Is now a good time to upgrade?

We reveal the hardware around the corner that's worth holding out for.

There's a school of thought that says there's never a good time to upgrade. There is always something better just around the corner. To be fair, there is some truth in these words - the development cycles of the major hardware players is fast enough that you'll never be too far away from the next great processor or graphics card. Even so, you've got to know when to take the punt on new hardware. That's where we come in, guiding you to the best kit at any given time, taking into account what's just around the corner and where the latest releases stand in respect to their peers. Having said that, there are times when the weight of what's coming throws out this delicate balancing act. And that's exactly where we find ourselves at the moment - not in every sector, but in enough areas to make it tricky to recommend certain builds purely because we know what will soon be hitting the shops. High-end PCs, for instance, are about to get a major shot in the arm thanks to the next release from Intel, while Nvidia's next graphics card isn't too far away either. Over the next few pages, we are going to be taking a look at the major upcoming releases that should be on your tech calendar. We'll investigate Intel and AMD's processor plans, then take a peek at what's lined up next in the graphics card market. We then look at storage, memory, peripherals and the major software updates. Being armed with all this information will ensure that you make the right decisions about when you upgrade, and what to keep in mind when you do.

Intel Haswell-E Q3 2014

Intel's ?E' processors have always been something of an oddity because they are essentially server processors that have filtered down to the desktop market. The next iteration of these chips, codenamed Haswell-E, is due for release shortly - we'll have the full breakdown in the next issue of PC Format. But what sets this particular series apart from its predecessors is the fact that it will be available in more than just the very highest SKUs. While we will still see a benchmark breaking eight-core, 16-thread i7 5960X clocked at 3GHz (Turbo-ing to 3.3GHz) that will set you back a cool $1,000. There will also be a couple of six-core, 12-thread versions that will be available at slightly more reasonable price points. These prices haven't been confirmed yet, but they are rumoured to be as low as $350 for the 3.3GHz i7 5820K. If true, this could really muddy the usually-clear water between the relatively more value-conscious desktop parts and these enthusiast chips. Haswell-E will also be the first time that DDR4 gets an outing on desktops, which may help explain why Intel feels the need to make some more affordable members of the family. This is because DDR4, much like its predecessors, will be incredibly expensive at launch and isn't set to offer significant advantages over DDR3 for a little while into its lifecycle. You'll need a motherboard packing Intel's X99 chipset for Haswell-E, so when you start adding up all of the platform costs, it certainly won't be a budget option. It will, however, pack some serious multi-threading performance into its chunky LGA 2011 packaging.

Intel Broadwell Q4 2014

The next Core architecture for desktops from Intel is codenamed Broadwell, and it is set to make an appearance at the end of the year. This is essentially a 14nm die shrink of Haswell, with the possibility of a few nips and tucks here and there, but it does nothing to fundamentally change what Haswell offers. Die shrinks have traditionally enabled Intel to up the clockspeeds of their predecessor chips, but as we've just had the Haswell refresh - codenamed Devil's Canyon - the pressure is on for Intel to really make these chips stand out from what it's done before. Broadwell was originally pencilled in to make an appearance much earlier than this, but the transition to the new production process hasn't been without issue. These problems have been sorted now, and the Broadwell processors are expected to make an appearance at the end of 2014, ramping into full production at the start of 2015. The good news about Broadwell is the motherboards that support it are already with us - Intel's Z97-powered motherboards were launched with the release of the Devil's Canyon chips a few months ago. This means that by the time Broadwell actually lands, these motherboards will be mature and performing well. You may need to get jiggy with a little BIOS flashing, but that isn't beyond the scope of you clever lot. The next completely new architecture from Intel, codenamed Skylake, will use the 14nm production process and is expected to appear in the second half of 2015. Not a lot is known about the chip just yet, but the architecture is expected to support both DDR3 and DDR4.

