We're late with our Ryzen 7 review for a number of reasons, and some might argue that the key story has already been told. We can confirm that the top-tier Ryzen 7 1800X does indeed match and occasionally even exceed Intel's mighty Core i7 6900K octo-core/16-thread monster on a great many benchmarks and heavily multi-threaded workloads. By this measure alone, AMD's return to a competitive footing in the x86 processor market is a remarkable achievement. But equally, our own tests confirm less impressive results posted by many: Ryzen 7's gaming performance does not match up to Intel's on most titles. The question is - why?
One aspect we have to take into account is that the AM4 Summit Ridge platform is in its earliest days. We used two boards during our testing, an MSI X370 XPower Gaming Titanium and an Asus ROG Crosshair 6 Hero - both supplied by AMD - but try as we might, the supplied Corsair 3000MHz DDR4 modules in our press package would not work at anything like their stated frequencies. AMD suggested loosened latency timings and lower bandwidth, but we still couldn't get it to work as it should (despite other reviewers with the same kit getting better results).
Our review would concentrate on gaming performance and as all of our recent CPU reviews have revealed, in-game frame-rates have a close relationship with clock-speed and memory frequency. Things are improving though - GSkill provided new Flare X-branded modules that feature low C14 latency and 3200MHz speed. On our MSI board, equipped with a fresh BIOS, everything just worked - 'A-XMP' set-up almost everything to spec, though DRAM voltage needed to be manually set.
Game on! Or rather... not. There's the sense that despite its lengthy gestation period, Ryzen isn't fully integrated with Windows yet, meaning that AMD wants us to make a number of changes for best performance. Dipping into the Windows 10 power profile, AMD recommends changing the setting from the standard balanced to high performance mode. The Windows high precision event timer needs to be disabled to add a few points to the performance scale. There are other tips AMD proffers, designed to push Ryzen performance higher in gaming scenarios, and we'll discuss those in due course.
Once set-up though, it's clear that Ryzen does indeed meet the hype in many respects, particularly with its multi-threading capabilities. The results across the web demonstrating this are legion, but just to cover the bases, below you'll find our Cinebench R15 results along with h.264 and HEVC Handbrake encoding tests, based on transcoding one of our demanding 4K videos. Ryzen is faster than the 6900K on h.264, but the Intel chip takes point with intense HEVC encoding. HEVC has more of a focus on AVX2 instructions - an area where AMD has an architectural disadvantage. Regardless, the Intel eight-core chip is only 11 per cent ahead. Scalability with HEVC seems to need some work though - the i7 7700K isn't so far behind. That particular i7 may lack cores, but its frequency is higher, which still seems to count for a lot.
But it's the gaming side of things where Ryzen has provoked the most controversy. In terms of single-thread and multi-thread performance, Cinebench tests - and many others - confirm that Ryzen 7 1800X is competitive with a chip Intel sells that costs twice as much, while Ryzen 7 1700X and stand a good chance of overclocking to something close to 1800X performance. With such fundamental validation on one set of number-crunching workloads, it stands to reason that Ryzen should perform in a similar manner on another: gaming - but it doesn't.
First up, a look at our test system and an explanation of how we benchmark CPUs. We're mostly using parts supplied by an AMD - MSI's X370 Xpower Gaming Titanium is a beautiful, top-tier motherboard, but we've swapped out the Corsair 3000MHz Vengeance LPX that we couldn't push beyond 2400MHz with GSkill's freshly minted low-latency C14 3200MHz Ryzen-validated modules. XMP isn't supported natively by AMD (it's an Intel format), which may have led to some of our compatibility issues, but all major manufacturers have their own emulation of the system. In the case of the MSI board, it instantly locked onto the correct timings and frequency for the GSkill modules, but the latest BIOS is required for this level of 'plug and play' compatibility. AMD also supplied an EK Predator 240 closed-loop watercooler - a monstrous piece of kit.
|Core i3 7350K||Core i5 7600K||Core i7 7700K||Core i7 5960X||Core i7 6900K||Ryzen 7 1800X|
|Cinebench R15 Single-Core||184||173||187||133||167||162|
|Cinebench R15 Multi-Core||450||654||963||1310||1460||1605|
|Handbrake 0.10.5 x264||6.4fps||9.6fps||13.1fps||17.5fps||18.5fps||19.6fps|
|Handbrake 0.10.5 x265/HEVC||2.8fps||4.8fps||6.2fps||6.9fps||7.2fps||6.5fps|
We believe that a CPU purchase should last for years, so rather than test processors with a particular GPU at standard gaming conditions, we opt instead to take the graphics hardware out of the test results as best we can and to attempt to concentrate more closely on a processor's gaming potential. The aim here is to ascertain relative performance between CPUs when running game engine code - this gives a better idea of how 'lastable' a potential processor may be. Check out this Kaby Lake CPU performance comparison, for example. It illustrates how much extra frame-rate you get from a CPU relative to the price you pay for it.
