in a separate chip on the motherboard, and upgrade to DMI 3.0 for much higher bandwidth intra-chip communication. Graphics hardware on all Skylake chips is expected to retire support for VGA in favor of HDMI and/or DisplayPort. DirectX 12 is also expected to be supported.
Performance expectations around Skylake are still somewhat vague - little is known at this time about individual product clock speeds or core configurations, though we do expect that no Skylake variant will have more than 4 cores or more than 20 PCIe 3.0 lanes, suggesting that Skylake won't be setting any new performance records with the enthusiast crowd (at least initially). The absence of anything revolutionary on the performance front may explain why Intel is content to launch Skylake and Broadwell almost simultaneously, providing something for both the large OEMs and the enthusiast markets, respectively, at the same time (after large delays in Broadwell's launch). Skylake is obviously a major architectural refinement, but, as an enthusiast who is most predominately concerned with performance, you may be better off looking to the Broadwell line for 14nm products, or even the Haswell-E line if you're looking to move to DDR4 on a very high-performing processor without regard to the manufacturing process
AMD Zen and K12
Anandtech's Introduction to AMD K12
WCCFTech's Zen Introduction Coverage
AMD has recently reinvigorated their leadership team with several hires from Apple (including CPU design guru Jim Keller and former AMD Graphics CTO Raja Koduri), and is simultaneously developing two new CPU architectures, both currently expected to launch in 2016.
K12 is a 64-bit ARM architecture running the ARMv8-A instruction set, specifically targeting the server and embedded markets. It will be built upon GlobalFoundries' 14nm process, and is expected to boast relatively high clock speeds. Zen is the x86-compatible sister chip to K12, and will target the desktop / enthusiast markets. Zen is expected to drop AMD's current CMT (a la Bulldozer) approach to multithreading in favor of a more traditional SMT design, which should result in products that feature fewer CPU cores overall with each individual core operating more efficiently compared to their current offerings. Keller's involvement in both projects has been largely interpreted as a very positive sign - he notably led development on the Athlon 64 family of chips in the early 2000s, a product line which was incredibly successful and very popular with the enthusiast market - but at this point there is still far too little known and far too much to prove for any real assessment to be made about the impact the K12 and Zen lines will have on the market. We recommend you pay attention, though, if for nothing else the fact that the people involved in these projects have enjoyed some huge successes in the past.
NVIDIA Maxwell Family GPUs
Tom's Hardware's GTX 970 and GTX 980 Reviews
Overclock.net's Maxwell Review Roundup
When it launched last summer, NVIDIA's new Maxwell core was first seen on a few modestly performing cards which featured very remarkable power efficiency - the GeForce GTX 750 and 750 Ti. After several years where it seemed like there was no limit to the quantities of power GPUs would draw, and jokes being made about every manufacturer being able to sell their wares as space heaters if the gaming industry collapsed, it was very refreshing to see "performance per watt" back on designers' minds. NVIDIA continued the pleasing trend with its new low power architecture when it launched the GTX 970 and GTX 980 later in the year, product lines for the enthusiast market which are almost a whole 100 watts less power-thirsty than their predecessors. That's remarkable, and the performance isn't too shabby, either. The GTX 980 is the fastest single-GPU card you can buy, and at under $600, the pricing is right in line with other high end performers. Its even very capably handles 4K resolutions (previously only the domain of ultra-expensive dual-GPU cards). The GTX 970 is a similarly good deal, offering performance that rivals much more expensive cards in AMD's lineup.
The Maxwell-based card family has collected many accolades, and they are still relatively recent entrants into the market. They're very capable performers with reasonable prices and power consumption improvements that are better than anything we can remember seeing recently - in other words, these cards are very worth paying attention to if you're shopping now or will be shopping in the near future. They've won a lot of awards and garnered a lot of hype, and they deserve every bit both.