AMD Excavator 2015

The current Steamroller architecture, which can be found inside AMD's range of KaveriAPUs, is set to be replaced by the Excavator architecture some time in 2015. Yet the actual specifics aren't as clear as Intel's regimented tick-tock timetable. What is known is that the first chips to use this fourth-generation Bulldozer core will be the Carrizo APUs, which will use the Socket FM2+ packaging and be produced with a 28nm production process. The core itself is expected to boast support for a number of new instructions, as well as introducing support for DDR4 alongside the current DDR3 memory controllers. Chips that use the Excavator core probably won't support both memory types at the same time, but will make the transition over to the new memory standard when it's more financially viable. The first Carrizo APUs will undoubtedly ship with just DDR3 support, but with the option of a refresh to support DDR4 when it makes financial sense to do so, possibly with a new socket too. As Carrizo is an APU, the graphics subsystem is obviously very important, but it hasn't been confirmed at this stage what form this graphics core will take. Currently, it's expected to be another outing for the Graphics Core Next tech that can be found in the current range of APUs and graphics cards. But still, we can't help hoping that these APUs are more powerful in order to keep Intel honest and lower power graphics cards competitive. Just giving more of the APU's silicon over to the graphics core may be enough here, although this may be a bit of an issue without the drop in production process.

AMD Project Skybridge 2015

AMD isn't just tying its future into the x86 ecosystem. Having already committed to designing an ARM-based, eight-core server system-on-a-chip (SoC), it is going one step further and introducing a fully ambidextrous silicon future. Project Skybridge is a design framework that's specifically set up to allow for either an ARM-based SoC or an x86-based chip to run effectively in the same platform. The ARM variant will be a 20nm Cortex A57 chip with an integrated Graphics Core Next GPU component, and will be pin compatible with an x86 SoC based on a next-gen core codenamed Puma+. When we say "pin-compatible", don't expect to be able to pick up a desktop board before dropping in either an ARM or x86 processor yourself. It's more likely that, at a manufacturing level, a platform will be able to have either one of the chips soldered down onto the same BGA motherboard. This will allow manufacturers to have a single design for a device and be able to offer both Windows- and Android-based versions of the same product without having to change anything apart from the chip itself. Project Skybridge's ARM component will be AMD's first official Android platform and as such both SoCs will be aimed primarily at the low-power mobile market.

AMD K12 and beyond 2016-2017

AMD has stated that this will be the last revision of its Bulldozer architecture before it is retired. One of the architectures following it is codenamed K12 and is expected to hit the shelves sometime in 2016. This follows on from the ambidextrous roadmap AMD laid out for Project Skybridge and shows it is treating its ARM development with the same respect it has for its x86 parts. The new K12 chip will be based on the ARM64 instruction set and will use a 14/16nm production process. It could potentially get AMD in a wide range of custom and embedded designs. But K12 is being designed as a higher-performance part compared with the low-end Project Skybridge SoCs. K12 isn't just being lined up for servers and tablets, but is also being touted for consumer laptops and Chromebooks too. Alongside the K12 ARMv8 chip will be a corresponding new x86-based 64-bit design. As yet, there are no concrete details surfacing about what AMD is going to do with its x86 license post-Bulldozer, but we can only hope a focus on improving the instructions per clock/single-threaded performance is made a priority. We have heard some rumours of a 14/16nm x86 design with the top model rocking some 20 cores. And those whisperings are quite specific about such cores not being AMD's compute cores either. It is also rumoured to be AMD's first LGA processor too, sticking the pins into the socket rather than on the chip itself. A high-performance FX CPU sounds like a pipedream right now, but it might happen.

Nvidia GeForce GTX 880 Autumn 2014

We've been waiting on the new Maxwell GPU architecture to replace the Kepler graphics cards for a while now. Nvidia has stuck with the (admittedly very impressive) Kepler silicon for years, rolling out newer, high-priced spins of its professional graphics GPU, the GK110, to keep upping the ante. We had our very first taste of the Maxwell GPU at the start of this year with the GTX 750 Tiand it was the first time Nvidia had launched brand new chip architecture on a low-end card. Have no fear, though. Nvidia is all set to drop a high-end Maxwell-based card - probably called the GTX 880 if previous naming convention is anything to go by - in September. We had expected the new architecture to be accompanied by a fresh production process too, giving us the hefty power saving and performance that the GTX 750 Tihas demonstrated in the Maxwell GPU architecture, as well as the extra boost that dropping from a 28nm production process to 20nm should deliver. TSMC, the actual physical manufacturers of Nvidia GPUs, have unfortunately run into problems with the 20nm node. Currently, we're unsure whether that's in terms of yield, cost or performance - it's possible there could be issues with all three. Whatever the real reason, the GTX 880 will still be based on a 28nm process, which will allow us to see clearly what benefits the new Maxwell silicon can offer over the existing Kepler chips on the market. We will probably see at least two new cards based on the GM204 GPU coming in the autumn, but as yet we don't have any final specifications on what that chip will entail. The rumours say that the top GPU could be housing anywhere between 1,536 and 3,200 CUDA cores. But if it wants to beat the existing GK110-based cards it's going to have to be something quite special.

