CPU by stefan @ 2017-03-04
The AMD Zen architecture has finally materialized into the Ryzen series of processors! We will take a closer look at the new platform from AMD which does incorporate the Ryzen 7 1800X flagship octa-core CPU along with the ASUS Crosshair VI Hero AM4 motherboard and see if it can dethrone the long-standing Core i7 6900K king from Intel Corporation.
At first, we would like to thank AMD for sending out their AMD Ryzen™ 7 1800X octa-core processor for testing and reviewing.
“Advanced Micro Devices, Inc. (AMD) is an American multinational semiconductor company based in Sunnyvale, California, United States, that develops computer processors and related technologies for business and consumer markets. While initially it manufactured its own processors, the company became fabless after GlobalFoundries was spun off in 2009. AMD's main products include microprocessors, motherboard chipsets, embedded processors and graphics processors for servers, workstations and personal computers, and embedded systems applications.
AMD is the second-largest supplier and only significant rival to Intel in the market for x86-based microprocessors. Since acquiring ATI in 2006, AMD and its competitor Nvidia have dominated the discrete Graphics Processor Unit (GPU) market.”
On the desktop CPU market scene, we are usually facing with a total domination from Intel, which did not happen only on the high-end with their Broadwell-E 14nm 6-to-8 core microprocessors but also on the mainstream with Skylake but also the recently launched Kaby Lake. Most of the forums that do recommend components for system builds are aiming towards Intel by default and AMD processors have been rarely a choice of enthusiasts; this lack of competition also caused an artificial price hike from Intel’s’ part and we have been used with seeing increased prices year-by-year.
In December 2016 AMD’s CEO Dr. Lisa Su did surprise us with the introduction of Zen architecture, which materialized into the Ryzen processors; the new architecture was described as a “clean-sheet” which means that it is basically completely new, a rare event in the semiconductor industry. These new SKUs are aimed towards the recent Intel counterparts, the first wave consisting of the Ryzen 7 1800X, Ryzen 7 1700X and Ryzen 7 1700 which do sport 8 core and 16 threads. These are backed by the same 20MB of L2+L3 cache, have different operating speeds, while the Ryzen 7 1700 SKU does also come with an incredibly low TDP of just 65W (judging by its total number of cores).
With the Ryzen 7 1800X flagship CPU ($499 MSRP), AMD is aiming directly towards the Core i7 6900K Broadwell-E, while the Ryzen 7 1700X ($399 MSRP) performance is placed by AMD from the internal tests between the Core i7 6800K and the Core i7 6900K.
The naming of the new Ryzen series has been carefully thought out:
Let us get into more detail! With these new processors, AMD is aiming towards PC gaming industry which is growing at a fast pace, multi-core processing which is great for game streaming, YouTube creators or 3D rendering environments such as Blender. DirectX 12 and Vulkan are new technologies that respond well to multiple cores and Ryzen was developed in order to serve content creators that means a competent IPC and very good multi-core scalability. For the mainstream users, going for an Intel solution which does sport more than four physical cores currently means a very big price increase, and this situation has created a serious limitation in the industry.
The new AMD Ryzen 7 1800X flagship CPU does feature the following highlights:
The new Zen microarchitecture from AMD does focus on four different areas: performance, throughput, efficiency but also scalability.
Regarding performance, the new Zen microarchitecture represents a very big leap in core execution capability versus the previous designs from the same company: Zen come with a 1.75X larger instruction scheduler window and 1.5X greater issue width and resources. This practically allows Zen to schedule and send more work into the EUs. Thanks to a new micro-op cache, Zen is allowed to bypass L2 and L3 caches when using frequently accessed micro-operations. The neural network-based branch prediction unit from the Zen microarchitecture does allow for more intelligent preparation of optimal instructions and pathways for future work.
Changes have been also made regarding the cache hierarchy with dedicated 64KB L1 instruction and data caches, we do have 512KB dedicated L2 cache per core and 8MB of L3 cache shared across four cores. The cache is enhanced with a learning prefetcher that speculatively harvests application data into the caches so they are practically available for immediate execution. These changes are assuring up to 5X greater cache bandwidth into a core. This type of design enhances the Zen architecture's throughput.