NVIDIA GM200
WCCFTech's Report on Initial GM200 Specs Leak
Leaked GM200 vs AMD Fiji Performance Comparisons
GM200, affectionately referred to as "Big Maxwell", is NVIDIA's next-generation graphics core powering products for expected release in late 2015 (presumably under the Titan II or GTX 980 Ti names, but, this is all pure speculation in absence of product name reveals from NVIDIA). GM200 will be built on a 16nm process with a 384-bit memory bus, and leaked benchmarks suggest it will come in limited and fully unlocked versions, with the latter featuring somewhere near 3000 CUDA cores compared to the limited versions ~2600. Also leaked (but keep in mind, totally unverified) is a performance number citing 34% improvement over the GTX 980.
AMD Fiji XT
WCCFTech's Fiji XT Preview
Leaked R9 390X (?) Benchmarks
The release of AMD's next-generation GPU line is expected in early 2015, and will be powered by the Fiji XT graphics chip under the AMD Radeon R9 300 series product name. This family of cards is expected to use 3D High-Bandwidth Memory from Hynx, which significantly outperforms GDDR5 (by as much as a 9x margin). The manufacturing process size is unknown, either 28nm or 20nm depending on which vague remarks from company executives you choose to read into, and the expected architecture is AMD's GCN 1.2. The R9 390X is expected to be the first single-GPU reference design with a liquid cooler from the factory. Little else is known about these cards at this time - but with the GPU performance race being what it is between AMD and NVIDIA, expect them to be serious contenders due to their memory technology alone.
Windows 10
Microsoft's Windows 10 Announcement
TechRadar's Windows 10 Overview
Microsoft's first announcements around Windows 10 (the next major version after Windows 8, don't worry, you didn't miss Windows 9) came over the summer. Building on Windows 8's introduction of ARM support, Windows 10 is being billed as the "one version of Windows that will run everywhere". This theme of unification goes well beyond just the device platforms the OS will run on, though, as Microsoft brings the familiar start menu and the live tile "Metro" interface into one hybrid menu. The classic start menu portion works just as you'd expect, and off to the side of it, a much smaller version of the formerly fullscreen live tile interface takes care of the App ecosytem. Fullscreen Metro-style apps are also gone on the desktop, in favor of presenting the standard mice-and-keyboard friendly view to users who are using mice and keyboards. One can probably also expect similar refinements to the touchscreen experience when running regular "desktop" apps on a tablet or phone interface. Other adjustments include a refined update system, dragging the command prompt into this century with such innovative features as native copy / paste support, and the introduction of DirectX 12 (see our DirectX 12 blurb below for more details on that). Windows 10 is expected to officially launch in the middle of 2015, with actual release dates expected to be announced towards the end of January. So far, it looks promising.
DirectX 12
ExtremeTech's DirectX 12 Announcement Coverage
MSDN's Intro to DirectX 12
The next major version of DirectX, Microsoft's suite of multimedia tools for developers, will be introduced alongside Windows 10. What we know so far is that DirectX 12's development was likely prompted by the success of AMD's Mantle technology, a set of low-level APIs for game developers that promised more direct access to graphics hardware than DirectX or OpenGL could provide, and that to that end DirectX 12 does intend to reduce CPU overhead, make better use of multithreaded hardware, and offer power consumption improvements for mobile users. Recent reports have conflicted information about whether or not DirectX 12 will be supported on operating systems older than Windows 8 - for right now, this is still largely undetermined. DX12 is also notably absent from technical previews of Windows 10.
On the hardware support side of things, NVIDIA has announced they will support DX12 on all "Fermi, Kepler and Maxwell architectural families" - aka every DX11 compatible GPU they've shipped. AMD has promised similar support for their GCN-based graphics cores, which is their 7000 series, R7 series, and R9 series GPUs (or a Kaveri APU). Intel support for DX12 will be present on their current-generation HD Graphics or Iris Graphics hardware. Additional details (or actual benchmarks) of DirectX 12 are still hard to find, but, it is expected that much more information will snowball into the September 2015 IFA show, leading to general availability for DirectX 12 titles sometime in late 2015. If DirextX 12 can deliver on what it promises, we should see all the benefits of Mantle (and more), in a much more widely accessible API. Though its fairly certain that almost any current generation graphics hardware you buy today will work fine with DX12, its definitely something you'll want to make sure any new hardware you pick up will support. Lots and lots of potential - if Microsoft delivers.