Nvidia Pascal 2016

Nvidia's GPU roadmaps always have to be taken with a pinch of salt. Years and architectures come and go without a word on what's happened, but this March we were told about the ?next big thing' from its GPU skunkworks. The Pascal GPU is the architecture that will come after Maxwell and is set to pack in some serious next-generation technology. The biggest change will likely be the introduction of NVLink. This is an Nvidiadesigned interconnect that offers significantly higher bandwidth than even the next iteration of PCIe has been designed to deliver - between five and twelve times the bandwidth of PCIe 3.0. NVLink is probably only going to be used on server-based pro cards and seriously high-end consumer GPUs as it will demand a radical change in motherboard design. Essentially, we're looking at a socketed GPU daughterboard laid parallel to the motherboard rather than standing vertically in a PCIe slot. NVLink is also set to deliver more than the 75W currently offered from PCIe, which could rid us of the sprawl of power cables we have today. We still think lower-end Pascal GPUs will be using existing PCIe layout because there is still support in the GPU design for the standard interconnect. As well as NVLink, Pascal is going to launch with 3D memory, which is similar to that which Samsung has been using for its newest SSD designs. These will be on-package memory chips, stacked on top of the GPU with connections drilling through the layers providing huge amounts of bandwidth between memory and chip. We don't know what process Nvidia is looking to use for Pascal, but there have been rumours it will be skipping the awkward 20nm node and will move straight on to 16nm later in Maxwell's lifetime. It stands to reason that Pascal's transistors will be as small as 16nm.

AMD Volcanic Islands 2.0 Autumn 2014

You simply cannot have one graphics card manufacturer releasing a slew of new cards without the other wanting to get involved. And while AMD is not likely to be offering a brand new graphics architecture, it will be refreshing its existing Volcanic Islands range of cards. These will essentially just be slightly tweaked versions - probably in order to reduce the manufacturing costs and power draw rather than boost actual gaming performance. It looks as if the first will be a GPU codenamed Tonga, which is likely to be a direct replacement for the existing R9 280. We've heard whispers from some AMD card manufacturers that AMD is looking to simply use the same name for the new card, although it's more likely to be some sort of R9 285. That sounds mighty confusing because if the expected specifications are realised, then the Tonga GPU will come with a 256-bit memory bus instead of the existing card's 384-bit bus. It is expected to have the same 2,048 cores as the R9 280X, along with a higher clockspeed, so it should have some pretty decent performance to make up for it. There is also meant to be a new lower-end part, possibly designed to take on the GM107 Nvidia GPUs. That is important because AMD doesn't currently have an answer to the combined low-power and high-performance of the GTX 750 Ti. The rumoured lower-end GPU is expected to be codenamed Iceland and should be a direct replacement for the Cape Verde GPUs that are available today. In terms of a higher-end refresh, there is a possibility of a higherclocked version of the HawaiiXT GPU that is currently in the R9 290X, so we could get a HawaiiXT2 in the same way we had a higher-clocked TahitiXT2 for the HD 7970 GHz edition a couple of years ago.