When talking about efficiency, the new Ryzen processors are built on the more power-efficient 14nm FinFET process; in more detail, the Zen architecture is using the density-optimized version of the Global Foundries 14nm FinFET process and this fact permits for smaller die sizes and lower operating voltages. The new Zen microarchitecture does incorporate some of the latest low-power design technologies:
-micro-op cache for reducing power-intensive faraway fetches
-aggressive clock gating to zero out dynamic power consumption in minimally utilized regions of the core
-a stack engine for low-power address generation into the dispatcher.
Moving on to the scalability aspect, Zen architecture does start with the CCX (CPU Complex) which is a native 4C8T module; each CCX does come with 64K L1 I-cache, 64K L1 D-cache, 512KB of dedicated L2 cache per core and 8MB of L3 cache shared across all cores. Each core that is contained in the CCX may optionally come with SMD for additional threads.
Depending on the SKU, more than one CCX can be present in a Zen-based product; when strictly referring to the Ryzen 7 Series, these do come with two CCXes, which does mean a total of 8 cores and 16 threads. Individual cores within the CCX may be disabled by AMD and the CCXes communicate across the high-speed Infinity Fabric. Thanks to this modular design, AMD can scale core, thread and cache quantities as necessary in order to cover all the market segments.
What is the previously mentioned Infinity Fabric you may ask? Well, it is a flexible and coherent interface that allows AMD to quickly and efficiently integrate an advanced IP portfolio into a cohesive die. These assembled components can utilize the so-called Infinity Fabric to exchange data between CCX modules, system memory and of course, other controllers such as I/O, PCIe and so on, present on the AMD Ryzen SoC design.
Thanks to the Infinity Fabric, Zen architecture is given powerful command and control capabilities that allows real-time estimations and adjustments to core voltage, temperature, socket power draw, clock speed and even more!
The presentation AMD held in December also mentioned of the AMD SenseMI technology: this is a package of no less than five related “senses” that are relying on learning algorithms and/or the command-and-control functionality of the Infinity Fabric, in order to empower AMD Ryzen processors with machine intelligence.
Pure Power
Thanks to the integrated network of smart sensors that are driving Precision Boost, processor power consumption can be carefully adjusted with any given workload. The telemetry data from the Power Power optimization loop does allow each Ryzen processor to inspect the characteristics of its own silicon, in order to extract individualized power management.
Precision Boost
By using current/temperature/load information fed by the Infinity Fabric, Precision Boost is modulating an AMD Ryzen processor in 25Mhz increments; this type of granular clock speed control is offering the Ryzen processor a greater operational freedom in order to reach ideal frequency targets and at the same time allows for finer dithering at that ideal target.
Extended Frequency Range or XFR
XFR is kicking in when high-performance cooling systems are installed on the AMD Ryzen processors and lifts the maximum Precision Boost frequency beyond the ordinary limits. This is achieved by reading and forecasting AMD Ryzen processor’s distance to TJMax, then converting the available headroom into extra frequency. For non-X SKUs (such as the Ryzen 7 1700), XFR will add 50MHz extra to the stock frequencies, while X SKUs (such as the Ryzen 7 1800X, Ryzen 7 1700X) will add 100MHz extra to the stock operating frequency when the said conditions are met.
Neural Net Prediction
Every AMD Ryzen processor holds a true artificial intelligence inside which harnesses a neural network for learning in real-time the applications’ behavior and speculate on its next moves. Thanks to this feature, the AI readies vital CPU instructions in advance for tackling a new workload.
Smart Prefetch
Thanks to the integrated sophisticated learning algorithms, internal patterns and behaviors of applications are understood, so they can anticipate what data is needed for fast execution in the future. Data is fed into local cache, so it is ready for immediate use.
AMD Ryzen and Streaming
Along with eSports, the streaming popularity has increased exponentially; at the heart of this activity lies the well-known Twitch platform that allows people at home to watch live gaming sessions but also interact in the chat room. Radeon ReLive is one of the tools that make possible broadcast of live game footage, audio, webcams, overlays and other multimedia contents to legions of fans.