To produce these results, we run a game at ultra settings or equivalent at 1080p resolution when paired with an overclocked Titan X Pascal. This ensures that the CPU is tested at its peak doing the jobs it is intended for - processing game logic, calculating animation and physics, as well as one of its most crucial tasks: preparing instructions for the GPU. It's fair to say that Ryzen is competitive in some titles here, but falls short in others. The raw numbers can be sobering though: the top-tier Ryzen is beaten by the i5 7600K in four out of seven titles, and it's significantly off the pace set by the i7s in games like Rise of the Tomb Raider DX12, The Witcher 3 and Ashes of the Singularity's brutal CPU benchmark.
Far Cry Primal is a game that is multi-core aware, but is powered by a dominant single thread. Here we see the more slowly clocked eight-core chips on Intel's older architectures easily beaten by both Kaby Lake processors, but Ryzen is also significantly behind. Crysis 3 remains one of the best multithreaded gaming workouts and acquits itself well with the 1800X (more so than the numbers suggest, as we shall discover) but the disappointing Tomb Raider result is strange: this game scales up adequately on the Intel eight-core processors - something of a rarity - but Ryzen disappoints. The frame-rate result is sub-par even compared to a stock 7600K, but we should stress that a look at frame-times shows less microstutter, a similar situation to an i7. Tomb Raider really does like threads - and lots of them.
Looking at the Crysis 3 test in more detail, the i7 7700K inches to victory over Ryzen, but the nature of the test result is a touch questionable. In detail-intensive, rich scenes, Ryzen pushes ahead. When the visuals are far less complex, the i7 7700K pulls way, way into the lead. Given the choice we would prefer a processor that handles the bottom end better - and in this case, that's Ryzen. However, based on our sample at least, this is an outlier. As things stand, the 7700K looks like the better bet. AMD's response to Ryzen's sub-par 1080p showing is to encourage reviewers and users to test at higher resolutions, where the processor's fortunes are healthier. We weren't so sure that this was a fair test - our initial thought is that this simply brings CPU scores more closely into line by adding the GPU as an artificial limiting factor. Well, that is indeed the case, but this is a new platform with its own strengths and weaknesses, and there's no harm in testing.
|1080p/Titan X OC||Ryzen 7 1800X||Core i5 7600K||Core i7 7700K||Core i7 5960X||Core i7 6900K|
|Memory Frequency||3200MHz DDR4||3000MHz DDR4||3000MHz DDR4||3200MHz DDR4||3200MHz DDR4|
|Assassin's Creed Unity, Ultra High, FXAA||119.4||121.4||132.2||124.4||122.1|
|Ashes of the Singularity, DX12, CPU Test||35.3||29.6||41.9||45.5||46.6|
|Crysis 3, Very High, SMAA T2x||137.5||99.4||138.2||140.2||150.8|
|The Division, Ultra, SMAA||125.3||132.0||133.8||130.1||131.4|
|Far Cry Primal, Ultra, SMAA||91.1||117.2||137.9||101.3||105.6|
|Rise of the Tomb Raider DX12, Very High, SMAA||85.8||89.7||126.5||128.0||129.0|
|The Witcher 3, Ultra, No Hairworks||118.8||97.7||139.4||132.8||134.3|
In re-benching at 1440p, we omitted Ashes of the Singularity's CPU test (it does not stress graphics much, and would not show much change) but by and large, you can see why AMD prefers that we use a higher resolution for gaming benchmarks. Even with an overclocked Titan X Pascal in play, frame-rates are depressed, giving more 'breathing room' to the processor. In short, more of the i7's headroom is left on the table, and Ryzen 7 is able to compete more closely. In a 'real world' gaming scenario, it's unlikely that you'll be using a GTX 1080 Ti or a Titan X, so you should see even closer performance on today's GPUs. However, if you're looking for the most 'lastable' CPU - the benchmarks basically work as a smokescreen in masking the actual top-end performance of each chip.