AMD Pirate Islands 2015

The real new AMD graphics cards are all set for a launch next year. If any of the rumours or leaks we've seen so far are to be believed, the R9 390X could need its own fusion reactor to keep it running. The R9 390X is going to be based on the Bermuda GPU, which is a GCN chip built on the new 20nm production process. Despite the smaller GPU lithography, it could end up being an absolutely enormous processor if the touted 4,224 GCN core figures are in any way accurate. If you thought the GK110 GPU in the GTX Titan was hefty, it looks like the Bermuda XTX could be even bigger. With the 512-bit memory bus on which the Bermuda GPU is expected to run, it looks unlikely that the new partnership between AMD and SK Hynix to produce stacked high bandwidth video memory is going to produce 3D VRAM for the first round of Pirate Islands GPUs. Like Nvidia, AMD is relying on TSMC to nail down whatever issues it has been having with the 20nm production process before we start to see actual volume manufacturing of AMD's new graphics cards. If AMD wants to pursue the 20nm lithography, it may have to start pushing the launch deep into 2015. There have also been some leaked documents about the Treasure Island and FijiGPUs. Treasure Island will be the basis for the R9 370X; with some 1,536 GCN cores it should provide a good deal of extra performance over the R9 270X it will effectively be replacing. The R9 380X will be the Fiji-based card and could be the serious sweetspot for AMD as it is rumoured to have a full 3,072 GCN cores and a 384-bit bus. This could indicate a similar strategy to the one Nvidia used for this last generation, with the R9 380X being the top card with the R9 390X being some crazy, ultra-enthusiast, $1,000 option.

NVM Express Q4 2014

NVM Express, or to give it its full title, Non-Volatile Memory Host Controller Interface (NVMHCI), is a specification for accessing solid-state drives over the PCiExpress bus in a much more efficient way than the current AHCi(Advanced Host Controller Interface) allows for. The specification is going to be a key ingredient for getting the most from the newer SATA Express and M.2 SSDs that are starting to appear on the market. AHCihas been with us for a long time, and is important for legacy compatibility, but it is holding SSDs back. NVMe, however, exploits the low latency and multiple lanes available from PCiExpress to really drive performance. This is helped by the fact that NVMe has been built from the ground up to make the most of how SSDs work. Multiple command queues, being able to handle multiple threads at the same time and specific optimisations for 4KB commands all work together to improve SSDs. SSDs may be speedy already, but once NVMe takes off, there should be a serious shift in their performance. There's an operating system element to NVMe as well, but the good news is that the NVMe Windows Working Group has been beavering away in the background and has managed to squeeze native support for NVMe into Windows 8.1. This means that if you're rocking Microsoft's latest OS then you just need the drive and a compatible motherboard when they do get released.

SandForce SF3700 Q1 2015

LSI's second-generation SandForce controllers redefined the SSD industry at launch. At one point, it felt like every SSD out there had a SandForce controller chugging away at its heart - even Intel jumped on the bandwagon. It is testament to LSithat it has managed to get its controller in so many drives, and testament to the performance on offer. Yes, the compressed data performance hasn't stood the test of time particularly well, and newer controllers from other manufacturers have left SandForce behind, but this is still a company worth watching and capable of some serious performance. The LSiSandForce third-generation controller, the SF3700, is shipping out to OEMs between now and the end of the year and we should start seeing new SSDs based on this controller by the start of 2015. LSiis focusing on combined read and write performance for this new controller - it will be optimised for reading and writing in smaller chunks as opposed to one massive read or write (as commonly tested in benchmarks). Reportedly, LSiis looking to allow manufacturers to disable compression algorithms to produce more consistent performance, regardless of the data. We expect the new SF3700 to stir up the market when it's released, leading to a flurry of launches from all the major players. Given the move over to SATA Express interfaces, this could make for some interesting times, and some seriously high performance.

DDR 4 Q3 2014

The next iteration of RAM, namely DDR4 SDRAM, is already out in the wild, although so far it has only been employed in servers. Even there, it's hardly widespread. Just as night follows day, so DDR4 will gently nudge DDR3 off the specification sheets and into the footnotes of IT history. The big push for desktops comes with the release of Intel's Haswell-E processors, which will support quad-channel DDR4 configurations. Intel Broadwell and AMD Excavator will also support the new standard eventually. DDR4 boasts the potential to increase memory bandwidth by a cool 50 per cent over DDR3, while at the same time slashing power consumption by 30 per cent. This latter point is a key reason why server administrators have been so keen to snap up the technology - it's going to save them money regardless of the initial premium that accompanies any new technology. Just to put some hard figures on this: the operating voltage of DDR4 ranges from 1.05V to 1.2V, while DDR3 starts at 1.2V and then tops out at 1.65V. At launch, you can expect to see the initial slew of memory operating at 2,133MT/s (million transfers per second), which is on a par with the top-end DDR3 memory sticks. However, the specification is rated up to 4,266MT/s, which is where the real performance boost comes in. The 284-pin DIMM modules look similar to the 240-pin modules used for DDR3 and are indeed the same width, but the connectors are smaller to cram them into the same space. The modules are slightly taller and thicker than DDR3 modules in order to make signal routing easier. Of course, there will be a price premium to start with, but with the likes of the Crucial Ballistix Elite 16GB PC4-21300C15 kit starting at ?240, the transition to DDR4 may not be quite as painful as the last generation.