Since some of the games are quite CPU-intensive to run and combining this with the fact that streaming adds quite a bit of load to the processor, we will get lots of input lag or other performance-related issues. Twitch is proposing a solution to this issue, which implies using two different PCs in order to split the workload: one system plays the game, while the second system with a capture card receives output from the GPU and serves as a dedicated broadcasting system to alleviate performance bottlenecks.
The GPU live video encoding does not seem to please the streamers since GPU encoders do need more bitrate to achieve the same quality as the CPU-based x264 encoder pre-configured on streaming packages such as OBS and XSplit so when talking about the tight 3500Kbps bitrate limits of Twitch, the GPUs are at a disadvantage.
AMD Ryzen 7 1800X CPU simply solves this issue by having 8 physical cores and 16 threads, so it can dedicate a full 4C8T to both encoding and gaming workloads.
The AMD AM4 Platform
The new AMD Ryzen processors would not operate by themselves, unless paired with the AM4 platform. This new platform consists of six chipsets that can be interchangeably paired with the new 1331 socket; this aspect allows motherboard manufacturers to craft different models in order to cover all market segments: entry level (A320 or A/B300), middle class (B350) and premium (X370 or X300).
These new solutions do incorporate the latest technologies such as NVMe PCIe 3.0 x4, SATA, SATA Express, dual channel DDR4, native USB 3.1 Gen 2 and more.
The Socket 1331 streamlines AMD’s socket infrastructure (AM3 and FM2+) into a single part which can host the AMD Ryzen processor, the 7th Generation APU or the future “Raven Ridge” APU based on the Zen architecture. AMD is intending on using this new socket through 2020 even with the introduction of new technologies such as DDR5 or PCI Express Generation 4.
The storage and I/O options that have been just described are extra to the SoC design of Ryzen CPU, which does also feature natively:
-4xUSB 3.1 Gen 1
-16 lanes of PCI Express Gen 3 for graphics (2x8 mGPU supported on X370)
-4x PCI-Express Gen 3 suitable for a high-speed NVMe SSD or other companion card
-4x PCI-Express Gen 3 for chipset communication (free for re-use along with X300 chipset)
Right from the introduction in December 2016, AMD had promised that all mid-range and high-end motherboards based on the mid-range B350 and high-end X370 chipsets will expose the full-range of multiplier voltage control built into the AMD Ryzen processors. Entry-level motherboards, which have weaker VRM designs, are built to run at stock speeds, so A320 chipsets are locking the voltage and multiplier adjustments.
With the presentation of the AM4 platform, we also need to clarify the cooler compatibility; AM4 comes with a wider bolt-through mounting pattern in order to accommodate the extra pins of the Socket 1331 versus the older FM2+ and AM3 designs. AMD has discussed with no less than 15 of the top cooling manufacturers such as EKWB, Phanteks, Noctua, Corsair, Cooler Master and so on in order to provide mounting kits which will enable usage of the previously-launched solutions on the new AM4 platform. What is interesting is that the current AM3/FM2+ coolers that use the clip system for attachment to the socket retention brackets are fully compatible with the AM4 platform, with no modification!
After a bit of delays, we were happy to see our door the parcel coming from AMD, containing the most recent Ryzen goodies!