Whether you're a longtime veteran of the scene who is trying to get back into things after a break, a complete newbie who is getting excited about computer hardware and overclocking for the first time, or an absolute expert who is giddy at the chance to call me out on something you think I got wrong, we hope this guide will be (almost) everything you need to get a great set of background info about the last year in computing, and what the upcoming year holds. Major categories are listed below, along with individual products / technologies in each. I've written a short summary of the most important details for each technology, and provided a few great external links where you can dig into the details more if you so desire.
Enjoy.
(image credit: The SSD Review )
Samsung 3D Vertical NAND Flash (aka V-NAND)
Samsung's Overview of V-NAND
AnandTech's Samsung 850 Series SSD Review
Samsung's new V-NAND technology is a method of arranging the individual NAND flash cells in vertical layers (as opposed to the strictly one-layer planar NAND processes we've seen for years). This approach is in contrast to the tried-and-true semiconductor improvement plan of just making everything smaller (which has indeed happened - over the last 15 year or so, as standard NAND flash elements have gone from 120nm to as small as 19nm). Samsung touts the the many benefits of this arrangement, including higher capacity, longer drive lifetimes, better power efficiency, and of course, improved performance. Sounds great, right? It indeed is. The first product line using this new tech, Samsung's SSD 850 Pro line, launched over the summer in most markets, and is blisteringly fast. In AnandTech's typically excellent review of the new line, their verdict was that if you can justify the cost, the SSD 850 Pro line knocks it out of the park in every way. V-NAND delivers on what it promises, and is absolutely worth paying attention to if you're SSD shopping. The Samsung 850 Pro series is at the top of the SSD market right now.
NVM Express (aka NVMe)
AnandTech's Intel P3700 review, which includes a nice intro to NVMe
SSDs redefined our expectations about how mass storage could and should work, and the continued refinement of these drives has now meant that in many cases they outperform the buses used to connect them to the rest of the computer - even 600 MB/s SATA II sometimes simply isn't enough anymore, and so manufacturers are increasingly jumping to another very common, much higher-bandwidth standard for connecting up their SSDs - PCI Express. Though high end (especially enterprise-oriented) SSDs have used PCI Express for some time now, they usually did so using nonstandard communications protocols. NVM Express (where the "NVM" is "non-volatile memory", aka flash) is a standardization effort designed to reduce the need for PCIe implementations of SSDs to require additional device drivers, or any configuration at all beyond what you'd expect when using an older-style SATA SSD.
At the end of 2013 we saw some of the first controller chipsets supporting NVMe introduced to market, but now at the end of 2014 we're seeing commercially available products incorporating these new chipsets and supporting the standard. NVMe is supported natively in Windows 8.1, with updates in Windows 7, and in Linux kernel versions 3.3 and up. Notably, though Apple has used PCIe-based SSDs in recent hardware like the current gen Macbook Air and Mac Pro, they do not support NVMe at this time. NVMe is clearly the way of the future - SSD performance improvements show no sign of slowing down, and with PCIe 3.0 16x offering up to 128Gbps of throughput, there is plenty of room to grow. With companies like Marvel and Phison already announcing 4GB/s SSD controllers, that room to grow can't come soon enough.
The M.2 Form Factor
Andandtech's M.2 / NGFF Form Factor Intro
Most well known as the replacement for mSATA and PCIe MiniCard, the M.2 form factor was previously called "NGFF" and defines a broad specification for expansion cards of many types, supporting both USB 3.0 and PCIe connectivity. This is a form factor in which we'll see many NVMe devices released (see the NVM Express section of this guide for more details). As our own Sean Webster put it, "M.2 is next big thing since sliced bread", and its a great example of a generic form factor that provides a ton of flexibility across both laptop and desktop platforms.
LSI 3700 Family SSD Controllers
AnandTech's LSI 3700 Introduction
LSI's 3700 Series Press Release and Feature Breakdown
The general theme for SSD technologies continues to be better performance across the board, and LSI's much-anticipated new line of 3700 series controllers is no exception. We've known about these for a while, and they're the first controllers to feature both SATA and PCIe support on one chipset, as well as being one of the first devices families for the consumer market which are nearly capable of saturating a PCIe 2.0 x4 connection - in other words, these chipsets will be used in some seriously fast SSDs, and the easy on-chip inclusion of both SATA and PCIe should help to drive adoption of PCIe in high-performance enthusiast drives. LSI is also introducing a new adaptive error-correction technology - called SHEILD - intended to prolong the useful lifetime of SSDs, especially with the higher error rates of smaller manufacturing processes for NAND flash.
