AMD launched the second generation of its high-end desktop processors, Ryzen Threadripper this year. The second generation of CPUs is said to be more powerful, with better frequency, and models with core count up to 32 cores. The 32-core and 16-core processors were launched back in August. AMD has now launched the 24-core 2970WX and the 12-core 2920X and we’ll be taking a look at how these chips perform.
The new Threadripper 2 processors use AMD’s latest Zen+ microarchitecture which provides a performance improvement of about 3 percent at the same clock speeds. Along with this, these second-generation chips are produced using the 12nm manufacturing node, which provides improved frequencies for an overall performance improvement of 10 percent.
Dynamic Local Mode
AMD has also released a new mode for the 24-core 2970WX and the 32-core 2990WX called the ‘Dynamic Local Mode’. This new mode is intended to improve the performance even further. This is achieved by ensuring the system’s most demanding threads are scheduled on local DRAM dies. Using a system service, Dynamic Local Mode measures the CPU time of all active threads and then arrange them in a descending order. The most demanding threads are placed onto the Zeppelin dies with local memory access.
Dynamic Local Mode is claimed to offer a performance boost of around 10 percent in some games at 1080p, and a boost of up to 20 percent in specific SPECwpc workloads.
Precision Boost Overdrive (PBO)
Precision Boost Overdrive is not a new feature but is quite important of these processors to perform the best possible. Precision Boost 2 system is used by AMD to determine the turbo frequency, instead of the older ‘cores loaded = set frequency’ model, it attempts to boost as much as it can within the power and current constraints.
PBO increases the limits on power and current, ensuring it’s within safe temperature limits. This increases performance over multiple scenarios, especially heavily multithreaded scenarios if there’s any scope in motherboard power delivery and in the cooling. AMD claims that tasks using PBO can experience around 13 percent improvement in performance. PBO is enabled using the Ryzen Master software.
AMD has claimed that PBO pushes the processor beyond the rated specifications and is basically an overclock, any damage caused by it won’t be covered under the warranty.
StoreMI is AMD’s take on caching technology. It allows users to combine an HDD, up to a 256 GB SSD, and up to 2 GB of DRAM into a single unified storage space. The software moves the data around which reduces the access time, and obviously, improves performance.
AMD claims that in a best-case scenario, StoreMI can improve loading times up to 90 percent over a normal hard drive. One demerit of this feature is that is one drive malfunctions, the whole chain is lost.
AMD Threadripper 2 processors will be compatible with the old X399 motherboards, with a BIOS update. Boards that are updated to support the 2990WX and 2950X will also support the 24-core and 12-core 2970WX and 2920X.
24 core Threadripper 2970WX
The 24 core processor has four silicon dies and each of them carries six cores. Big rises can be seen at 7-8 core load, 13-14 core load and 19-20 core load. The reason is that we are crossing die boundaries when more threads are added.
The same pattern is much more clearly visible in the per-core charts. At 7-8 core load, the power jumps from 12 W per core to 15 W per core. At 13-14 core load, the change is from 10 W per core to 12-13 W per core.
The non-core power was around 53-55 W of the total power. This is about 20-22 percent of the total package power.
12 core Threadripper 2920X
The power curve is smooth and linear, a TDP of 180 W is reached at 9-10 core load.
The per-core values show that a per-core power drop doesn’t occur until 8 core load. At one core load, the power drawn is at 18 W, and at 7 core load, the power drawn is still between 17 to 18 W per core.
Peak Power Comparison
System response time is something every user takes very seriously. To test the loading speeds, actual time to load a particular application can be measured. For this, an open-source online photo editing tool named GIMP was used. A large 50 MB design template was loaded 10 times in intervals of 10 seconds and the results are below.
FCAT: Image Processing
The FCAT software is designed to detect micro stuttering, dropped frames, and run frames in graphics benchmarks. The video used for the test was 90 seconds of Rise of the Tomb Raider at 1440p, FCAT uses this video and processes the color data into frame time data so the system can give out an ‘observed’ frame rate.
As you can see, both processors tested are in the middle of the chart.