I bet you have seen online many unboxing photos and videos but in our case, the bundle was contained inside the motherboard box, which in our case was the high-end ASUS Crosshair VI Hero. ASUS know their stuff with the ROG products, so the product comes with a specific box art:
After lifting the top box cover, we have discovered the Ryzen CPU sitting inside a small cardboard enclosure, a small leaflet that is inviting the publications to try out the latest AMD technology and, of course the ASUS motherboard:
On the back side of the previously mentioned leaflet, AMD links us to some more resources:
The original AMR Ryzen 7 1800X box is a bit bigger (was not included in our bundle), but we have received the main essentials, which were located in there: the CPU along with the case sticker:
These two items are to be found in a blister packaging, after removing the cardboard layer:
The included Ryzen 7 sticker looks very cool on any new computer system case:
A closer look at the flagship Ryzen 7 1800X CPU does reveal a central logo that uses familiar fonts (from the presentation in December 2016). The Ryzen CPUs are using solder between the HSF and the cores, for an optimal heat conduction to the CPU cooler:
In the top left corner, we can read the full name of the processor:
The lower left corner does house some serial numbers:
On the back side of the CPU PCB, we will find 1331 golden pins, and its specific pin placement does not allow wrong installations on the motherboard socket:
After removing the CPU from its original packaging, it was the time of the motherboard to be inspected in order to build the system:
The ASUS Crosshair VI Hero comes with the X370 high-end chipset from AMD, but also includes some of the latest technologies such as ROG Water Cooling Zone, Aura Sync RGB LED, SupremeFX audio, Intel Gigabit Ethernet with LANGuard and GameFirst technologies, USB 3.1 Front Panel connector, M.2 but also USB 3.1 Type A/Type C. The board does also support multi-GPU SLI/CFX setups and lists official memory support up to DDR4 3200MHz in overclocked mode:
What we did also like about this motherboard is the fact that it does also support AM3 cooling systems, so you can re-use your high-performance air/water cooler from the previous build:
We won’t concentrate our attention too much in this article on the motherboard, because it will have a dedicated article later on. Mounting the Ryzen 7 1800X CPU inside the socket is a piece of cake:
For cooling this platform, Cooler Master has provided us with their MasterLiquid Pro 280 AIO, so we went ahead and installed the backplate along with the long screws:
After applying a bit of thermal compound, we have secured the pump into place:
Another bundle component you have probably seen from the previous Ryzen unboxings is the Corsair Vengeance LPX 2x8GB 3000MHz kit; we have also received in in order to use it along with the flagship CPU:
The memory kit is also supplied in a blister packaging:
On the frontal area of the modules, we will get to see a nice Vengeance LPX logo:
The back side though is listing the full technical specifications of the modules (3000MHz operating frequencies, 15-17-17-35 latencies and 1.35 operating voltage). The AMD platform does not support Intel XMP so we are invited to dial-in these timings manually:
The next step was to install the system inside the case, which did also meant we have had to secure the water cooling system and install the video card along with the PSU cables, extra fans, frontal panel connectors, storage drives and so on:
Before actually testing out the system, we made sure that we were operating with the latest ASUS BIOS from the manufacturer’s website:
We then navigated inside the BIOS and made sure that everything was left on AUTO, since we did intend on using AMD’s own overclocking utility, which is named “Ryzen Master”. Ryzen Master is a very handy tool for modifying the core speeds of the Ryzen processor on-the-fly, along with the CPU operating voltage, MEM VDDIO, MEM VTT or VDDCR SOC voltage. Some operations such as memory clock adjustment, core deactivation or memory latency setup do require a system reboot. The application does also display the current Peak Speed and please keep in mind that when you start system oveclocking, Precision Boost and XFR (which are a part of SenseMI) will be disabled:
After benching the system at its stock speeds, our aim was also to see how high we can go in overclocking, by setting a higher multiplier on all cores.
First, we did run Prime95 stress test at stock settings and we have seen with this occasion Precision Boost at work, which does allow us to go from the base clock of 3.6GHz to 3.7GHz on all cores. Under lighter CPU loads, two of the cores may jump up to 4.1GHz, if the thermal conditions are also met. By performing this test, we have seen that the MasterLiquid 280 AIO from Cooler Master did its job wonderfully, by succeeding to keep the processor at 65 degrees Celsius maximum at unrealistically high loads such as Prime:
Our next attempt was to set a speed of 4GHz on all cores, set a starting voltage of 1.35V and restarted Prime95; after just a few moments, we got a black screen and the system became unstable. We have then established a voltage of 1.362, 1.375, 1.387, even got to 1.4V and the system still was not fully stable in Prime95. Our last attempt for 4GHz was at the 1.412 voltage, in which case the CPU package got to about 81 degrees from the first few minutes, so we have abandoned this task, in order not to degrade the CPU.