DDR4 SDRAM
Micron's DDR3 to DDR4 Comparison Chart
Overclock.net and Corsair's DDR4 Panel Discussion / QA Session
It isn't very often that we see a new generation of DRAM come around, so this one is notable almost by default - the fourth generation of the DDR SDRAM standard includes many of the typical improvements we've come to expect: lower supply voltages, higher frequencies, and larger chip densities. Additional improvements should come in the form of a work-in-progress low power version of the standard, which isn't finalized, but will shave a few tenths of a volt off the already low. Micron has published an excellent chart detailing the differences between DDR3 and DDR4 for those of you who are very familiar with DDR3 already. Q2 2014 saw some initial DDR4 product releases for the server market, while the desktop and personal computing market took another few months to get things together and didn't see DDR4 product releases until Q3 2014. Back in August, we hosted a QA discussion with some R&D and Marketing folks from Corsair, answering many of the questions members had about the real-world impact of DDR4, what it means for overclocking and gaming, and misc other bits of curiosity. Whats the takeaway? For right now, DDR3 and DDR4 performance largely overlaps - which is pretty typical for any new architecture coming to market and competing against a highly refined version of its predecessor. The good news is DDR4 is expected to have a long runway for performance improvements into the future, from both a speed standpoint and a power usage standpoint. There are other immediate benefits (like the new connector shape designed to reduce stress on both RAM DIMMS and the board slots they plug into), but these probably aren't applicable to all but the most extreme subset of users who very frequently insert / remove DIMMS from motherboards, etc.
What are your options if you want to use DDR4 right now? Intel's Haswell-E platform launched at the end of August, and it was the first consumer platform to make use of DDR4. Its high end, high performance, and high price - but you can buy it and use it right now. If you're an AMD fan, at this time we know very little about their next CPU architecture (though it presumably will support DDR4, as well). Curious to know more about Haswell-E or Skylake, Intel's upcoming mainstream platform with DDR4 support? Check out the CPUs and Motherboards section of this guide.
Intel Haswell-E CPUs and X99 Chipsets
Tom's Hardware's Haswell-E Reviews
Overclock.net's Haswell-E Review and Article Roundup
Anandtech's Guide to X99 Motherboards, Holiday 2014
August 29th at PAX Prime in Seattle was a very busy day for Intel - they introduced a new line of i7-5000 series CPUs, codenamed Haswell-E, their new X99 series motherboards using the new LGA2011-3 socket, and those two lines combined also were Intel's first step into consumer-level DDR4 support. (For more on DDR4, check out the Storage and Memory section of this guide). The new i7-5000 series chips keep the same 2011 pin count as their predecessors, but, Intel has added a slightly different physical footprint to the chip to make sure that the "new" 2011 can't be mixed up with the "old" 2011 - these are not compatible platforms. You can use the same cooling hardware between them, but that's it. Sandy Bridge-E and Ivy Bridge-E will not fit into the new X99 motherboards.
Lets talk about X99 on its own, which Intel is calling the "ultimate desktop PC platform for extreme gamers, enthusiasts, and content creators" - positioning it squarely in the HEDT (high end desktop) market segment. X99 based motherboards are designed as enthusiast and prosumer level products; they're expensive, high performance, and full of features. The new chipset is the first to chip with the LGA2011-3 socket, and thus is the first to support Haswell-E processors. At present, that means only i7 chips - there are no lower level i5 or i3 processors available on the platform. The X99 chipset family supports 10 SATA 6 Gbps ports, up to 6 of them in RAID configurations, 14 USB ports, six of them supporting USB 3.0, and the relatively standard package of onboard audio, onboard gigabit Ethernet, and a few other things stuck onto a DMI 2.0 bus. Nothing revolutionary here, but, a nice refresh in Intel's slow-moving desktop chipset line. All of the major motherboard makers are churning out products based on X99 - for a detailed roundup of the individual products, take a look at AnandTech's 2014 Holiday X99 guide, linked above.
