1. Intel Optane 900P SSD review

Cast your mind back to the autumn of 2015 and you might just remember how Intel and Micron were claiming that their new memory technology, 3D XPoint Memory, would revolutionise computer storage, potentially placing both NAND in SSDs and DRAM in system memory.

For those of you who can’t remember, 3D XPoint is so exciting because it addresses the downsides of NAND and DRAM. For example, while NAND is thousands of times faster than a HDD, is non-volatile so retains data when the power is cut and is much cheaper than DRAM, it has limited endurance and is much slower than DRAM. In contrast, while DRAM is much faster than NAND and has essentially unlimited endurance, it is much more expensive and is volatile, so loses data when the power is cut.

In contrast, 3D XPoint stores data in individually accessible cells that are formed from an array of cells that are written to via a change in voltage, so unlike NAND it doesn’t have to refresh a whole page of memory even if just a single bit is being written. As a result, 3D XPoint has much lower latency and greater endurance than NAND.

While 3D XPoint has yet to make it into DIMMs yet, the first Intel SSDs using the new memory technology are now available and are known as the Optane 900P series. The initial line up comprises three models, two half-height 4x PCI-E cards and a 2.5in U.2 drive.

The table below highlights the key specs of the new Intel Optane 900P SSDs versus the Samsung 960 Pro, the most popular high-end NAND SSD on the market.

The specs reveal some intriguing differences between the Optane 900P and the 960 Pro. To start with, the 960 Pro has a significantly faster sequential transfer rate, although you need to careful comparing numbers from different manufacturers, as they don’t always specify the conditions under which performance was measured. For example, SSD manufacturers typically quote very high queue depths and block sizes as this generates very high data transfer rates.

One of the key selling points of 3D XPoint is its lower latency than NAND, and Intel’s claims of 10µs are pretty gobsmacking. Unfortunately Samsung doesn’t publish latency numbers for the 960 Pro, but to put the 900P’s 10µs into context, Intel’s own fastest NVMe NAND drive the 750 has a latency of 20µs while its fastest SATA NAND drive the 525 has a latency of 80µs, both of which are significantly slower than the 900P. What’s more NAND SSDs also suffer from additional latency as writes are buffered, whereas the 900P starts writing data straightaway.

The specs also show that while Intel and Samsung’s drives have a 5-year warranty, there’s an enormous difference in the endurance rating. This number is the total amount of data that can written to a drive before the cells wear out and in this instance shows one of the key benefits of 3D XPoint over NAND, with the Optane 900P have over 1000% more endurance than the 960 Pro. While this amount of endurance is arguably overkill and unnecessary for a typical gaming PC, if you’re doing content creation work such as editing videos and graphics then a high endurance drive will be of benefit.

To see how all those tech specs translate into real world performance we benchmarked the new Intel Optane 900P SSD head to head some of the most popular drives on the market. The other drives on test were the aforementioned 512GB Samsung 960 Pro SSD, 500GB Samsung 850 Evo SSD and a 2TB Seagate BarraCuda hard drive. All the cards were tested in one of our 3XS gaming PCs which had an Intel Core i7 8700, Asus Prime Z370-P motherboard and 16GB of Corsair Vengeance DDR4 memory.

The first benchmark that we ran was the synthetic storage application CrystalDiskMark. The first two sets of results are run with a transfer size of 128K, queue depth of 32 and single thread, with the 960 Pro posting the best read by far and the 900P the fastest write by a small margin. These results are interesting and pretty much mirror the marketing specs of each drive, but almost no desktop applications or games operate in this way so these numbers aren’t really indicative of real world performance, more on this later.

The second scenario in CrystalDiskMark drops the transfer size down to 4K, although the queue depth of 32 is still extremely high. Even so, this more strenuous test begins really started to separate the men from the boys, with the 900P taking clear first place in both the read and write tests with a clear 20% lead over the 960 Pro.