Oculus Rift 2015

When Facebook wanders in and drops a cool $2bn on a headset, you know that there's something more to this whole second coming of virtual reality than just a neat peripheral for a few specific games. And sure enough, even before Facebook appeared on the scenes, plenty of games developers were lining up to get their games in on the action. With the likes of Elite: Dangerous, Star Citizen and Project CARS all jostling for a piece of the Rift, this might be the coolest tech you ever buy. In case you've managed to avoid being sucked into the whole excitement surrounding the device, Oculus Rift is a head-mounted 3D display that tracks the movement of your bonce to give the impression that you're actually in the virtual world, not just viewing it through a window. The final models will boast a 7-inch display with native resolution greater than 1080p, have a responsive head tracking system and be comfortable for long play sessions. We tried the first iteration of the developer kit, and have to say that some of the demos are very impressive, although not without their issues. The screen resolution of the first-generation development kits wasn't great, there was some tracking lag, and we found being unable to see our keyboard and mouse frustrating. The second-gen development units are shipping now, and the head-tracking and screen resolution are improved, but controlling yourself in-game is still an issue. Expect to pay about ?250 for the finished model, but price and release date are not confirmed.

Microsoft Windows 9 Q3 2015

There's no official word on the release date for Windows 9, but several analysts suggest that it should be out before autumn 2015. Microsoft blogger Paul Thurrott has gone one step further, suggesting that it could be with us in April 2015. This is somewhat due to the bad name with which Windows 8 has managed to tarnish itself. It's fair to say that Windows 8 hasn't had it easy, but that's partly because Windows 7 nailed what most of us wanted from an OS. Windows 9 could be the next logical update for many though, so it does make sense that it'll be an earlier release. Windows 9 is expected to keep building on the ethos set out by Windows 8 - that is, unifying the operating system across a wide range of devices. Key to this concept is the idea of creating Universal Windows Apps that function the same across a wide range of hardware. Given the guff that currently fills the Windows Store, this may not be particularly exciting, but we can see that it is essential for realigning Microsoft's ecosystem to its new cross-platform ethos. As for the ill-fated Start Menu, a preview of Windows 9 was shown at Build 2014 that showed a return of Start Menu, albeit one that contained Modern Uielements. Either way, we're seeing that as a win for users. Properly functioning Windows Modern Uiapps are expected to be included at release, along with improved power management code, better integration with the cloud and enhanced support for gestures and other methods of interacting with our devices. We will probably see another interim update for Windows 8 - Windows 8.2, no doubt - before we hit the next big release. This could even be a chance to try out some of the things that Microsoft wants to squeeze into the Windows 9 OS.

DirectX 12 Q4 2015

Microsoft first gave concrete evidence of DX12 at this year's annual Games Developer Conference. The demonstration featured Xbox One racing title, Forza, running on a Titan Black-powered PC. It was a remarkable feat on many levels. Not only is Forza very much an Xbox exclusive IP, but the demo aptly showed just how capable the alpha build of DirectX 12 was. And while you could say that the Xbox One is essentially a PC anyway, that's only from a hardware level - there is a whole mess of operating system to consider. In keeping with the multi-platform plans of Windows 9, the next version of DirectX is being designed to run on a wide variety of platforms. As well as PCs, it will work on the Xbox One, tablets and even phones. Platform-agnostic, if you will. The other major focus of its development is to be a more efficient and thinner API. It's impossible to talk about this side of things without mentioning Mantle, which is AMD's rethinking of graphics APIs to cut down on some of the overheads and make better use of the hardware in our machines. DirectX 12 is going to do for every graphics subsystem (from Nvidia, Intel and AMD) what Mantle can only hope to do for AMD's cards.