On to Haswell-E, now more properly known as the i7-5000 series line of processors: These are high-end enthusiast CPUs, including Intel's first 8-core processor in a consumer socket, the Core i7-5960X (8 cores at 3.5Ghz, with 8 cores at 3.0GHz). Others in the line include the Core i7-5930K (6 cores at 3.5GHz), and the Core i7-5820 (6 cores at 3.3GHz, but, with only 28 PCIe 3.0 lanes instead of the two higher-end models' 40 PCIe 3.0 lanes). All three models operate with a 140W TDP, and all are either Extreme Editions or K models, so they're very overclockable. So whats the verdict? The Core i7-5960X does exactly what it is supposed to do - perform fantastically well on parallelized workloads like rendering, software builds, multimedia encoding, etc - but despite being the cheapest 8-core chip in Intel's entire product lineup, its still very expensive. For many tasks like gaming, the benefits of the extra cores are overshadowed by the higher clock speeds of the two other i7-5000 series chips (especially the Core i7-5820, which manages to offer a very compelling value for the gamer or overclocker, which its only main shortcoming being its relative lack of PCIe 3.0 lanes). If you want a DDR4 rig, and you want it now, this is the product line you should look to - and despite a lack of anything groundbreaking on either the chipset or CPU front, overall great performance and a nice stepping stone into DDR4 (with rumored forward-compatibility with future Broadwell chips on the same socket) make the X99 and Core i7-5000 combo very appealing to the enthusiast, gamer, or overclocker building a new high-end Intel rig right now with an eye towards the future. For detailed specs, see our compilation of many Core i7-5000 series reviews linked above.
Intel Broadwell CPUs
Anandtech's In-Depth Look at Intel's 14nm Process
PC World's "Broadwell or Skylake" Opinion
"Broadwell" is Intel's codename for its 14nm successor to the Haswell family of chips, and Intel's first foray into 14nm production processes in general. As we expect from a process shrink like this, Broadwell delivers significant power savings (early numbers from Intel cited 30% improvements on mobile device battery life) and other incremental improvements to instruction sets, onboard graphics hardware, and modest bumps to raw performance numbers. The Broadwell chips which are initially available on the market target mobile devices. Intel has not only shrunk the power demands, but also the physical footprint of the chip, and even advertises fanless configurations - but though the emphasis is on mobile, these are still high-performance processors destined for high-end devices. Broadwell is not designed to replace the Atom family or compete directly with ARM's offerings in the mobile space. Broadwell chips can currently be purchased in mobile devices advertising the Intel Core M label.
Though the focus is on mobile, Intel has not forgotten about the desktop. Early reports in 2013 prompted fear that there would be no socketed chips in the Broadwell family for use in desktop configurations, but, Intel eventually announced the Broadwell Unlocked line. Broadwell Unlocked targets the enthusiast market, is pin-compatible with existing Haswell motherboards, and features unlocked multipliers (much to the delight of overclockers everywhere). Expected to launch in mid 2015, Broadwell Unlocked is also likely to be the last enthusiast product release on the LGA 1150 socket, with a launch that closely coincides with Skylake's general availability.
Intel Skylake CPUs
ExtremeTech's Intel Roadmap Update (Skylake Focus)
Anandtech's Coverage of Intel's Skylake Demo
Skylake is positioned to be the "tock" to Broadwell's "tick" in the standard Intel major-minor tick-tock product release cycle. Shipping on a new socket (LGA 1151), Skylake will bring a number of updated technologies to mainstream platforms - DDR4 (with simultaneous DDR3 support available through the UniDIMM socket), Thuderbolt 3.0, SATA Express, and the Rezence wireless charging standard. Skylake is expected to come in 5 major variants ("S", "H", "U", "Y", and "K"), with the first 4 available at launch and the unlocked, overclocker-targeted "K" version arriving later in the product lifecycle. Only the "S" and "K" variants will use an LGA socket, the other variants will be soldered BGA form factors for OEM use. The BGA variants also move the Platform Controller Hub (PCH) onto the CPU die, creating a true SoC configuration. Socketed versions leave the PCH in a separate chip on the motherboard, and upgrade to DMI 3.0 for much higher bandwidth intra-chip communication. Graphics hardware on all Skylake chips is expected to retire support for VGA in favor of HDMI and/or DisplayPort. DirectX 12 is also expected to be supported.