3-D Particle Movement v2.1: Brownian Motion
Dolphin 5.0: Console Emulation
DigiCortex 1.20: Sea Slug Brain Simulation
y-Cruncher v0.7.6 : Microarchitecture Optimized Compute
Agisoft Photoscan 1.3.3: 2-D image to 3-D Model Conversion
Corona 1.3: Performance Render
Corona scales well with core count and memory bandwidth. The 24 core 2970WX is only behind the 32 core 2990WX and the 12 core 2920X is behind the 16 core 2950X and sits in about the middle of the chart.
Blender 2.79b: 3D Creation Suite
For this test, bmw27 scene was run in CPU only mode, and time to complete the render was measured.
As it’s visible from the results, the position of the two processors is about similar to what we saw in the last test.
LuxMark v3.1: LuxRender
The ‘Ball’ scene was run on both C++ and OpenCL code paths, in CPU only mode.
POV-Ray 3.7.1: Ray Tracing
PCMark measures and scores 14 different tasks and areas, like web, spreadsheets, photo editing, rendering, video conferencing, application startup, and physics. But what actually reflect a processor’s capability is the overall score.
PCMark has a bit of uniformity of scores with almost all processors except the Intel Core i9-9900K which takes a slight lead over other chips.
This test uses windows instructions to compile Chrome 56 build from March 2017.
Both the processors perform quite well against the competition but the 16 core, 32 thread AMD TR2 2950X and Intel Core i9-9900K share the pole position with the 2920X behind these two on second.
This test was conducted on different scenes like Cloud Gate, Sky Diver, Fire Strike Ultra, and Time Spy.
Handbrake is an open source video conversion tool. It supports conversion from any format to any format.
The test was split into several tests, and a Logitech C920 1080p 60 native webcam was used for recording.
The output settings used are:
- 720p60 at 6000 kbps constant bit rate, fast setting, high profile
- 1080p60 at 3500 kbps constant bit rate, faster setting, main profile
- 1080p60 HEVC at 3500 kbps variable bit rate, fast setting, main profile
This test is used to benchmark the command line, reporting the compression, decompression, and a combined score.
Interestingly, Intel CPUs have better scores on compression while AMD has better scores in decompression.
A compression over a set directory was run consisting of over thirty 60-second video files and 2000 small web-based files at a normal compression rate.
In the AES encryption test, TrueCrypt was used to test several encryption algorithms directly In memory.
WEB AND LEGACY TESTS
This test was run four times, and an average was taken. The benchmarking includes various aspects like user interaction, office computing, graph generation, list sorting, HTML5, image manipulation, and some AI testing.
This is the older version of WebXPRT 3 we just used. This test focuses on a different set of web technologies and frameworks.
The results are given in ‘rpm’.
Google Octane 2.0
Octane is a web test developed by Google. This version of the test performs tasks such as regular expressions, cryptography, ray tracing, emulation, and Navier-Stokes physics calculations.
Mozilla Kraken 1.1
This is an older test than the Octane and this one is developed by Mozilla. This test also performs similar computational mechanics, like audio processing or image filtering.
World of Tanks encore
Final Fantasy XV
Shadow of War
Grand Theft Auto V
Far Cry 5
Shadow of the Tomb Raider
CONCLUSION and PRICING
The 12 core 2920X and the 24 core 2970WX maybe not be as powerful as the previously launched 32 core 2990WX and 16 core 2950X, but they still have serious potential and their own market.
Performance wise, the 2920X is just below the 2950X, and the 2970WX is just below the 2990WX, which is expected, because of the lower core count on these processors with either lower or same frequencies.
The 2970WX is going to retail at $1299 which is $500 less than the 2990WX, this makes it quite competitive in the price to the performance metric. In some of the tests, the 2970WX actually performs better than the 2990WX, like in PhotoScan, 3DMark Physics, and GeekBench. This is probably because of the fewer cores competing for the inter-core bandwidth. However, in overall performance, 2990WX still does better.
The 12-core 2920X is the most basic chip in the whole Threadripper 2 series. Here’s a comparison of it with the Intel Core i9-7920X which is also a 12 core chip.
The i9-7920X chip performs well in the AVX-512 but lags behind in compile and web tests performance. Both the chips are very similar in terms of overall performance, the price is what separates them. The AMD costs around 45 percent less than the Intel chip.
Both of these chips were launched after the 16-core and 32-core models, which is probably because these chips are not as powerful, but are still very aggressive on the price to the performance metric.