Architectural Details

Up-to 40 MB L2+L3 Cache

The caches; all parts are based on the same architecture, anything Ryzen. And considering there is 20 MB for an 8-core processor for the L2 and L3 cache, that can be multiplied by two and thus comes to 40MB of cache on the 16-core part and 80MB for the 32-core part.

	Ryzen Threadripper 2990WX	Ryzen Threadripper 2970WX
TDP	250W	250W
Core Topology	4+4 (die 0/1) / 4+4 (die 2/3)	3+3 (die 0/1) / 3+3 (die 2/3)
L2 Cache	512K per core (16MB total)	512K per core (12MB total)
L3 Cache	16MB per die (64 MB total)	16MB per die (48MB total)
Base Frequency	3.0 GHz	3.0 GHz
All Core Boost	tba GHz	tba GHz
XFR Requency	Up to 4.2GHz XFR (4 cores)	Up to 4.2GHz XFR (4 cores)
SMT	64 threads	48 threads
PCIe Gen3 Lanes	64	64

Some examples: The 32-core part with eight 4-core CCX units gets a per core L1 data cache size of 32 KiB, an L1 instruction cache size of 64 KiB and then an L2 cache size of 512 KiB per core.

• L1 32 x 32 Kb Data
• L1 32 x 64 Kb Instruction
• L2 32 x 512 Kbytes
• L3 8 x 8 Mbytes L3

The 16-core part with four 4-core CCX units gets a per core L1 data cache size of 32 KiB, a L1 instruction cache size of 64 KiB and then a L2 cache size of 512 KiB per core.

• L1 16 x 32 Kb Data
• L1 16 x 64 Kb Instruction
• L2 16 x 512 Kbytes
• L3 4 x 8 Mbytes L3

The L3 cache remains an open bigger pool at 8 Mbytes per CCX cluster. Again, one Core Complex Unit holds four processor cores. The processor has quad-channel DDR4 support, AVX2, AES, FMA3, AMD-V SSE 4.1 and 4.2 instruction sets etc. The bus frequency is 100 MHz multiplied by whatever the processor fires off at it. Ryzen Threadripper Gen2 is built with a 12nm FinFET+ fab node, this greatly helps where AMD is with performance and power consumption. 

You can count them, that's 48 threads ... a staggering thing to see - let alone experience.

The technology inside the processor

AMD implemented technology that will make sure that this processor runs applications efficiently and optimized. SenseMI is a set of sensing and adaptive technologies, including an artificial network inside every “Ryzen” series processor to anticipate future decisions, preload instructions, and choose the best path through the CPU. AMD is also introducing a new interconnect called AMD Infinity Fabric. This is a new and fast way of connecting various parts within a SoC. Infinity Fabric is not just used in the Ryzen generation processors, it is also found in future (Vega) AMD GPUs and (almost) all other AMD chips in the near future. Infinity Fabric allows for faster and better secure connections within a chip. The inter- and intra-chip connector will be standardized and used in many AMD products and, as such, AMD can easily communicate over that very same interconnect. Pure Power is a technology that allows the Ryzen chips and other Ryzen variants to work as efficiently as possible. Being part of the Ryzen family, Threadripper processors are just that; CPUs based upon Ryzen architecture. While a lot of IO changes have been made to facilitate it the reality is simple: in the processor package you'll spot two or four 8-core Ryzen ZEN+ processor dies activated, much like the Ryzen 7 2700X has two CCxes, each holding four cores. This means that Ryzen Threadripper processors have four or eight CCXes. Which allows them to go to 16 or 32 cores in a 4+4 CCX configuration, or 12 and 24 cores on a 3+3 CCX configuration.

You've already seen a number of X399 motherboard announcements and it is no secret that processors will be released in the HEDT (High-End Desktop) and more professional creator segment. Available starting August 13th, 2018, AMD will initially release the 32-core 64-thread model and, later on, the three other models. All offer you quad-channel memory and 64 lanes of PCIe gen 3.0.