That said, AMD told us that games could run faster than Core i7 at 1440p even if beaten at full HD, and we have been able to replicate this. Take a look at the Crysis 3 test - there's a telling shot below. Ryzen 7 1800X is very slightly off pace at 1080p, but takes a five per cent lead at 1440p. How is this possible? Well, as we mentioned, non-complex scenes see the 7700K shoot ahead of Ryzen at 1080p, while the AMD chip commands a valuable advantage at the lower end in much more taxing rendering scenarios. By running at 1440p, the GPU overhead cuts off those sky-high leaps the Intel chip enjoys in simpler areas of the test scene, and those complex areas where Ryzen takes point weigh more heavily into the overall average.
The Crysis 3 result is fascinating because the GPU cutting off the i7 7700K's advantage won't just be noticeable at 1440p - it should also manifest with a less capable graphics card at 1080p too. This is a fascinating route forward for further testing with Ryzen, and we've approached AMD for its views on other games that may exhibit this behaviour. But we should be clear that it's not a silver bullet that makes Ryzen 7 the better chip - only Crysis 3 validated AMD's point out of our test titles and we would expect the Core i7 7700K to maintain its leadership in most games.
On top of that, we should also consider the 'stock for stock' nature of this testing. We've paired all systems tested here with fast RAM in order maximise CPU-bound performance on stock clocks, but this actually does the Intel K chips a disservice. For years now, all enthusiast-level boards we've come across automatically invoke all-core turbo when you plug-in overclockable memory and activate XMP. For the Core i7 7700K, this takes the processor to 4.5GHz out of the box with no further user tweaks required. On top of that, overclocking Kaby Lake to 4.8GHz is usually a simple matter of typing '48' into the core multiplier box on any Z170 or Z270 board - some samples even reach 5.0GHz. Put simply: there's headroom here. So how does Ryzen compare?
|1080p vs 1440p/Titan X OC||Ryzen 7 1800X 1080p||Core i7 7700K 1080p||Ryzen 7 1800X 1440p||Core i7 7700K 1440p|
|Memory Frequency||3200MHz DDR4||3000MHz DDR4||3200MHz DDR4||3000MHz DDR4|
|Assassin's Creed Unity, Ultra High, FXAA||119.4||132.2||87.5||91.2|
|Crysis 3, Very High, SMAA T2x||137.5||138.2||115.9||110.4|
|The Division, Ultra, SMAA||125.3||133.8||96.2||101.0|
|Far Cry Primal, Ultra, SMAA||91.1||137.9||91.6||107.1|
|Rise of the Tomb Raider DX12, Very High, SMAA||85.8||126.5||86.1||100.5|
|The Witcher 3, Ultra, No Hairworks||118.8||139.4||108.1||112.1|
It's safe to say that even an overclocked Titan X Pascal running at 1080p will find its limits eventually, but what's surprising is that it happens on so many games. Assassin's Creed Unity and The Division are topped out with the 7700K at 4.8GHz, while the other titles showing diminishing returns. As easy as it is to overclock the i7 7700K, simply plugging 3000MHz DDR4 and engaging XMP will do the vast majority of the work for you, giving an easy 4.5GHz all-core lock, paired with the memory bandwidth required to get the most out of it. Multi-core aware titles like Ashes of the Singularity, Crysis 3 and Rise of the Tomb Raider do see increases though.
The Core i7 6900K sees some decent improvement with its 4.4GHz overclock too - single-core dependent titles like Far Cry Primal gain significant momentum, while multi-core games like Ashes of the Singularity and Crysis 3 push on to achieve results at the top of their class. It's not a ringing endorsement bearing in mind the processor's colossal price-point, but we should see similar results from the six-core i7 6800K, which is a fair bit cheaper (unfortunately a six-core sample was not available for testing for this review).
Ryzen's stock clocks are pushed far closer to its limits, and there's not as much overclocking headroom. We started to worry about voltage and heat at 4.1GHz (Ryzen was toastier than the 6900K here, even though it had a meatier cooler) so we settled at 4.0GHz across all cores. Frame-rates improve, but not to a gaming-changing degree. Even Far Cry Primal - which craves single thread performance boosts - only sees a seven per cent boost. Overall, where Ryzen is competitive with the i7s, it remains competitive when comparing overclocked results, but you can't OC to significantly overcome performance deficits in titles where the new CPU falls short.