The 3.9Ghz speed on all cores was our next aim so we have started with a voltage of 1.4V and the system has proven itself as fully stable in Prime95. Afterwards, we have tried with lower voltages such as 1.387, 1.375….all the way down to 1.325V: not only the system was fully stable at this voltage, but the maximum-recorded temperature was decent too! (72 degrees Celsius)
We have also validated the results with the CPU-Z utility:
Not long ago, we have reviewed the mainstream Intel platform that is based upon the Z270 chipset and with this occasion, we have tested and recorded results in many benchmarks. The main highlight in this environment was the Intel Core i5 7600K (Kaby Lake), which can be seen as a product from Intel's lineup with one of the best price/performance ratios. In order to compare the platform with newer architectures/systems, we have used a 16GB memory kit @ 2133MHz with default timings, a KFA2 GTX 1060 OC 6GB video card, one Cooler Master 850 PSU, but also an OCZ Vector 150 240GB SSD. The platform was running on a fresh Windows 10 Anniversary installation while all hardware was mounted inside a Cooler Master ATCS 840 Tower case.
The first comparison stage in this review will be between the AMD Ryzen 7 1800X at stock/overclocked on water versus the Intel Core i5 7600K stock/overclocked on water in order to see how they compare. The water cooling is handled by the MasterLiquid Pro 280 from Cooler Master. The memory timings were adjusted to match the Intel platform, as well as the frequency which was 2133MHz.
The extra cores of the Ryzen 7 1800X do simply crush the i5 7600K in CineBench suites, even when overclocked at 4.8GHz. Blender benchmark does say the same story, by maintaining the Ryzen CPU far ahead. PCMark Vantage does not seem to benefit much from the extra cores, but we can see a healthy performance increase with the overclock at 3.9GHz. In PcMark 7, the system overclock does not seem to matter too much, which tells us that SenseMi does it job very well in productivity applications. Another application that does not seem to make use of all cores is PCMark 8.
SuperPi tells an interesting story: since it is a single-theaded benchmark, it can tell us how powerful a core is in every processor. By running at a higher frequency of 4.2GHz in Turbo Boost, Kaby Lake i5 7600K is taking the lead and the Ryzen cores cannot match it even with SenseMi activated. When each core is overclocked at a stable speed of 3.9GHz, we see a performance drop because SenseMi is not performing the automatic overclocking over 4GHz on two cores anymore.
X265 benchmark does again show a healthy performance boost for the Ryzen 7 1800X, the i5 7600K @ 4.8GHz not being able to match it.
3DMark vantage is an absolute win for the Ryzen CPU, since it can make use of all available cores.
3DMark 11 gives another win to the Ryzen 7 1800X, while 3DMark 2013 shows a healthy performance boost when the CPU is overclocked at 3.9GHz.
Unigine Valley Benchmark is showing a slower performance in case of the Ryzen 7 CPU and here we can see that the maximum FPS has the most impact.
Ashes of Singularity in DX12, CPU Focused Benchmark brings another win for the Ryzen.
As a second stage, thanks to our colleague reviewer colleague Albrecht which has lend a hand with his database, we are going to also pit the AMD Ryzen 7 1800X against the octa-core 6900K Broadwell-E and check out the performance differences. In order to produce comparable results, we have used the same memory timings as he did (DDR4 at 2133C15-15-15-35 2T), while exchanging the KFA2 GTX 1060 OC 6GB video card with a HIS R9 290X 4GB video card.
CineBench R10 CPU rendering benchmark shows a clear win for the Ryzen 7 1800X; same thing goes for CineBench R11.5 CPU rendering benchmark, which clearly tells us that the Zen architecture is the better one for rendering purposes.
On the other side, Fritz Chess provides the Intel Core i7 6900K with a definite win! With the RAM running at the recommended 2667MHz, the Ryzen CPU still cannot match the Intel counterpart.
By running SuperPi, we can see that the Ryzen 7 1800X and the Intel i7 6900K are trading punches.
Handbrake is another very pleasant surprise from AMD! While the average FPS with the Intel i7 6900K is 39.9, we can see an average of 43.5 FPS on the stock 1800X and if we run the RAM at 2666MHz, the average framerate increased at 43.94 FPS.