The final scenario test we ran in CrystalDiskMark kept the transfer size at 4K but dropped the queue depth to one, simulating how many games load data when starting up. Under these conditions the 900P was the clear winner when reading, transferring data a ludicrous 522% faster than the 960 Pro. The results were closer when writing, but even so the 900P was still a massive 43% faster than the 960 Pro.

We also ran the storage tests that form part of the SPECwpc benchmark that is used to measure the performance of workstations. The storage tests in SPECwpc use a trace of different applications which are then are simulated by the Iometer utility across four different application types. In all four applications types the 900P turned in by far the best results with over a 300% performance lead the 960 Pro, let alone the other drives. These are important benchmarks for the 900P to perform well in as one of the main target markets of Optane SSDs is workstations, so its great to see the drive excelling at these workloads.

As the 900P turned in a mixed set of results in the synthetic CrystalDiskMark benchmark with some great and some middling results we also wanted to see if it could accelerate loading games. The two games we tested were chosen because they have huge datasets and so are a big challenge for any storage device as shown by the snore inducing it took the hard disk to load both games. What we discovered is that pretty much all SSDs take the same amount of time to load games. Sure, the more expensive 900P and 960 Pro were a bit faster than the 850 Evo, but less than a handful of seconds isn’t enough of a difference to get too excited about.

Conclusion
It’s great to see SSDs using 3D XPoint Memory finally making their way onto the market, as the hype around this new memory technology was enormous a couple of years ago. The benchmarks for the new Intel Optane 900P SSD paint an interesting picture, and one that needs to careful evaluation. For instance, in commonly used synthetic tests with huge queue depths the 900P is faster at writing, but slower at reading than the market leading high-end NAND SSD the Samsung 960 Pro. However, at the lower queue depths that most desktop applications use, the 900P is much faster than the 960 Pro, so it really depends on what you use your PC for.

Digging deeper, the SPECwpc benchmarks gave the 900P a massive lead over the 960 Pro showing that for workstation applications there really is a big difference in performance between the drives. On the other hand, the 900P was only marginally faster at loading games than the 960 Pro.

At the time of writing the 900P will set you back £1.23 per GB, whereas the 960 Pro costs £0.58 per GB, so clearly you’re being asked to pay a significant premium for 3D XPoint memory. In terms of pure performance that premium makes it hard to justify buying a 900P for a gaming PC unless you’ve already maxed out every other component in your PC, as upgrading the CPU or GPU is likely to make more of an appreciable difference. That said, if you’re speccing up your dream PC then a 900P should clearly be at the top of your list.

For workstations the 900P is more compelling, especially when you consider that 3D XPoint has such amazing endurance that you’ll be able to get more use out of the drive compared to a conventional NAND-based SSD. These same characteristics make the 900P really interesting for servers, although the 280 and 480GB capacity is rather limiting. Still, we’re sure that Intel have bigger 3D XPoint drives on the way. In the meantime, the Intel Optane 900P, if you can afford it, is the new king of performance SSDs.

2. NVIDIA TITAN V

NVIDIA won the prize for announcing the last major product of 2017 with its new Titan V graphics card. Based on the same Volta architecture as the mighty Tesla V100, the Titan V is aimed squarely at deep learning and AI workloads, boasting an incredible 110 teraFLOPS of performance in FP16 calculations, compared to just 0.19 teraFLOPs for the existing Titan Xp.

This incredible throughput is thanks to the Volta architecture having both CUDA cores as well as a new programmable type of core, known as Tensor, that run alongside standard CUDA cores. Tensor cores can perform 4x4 Matrix operations in one unit, significantly boosting performance in FP16 and FP32 calculations. For instance, a single Tensor core produces the equivalent of 64 FMA operations per clock, equivalent to 1024 FLOPs per SM, compared to just 256 FLOPs per SM for standard CUDA cores. The Titan V is armed with 5,120 CUDA cores, 640 Tensor cores plus the latest HBM2 memory which provides significantly more bandwidth than standard GDDR5 or GDDR5X.

Retailing for a cool £2,700 the Titan V is considerably more affordable than a Tesla V100, so is a great option for developing deep learning and AI without paying the premium for a datacentre solution.