While the Ryzen 7 1800X is going to be the top-binned processor in the stack, initial tests suggest that overclocking headroom on the 1700X and even the much cheaper 1700 should be similar to the top-tier product. We'll be reporting back on that shortly, but assuming we get similar results, it may well be the case that you could save a lot of money by investing in a cheaper chip - the caveat being that you should expect significantly higher power consumption and more heat. Regardless of which Ryzen 7 you choose, you will require a decent cooler if you're looking to achieve a good all-core overclock.
|1080p/Titan X OC||Ryzen 7 1800X||Ryzen 7 1800X 4.0GHz||Core i7 7700K||Core i7 7700K 4.8GHz||Core i7 6900K||Core i7 6900K 4.4GHz|
|Memory Frequency||3200MHz DDR4||3200MHz DDR4||3000MHz DDR4||3000MHz DDR4||3200MHz DDR4||3200MHz DDR4|
|Assassin's Creed Unity, Ultra High, FXAA||119.4||120.8||132.2||132.9||122.1||122.6|
|Ashes of the Singularity, DX12, CPU Test||35.3||36.8||41.9||44.1||46.6||52.3|
|Crysis 3, Very High, SMAA T2x||137.5||143.8||138.2||145.5||150.8||163.5|
|The Division, Ultra, SMAA||125.3||128.4||133.8||133.9||131.4||133.1|
|Far Cry Primal, Ultra, SMAA||91.1||97.2||137.9||140.1||105.6||118.2|
|Rise of the Tomb Raider DX12, Very High, SMAA||85.8||89.9||126.5||131.0||129.0||129.6|
|The Witcher 3, Ultra, No Hairworks||118.8||121.3||139.4||145.2||134.3||146.2|
As the results came in, one had to wonder: at the end of the day, a CPU is a number-cruncher. Ryzen acquits itself beautifully in a vast range of workloads, but falls short to a certain degree in gaming. And yet, running game logic and preparing draw calls is still an exercise in number-crunching, so why do we see this disparity? The eight-core Ryzen processor is essentially two quad-core modules (CCXs) integrated and connected via a fabric that runs at a speed determined by the DDR4 memory. The theory goes that faster memory could see dramatic gains in gaming performance as the two CCXs could communicate faster.
However, across our test games, the boost in performance from extra memory bandwidth was ballpark with the Core i7 7700K - which has no such interconnectivity fabric. The Witcher 3 loves memory bandwidth and gets a 20 per cent boost with faster RAM on both systems. Crysis 3 has lower dependency - there's a four per cent boost on the 7700K and seven per cent with the 1800X. This does not seem like much of a smoking gun when two very different CPU architectures hand in very similar results. It points more to the software's dependency on memory bandwidth.
More complex is the topic of how Windows allocates CPU threads, and much suspicion has been pointed at Windows' core scheduler (for its part, AMD has ruled this out). Two key pieces of evidence are worthy of investigation. Firstly, AMD itself has recommended that we test with SMT (hyper-threading essentially) on and off, reporting performance gains in some scenarios - this would essentially force Windows to allocate code to physical cores, not the 'hyper-threaded' ones. Secondly, some users suggest that Ryzen runs games faster on Windows 7 x64. Both of these factors could be tested and from there, we should be able to figure out if there is indeed an issue with how work is distributed across cores.
There is clear variance between Windows 7 and Windows 10 performance, as well as improvements with SMT on or off - but again, no smoking gun suggesting a common problem with the core scheduler. Again, Far Cry Primal shows up some fascinating results in each permutation, while Crysis 3 on Windows 7 actually seems to run as fast as the 4.0GHz overclock in Windows 10. However, the notion that Microsoft's newer operating system is slower for Ryzen than its older offering is not backed up by our testing - some games are faster, fewer are slower. It's worth pointing out that Windows 7 has no official Ryzen support - and while disabling SMT can produce increases in Ryzen performance, the notion of doing so on a per-game basis is overly onerous for the user. Our advice? Stick with Windows 10, turn off the high precision event timer, engage the high power profile, keep SMT on and consider fast, Ryzen-validated DDR4.