In 3D-based benchmarks, such as FutureMark Fire Strike, we can see that the Intel 6900K CPU is trading again punches with the Ryzen, and if we did set the RAM at 2666MHz, we received a noticeable performance increase.
Another 3D benchmark we have ran is the HWBOT Unigine Heaven benchmark, which shows the Ryzen falling behind with the RAM at 2133MHz, but is very close to its Intel counterpart at 2666MHz.
Finally, we have Ashes of Singularity, which does give another win to the Intel systems (in DX11 mode, GPU focused).
By using the stock memory timings of the AMD-supplied kit (15-17-17-35), we will check out how the performance of the AMD Ryzen 7 1800X does scale when the memory frequency increases. For this aspect, we have included games, CineBench R15 suite for rendering, HWBOT X265 Encoding suite but also SuperPi 32M.
The following benchmarks are clearly showing that the Ryzen 7 1800X is scaling very well at 1080P with increased memory frequencies, while at 4K resolutions it is self-explanatory that the GPU is used more and limitations start to appear.
Rendering applications/benchmarks such as CineBench R15 but also video encoding does show a performance increase with the higher memory frequencies, while single core performance in SuperPI benefits from a healthy increase.
With the help of the AIDA64 utility, we have been able to check out the memory bandwidth differences at the speeds we have tried during the DDR scaling tests.
After finalizing the tests with the Ryzen 7 1800X and comparing with the Intel counterparts we can finally draw a conclusion regarding its performance: the new AMD competitor offers enough performance in order to beat the Core i7 6900K in rendering and video encoding tasks which is a massive achievement in the case of productivity workstations. If we do also take into account the fact that the flagship is only $499, then we can declare it as a clear winner! In the case of games, Intel seems to still have the lead and this is mainly because a ton of titles are optimized for its platform; the latest discussions on the tech forums also seem to bring more proof regarding this aspect because some of the game titles are running at higher FPS when HT is disabled on the Ryzen. Via patches, AMD will be able to fix this small performance issue and have already announced partnerships with Oxide Games, Creative Assembly, Bethesda Softworks and so on.
Ryzen 7 1800X seems to scale well with the increase of the memory frequency and not only in productivity applications, but in games too! With our ASUS test board, we have had no issues running the RAM up to 2933MHz, but the whole system started acting weird after trying to go further. The current BIOS revisions still need some issues ironed out in order to be able to achieve even higher memory frequencies.
For overclocking and in order to avoid CPU degradation, you need a pretty decent AIO water cooling system and depending on the luck of the draw, you may be able to achieve up to 4.1GHz at decent voltages. Our sample seems to have hit a wall after 3.9GHz, since at 4GHz we needed much more volts and the processor temperatures had increased considerably. We have seen some other publications reaching 4.0GHz and 4.1GHz with the Ryzen 7 1800X, but used over 1.4V in order to attain fully stable systems (AMD recommends a maximum of 1.45V only for occasional benchmarking). Having a smaller overclocking headroom versus the Kaby Lake processors is understandable, simply because we are dealing with the double number of cores that need to operate at the same frequency in a small package.
For the current pricing of the Ryzen 7 1800X, we can clearly say that AMD has done an amazing job beating Intel at their own game in the productivity segment and we are really looking forward seeing the improvements on the gaming side via patches, future optimizations and motherboard BIOS updates!
Thought that is all? Well….we’ll be done in just a bit! Remember the fact that we are dealing with an octa-core processor that runs at speeds up to 4GHz (4.1GHz max on two cores on light loads); we were very surprised to see the overall system low power consumption, even when overclocked:
Stock Speed (SenseMi activated):
IDLE: 55W
Movie: 64.8W
Handbrake Encoding: 146W
Game: 133W
3.9 GHz all-core Overclock
IDLE: 56.3W
Movie: 67.8W
Handbrake Encoding: 163W
Game: 147W
Later updates: AMD has confirmed that the Ryzen platform does support ECC memory and only the motherboard vendors need to update their BIOSes in order to enable this feature.
The latest chipset drivers do also support Windows 7 OS:
We would like to thank again to AMD for making this review possible!