Performance expectations around Skylake are still somewhat vague - little is known at this time about individual product clock speeds or core configurations, though we do expect that no Skylake variant will have more than 4 cores or more than 20 PCIe 3.0 lanes, suggesting that Skylake won't be setting any new performance records with the enthusiast crowd (at least initially). The absence of anything revolutionary on the performance front may explain why Intel is content to launch Skylake and Broadwell almost simultaneously, providing something for both the large OEMs and the enthusiast markets, respectively, at the same time (after large delays in Broadwell's launch). Skylake is obviously a major architectural refinement, but, as an enthusiast who is most predominately concerned with performance, you may be better off looking to the Broadwell line for 14nm products, or even the Haswell-E line if you're looking to move to DDR4 on a very high-performing processor without regard to the manufacturing process
AMD Zen and K12
Anandtech's Introduction to AMD K12
WCCFTech's Zen Introduction Coverage
AMD has recently reinvigorated their leadership team with several hires from Apple (including CPU design guru Jim Keller and former AMD Graphics CTO Raja Koduri), and is simultaneously developing two new CPU architectures, both currently expected to launch in 2016.
K12 is a 64-bit ARM architecture running the ARMv8-A instruction set, specifically targeting the server and embedded markets. It will be built upon GlobalFoundries' 14nm process, and is expected to boast relatively high clock speeds. Zen is the x86-compatible sister chip to K12, and will target the desktop / enthusiast markets. Zen is expected to drop AMD's current CMT (a la Bulldozer) approach to multithreading in favor of a more traditional SMT design, which should result in products that feature fewer CPU cores overall with each individual core operating more efficiently compared to their current offerings. Keller's involvement in both projects has been largely interpreted as a very positive sign - he notably led development on the Athlon 64 family of chips in the early 2000s, a product line which was incredibly successful and very popular with the enthusiast market - but at this point there is still far too little known and far too much to prove for any real assessment to be made about the impact the K12 and Zen lines will have on the market. We recommend you pay attention, though, if for nothing else the fact that the people involved in these projects have enjoyed some huge successes in the past.
NVIDIA Maxwell Family GPUs
Tom's Hardware's GTX 970 and GTX 980 Reviews
Overclock.net's Maxwell Review Roundup
When it launched last summer, NVIDIA's new Maxwell core was first seen on a few modestly performing cards which featured very remarkable power efficiency - the GeForce GTX 750 and 750 Ti. After several years where it seemed like there was no limit to the quantities of power GPUs would draw, and jokes being made about every manufacturer being able to sell their wares as space heaters if the gaming industry collapsed, it was very refreshing to see "performance per watt" back on designers' minds. NVIDIA continued the pleasing trend with its new low power architecture when it launched the GTX 970 and GTX 980 later in the year, product lines for the enthusiast market which are almost a whole 100 watts less power-thirsty than their predecessors. That's remarkable, and the performance isn't too shabby, either. The GTX 980 is the fastest single-GPU card you can buy, and at under $600, the pricing is right in line with other high end performers. Its even very capably handles 4K resolutions (previously only the domain of ultra-expensive dual-GPU cards). The GTX 970 is a similarly good deal, offering performance that rivals much more expensive cards in AMD's lineup.
The Maxwell-based card family has collected many accolades, and they are still relatively recent entrants into the market. They're very capable performers with reasonable prices and power consumption improvements that are better than anything we can remember seeing recently - in other words, these cards are very worth paying attention to if you're shopping now or will be shopping in the near future. They've won a lot of awards and garnered a lot of hype, and they deserve every bit both.