|1080p/Titan X OC||Ryzen 7 1800X/SMT On||Ryzen 7 1800X/SMT Off||Ryzen 7 1800X/SMT On||Ryzen 7 1800X/SMT Off||Ryzen 7 1800X/SMT On|
|Memory Frequency||3200MHz DDR4||3200MHz DDR4||3200MHz DDR4||3200MHz DDR4||2133MHz DDR4|
|Assassin's Creed Unity, Ultra High, FXAA||118.1||123.5||119.4||123.0||112.2|
|Crysis 3, Very High, SMAA T2x||143.1||132.9||137.5||123.6||128.6|
|The Division, Ultra, SMAA||123.8||124.3||125.3||129.8||125.5|
|Far Cry Primal, Ultra, SMAA||100.9||107.4||91.1||105.0||79.2|
|The Witcher 3, Ultra, No Hairworks||109.2||111.4||118.0||115.8||98.9|
If there is an issue about how gaming workloads are distributed across cores, the most interesting evidence we uncovered concerned gaming performance results when we used the MSI BIOS to disable one core in each CCX, giving games six cores and 12 threads to work with. This is the '3+3' configuration that's being deployed on the Ryzen 5 1600/1600X, arriving next month. The 1600X actually has access to the same amount of cache memory as the 1800X and runs at the same base and boost frequencies. The results of the BIOS tweak are quite remarkable. Shorn of two cores and four threads, games only lose between three to seven per cent of the performance of the fully enabled eight-core chip. Assuming this is indeed representative of the upcoming Ryzen 5 1600X's turnout (and this can only be confirmed with actual hands-on time with the product itself), AMD may well have a highly compelling couple of mainstream products waiting in the wings that could make you think twice about a prospective Core i5 or indeed a Ryzen 7 purchase - but that's a discussion for another time.
In the here and now, Ryzen remains a capable gaming performer, but the fact it's so good at some workloads while falling short in others is a puzzle wrapped in an enigma, and the disparity has yet to be explained - and we did ask. AMD released a couple of statements from Oxide Games (Ashes of the Singularity) and Creative Assembly (Total War) telling us to expect optimisation pushes for Ryzen to get more performance from the new architecture. Obviously this is welcome, but PC owners go into an upgrade expecting a big out-of-the-box improvement, not 'jam tomorrow'. Back catalogue titles likely won't receive this attention, while the immediate future for platform-specific optimisations is uncertain. Longer term, we can pretty sure that Ryzen optimisation will gain traction - we're confident that the architecture will appear in the next console generation.
Ryzen 7 1800X - the Digital Foundry verdict
AMD clearly gets a lot right with Ryzen. The 1800X absolutely trounces productivity and multimedia workloads, and Intel's much more expensive chips only remain at the races here owing to the ability to hit higher all-core overclocks. We do a lot of video editing and encoding work at Digital Foundry, and everything we've seen so far suggests that Ryzen 7 offers the best route forward when it comes to upgrading our systems. There are other key advantages that cannot be overlooked either: all Ryzen processors are multiplier-unlocked, while memory/processor overclocking isn't restricted to the the top-tier enthusiast motherboards as it is with the competition.
Other aspects also deserve praise: Ryzen's eight-core TDPs are lower than Intel's, power consumption under load on our 1800X system was identical at 184W, while idle power consumption was much, much lower than the Intel X99 rivals (motherboard choice may be a factor here, of course). This bodes well for Ryzen's success in enterprise markets and elsewhere: cybercafes, for example. AMD's lower overclocking headroom is unlikely to be a concern for those users.
It's also worth stressing that while gaming performance does lag behind Intel by and large, this does not make Ryzen 7 processors 'bad for gaming'. The Witcher 3's Novigrad City is one of the most processor-intensive areas of the game and the 1800X is still delivering a 119fps average. Our results show that the i7s AMD is targeting with Ryzen 7 are potentially longer-lasting for gaming - a key factor in any CPU purchase - but in the here and now, it's abundantly clear that Ryzen still does the job. Clearly PCs are used for more than just gaming, so if play is just one part of your requirement from a processor, Ryzen's undoubted capabilities elsewhere may make it the more logical choice than an i7 - and at these prices, the Ryzen 7 line's excellent overall performance simply cannot be ignored.