3. Intel with AMD inside

Intel recently unveiled that it’s joining up with AMD to produce a new range of Intel CPUs with integrated AMD GPUs.

This is made possible by the new EMIB (Embedded Multi-Die Interconnect Bridge), allowing the two processors plus the VRAM to be mounted together in one package, communicate with one another and share power. The combined package will take up much less space than the components inside a traditional gaming laptop, in which the motherboard has three physically separate items, the CPU, GPU and VRAM.

Intel is planning to launch four Core i7s and a single Core i5 processor with the new integrated AMD graphics. At a CPU level all the processors are remarkably similar, all being quad-core with HyperThreading technology with support for dual-channel DDR4 memory. Where the differences arise are in the GPU. The two top-end Core i7s will be equipped with a custom AMD Radeon RX Vega M GH GPU, which has 1536 stream processors running between 1063 and 1190MHz. The two lower-end Core i7s and the single Core i5 will have a somewhat less powerful GPU, the AMD Radeon RX Vega M GL GPU, which has 1280 stream processors running between 931 and 1011MHz. Both GPUs will be equipped with 4GB of HBM2 memory.

To put these numbers into context, the least powerful discrete Vega GPU, the Vega56 has 3584 stream processors and 8GB of HMB2 memory and provides good frame rates on a 2560 x 1440 monitor. It’s not unreasonable therefore to expect to see good frame rates from the new integrated Vega GPUs at 1920 x 1080. In fact, Intel’s press briefing claimed that in games such as Hitman and Deus Ex: Mankind Divided a Core i7 with Vega M GH graphics is up to 40% faster than a Core i7 laptop with NVIDIA GeForce GTX 1050 graphics.

You can expect to see laptops and small form factor desktop PCs with the new Intel processors sometime in the first half of 2018. Watch this space for more info.

4. AMD unveils its plans for 2018 and beyond

AMD had a very busy year in 2017, with a long awaited comeback in the CPU market with its Zen based Ryzen processors and the Radeon Vega graphics cards. 2018 looks to be another busy one, with lots of new products planned.

The first new products out of the gate will be two new APUs, codenamed Raven Ridge. Based on the Zen architecture, these quad-core processors will include Radeon Vega graphics and could be an interesting option for entry-level gaming PCs. You can expect to see the Ryzen 5 2400G and Ryzen 3 2200G from February.

Next on the calendar is Zen+, which is a die-shrink for Ryzen 3, 5 and 7 CPUs, moving from the existing 14nm process to a smaller 12nm process. This should help to boost clock speeds, but there will also be a few architectural tweaks such as Precision Boost 2 which will allow the CPU to dynamically boost individual cores, whereas Precision Boost can only boost one, two or all cores. This should significantly improve Ryzen’s game performance, especially at lower resolutions, so it will be very interesting to see how much of a real world benefit it actually provides. Zen+ CPUs should work in existing Ryzen motherboards, but AMD is also lining up a new X470 chipset although it’s not clear what the benefits over X370 are at this stage.

AMD also plans to expand its mobile line-up in 2018, with more Ryzen U-series chips on the way, plus a cut-down RX Vega CPU for gaming laptops. Finally, while there’s still no sign of a new gaming GPU, AMD is working on a new Vega GPU for deep learning and AI. There aren’t any specs for this GPU yet, but given that it’s going to made using a 7nm process rather than the 14nm of current Vega GPUs it potentially could be extremely fast and power efficient.

5. New and improved HTC Vive Pro VR headset

With its room-scale experience and huge software library the HTC Vive has proven to be one of the most popular VR headsets. HTC has announced that a new and improved version is in the works, the Vive Pro. The most notable improvement will be clarity, with the Vive Pro having a resolution of 1440x1600 per eye, compared to 1080x1200 per eye for the original Vive. The Vive Pro will also have improved ergonomics and a headphone amplifier with active noise cancellation. There’s no word on pricing or a launch date yet, but as soon as we know more we’ll let you know.