NVIDIA GM200
WCCFTech's Report on Initial GM200 Specs Leak
Leaked GM200 vs AMD Fiji Performance Comparisons
GM200, affectionately referred to as "Big Maxwell", is NVIDIA's next-generation graphics core powering products for expected release in late 2015 (presumably under the Titan II or GTX 980 Ti names, but, this is all pure speculation in absence of product name reveals from NVIDIA). GM200 will be built on a 16nm process with a 384-bit memory bus, and leaked benchmarks suggest it will come in limited and fully unlocked versions, with the latter featuring somewhere near 3000 CUDA cores compared to the limited versions ~2600. Also leaked (but keep in mind, totally unverified) is a performance number citing 34% improvement over the GTX 980.
AMD Fiji XT
WCCFTech's Fiji XT Preview
Leaked R9 390X (?) Benchmarks
The release of AMD's next-generation GPU line is expected in early 2015, and will be powered by the Fiji XT graphics chip under the AMD Radeon R9 300 series product name. This family of cards is expected to use 3D High-Bandwidth Memory from Hynx, which significantly outperforms GDDR5 (by as much as a 9x margin). The manufacturing process size is unknown, either 28nm or 20nm depending on which vague remarks from company executives you choose to read into, and the expected architecture is AMD's GCN 1.2. The R9 390X is expected to be the first single-GPU reference design with a liquid cooler from the factory. Little else is known about these cards at this time - but with the GPU performance race being what it is between AMD and NVIDIA, expect them to be serious contenders due to their memory technology alone.
Windows 10
Microsoft's Windows 10 Announcement
TechRadar's Windows 10 Overview
Microsoft's first announcements around Windows 10 (the next major version after Windows 8, don't worry, you didn't miss Windows 9) came over the summer. Building on Windows 8's introduction of ARM support, Windows 10 is being billed as the "one version of Windows that will run everywhere". This theme of unification goes well beyond just the device platforms the OS will run on, though, as Microsoft brings the familiar start menu and the live tile "Metro" interface into one hybrid menu. The classic start menu portion works just as you'd expect, and off to the side of it, a much smaller version of the formerly fullscreen live tile interface takes care of the App ecosytem. Fullscreen Metro-style apps are also gone on the desktop, in favor of presenting the standard mice-and-keyboard friendly view to users who are using mice and keyboards. One can probably also expect similar refinements to the touchscreen experience when running regular "desktop" apps on a tablet or phone interface. Other adjustments include a refined update system, dragging the command prompt into this century with such innovative features as native copy / paste support, and the introduction of DirectX 12 (see our DirectX 12 blurb below for more details on that). Windows 10 is expected to officially launch in the middle of 2015, with actual release dates expected to be announced towards the end of January. So far, it looks promising.
DirectX 12
ExtremeTech's DirectX 12 Announcement Coverage
MSDN's Intro to DirectX 12
The next major version of DirectX, Microsoft's suite of multimedia tools for developers, will be introduced alongside Windows 10. What we know so far is that DirectX 12's development was likely prompted by the success of AMD's Mantle technology, a set of low-level APIs for game developers that promised more direct access to graphics hardware than DirectX or OpenGL could provide, and that to that end DirectX 12 does intend to reduce CPU overhead, make better use of multithreaded hardware, and offer power consumption improvements for mobile users. Recent reports have conflicted information about whether or not DirectX 12 will be supported on operating systems older than Windows 8 - for right now, this is still largely undetermined. DX12 is also notably absent from technical previews of Windows 10.
On the hardware support side of things, NVIDIA has announced they will support DX12 on all "Fermi, Kepler and Maxwell architectural families" - aka every DX11 compatible GPU they've shipped. AMD has promised similar support for their GCN-based graphics cores, which is their 7000 series, R7 series, and R9 series GPUs (or a Kaveri APU). Intel support for DX12 will be present on their current-generation HD Graphics or Iris Graphics hardware. Additional details (or actual benchmarks) of DirectX 12 are still hard to find, but, it is expected that much more information will snowball into the September 2015 IFA show, leading to general availability for DirectX 12 titles sometime in late 2015. If DirextX 12 can deliver on what it promises, we should see all the benefits of Mantle (and more), in a much more widely accessible API. Though its fairly certain that almost any current generation graphics hardware you buy today will work fine with DX12, its definitely something you'll want to make sure any new hardware you pick up will support. Lots and lots of potential - if Microsoft delivers.