
Intel Xeon W - when you want more help
In the Cascade Lake family replenishment: after the second generation Intel Xeon Scalable, new models of the Xeon W series are introduced . Let me remind you that processors with the W index are designed for high-performance workstations - in particular, these new products will be included in the freshly announced Apple Mac Pro , although they are suitable for any other computer, even in the wall-mounted version.

A total of 9 models are presented in the family, all of them are almost identical in functionality and differ only in the number of cores and frequencies. Some models have an additional index M - it means the presence of a more advanced memory controller that supports up to 2 TB of RAM. All others are limited to “just” one terabyte. With younger models, DDR4-2666 memory is used, while older ones can work with DDR4-2933. As for the socket, it is the same as Scalable - LGA 3647.
Among the functional innovations, I would like to mention Intel Deep Learning Boost - an additional set of instructions AVX-512, which accelerates machine learning and Deep Learning due to vector operations of scalar product 8- bit and 16-bit values with 32-bit summation.
Numerical operational characteristics are given in the table. Note that compared with the previous generation, the choice of models with more than 20 cores has increased - previously there were only one, now 4. Thanks to improvements in the manufacturing process, we managed to lower TDP: earlier it was 255 W for 28 cores, now it is 205 W. So, within the 14 nm framework, there is still something to do.
It seems that with such characteristics, the new Xeon W will suit the most fastidious user. On sale, they will appear in the very near future.

A total of 9 models are presented in the family, all of them are almost identical in functionality and differ only in the number of cores and frequencies. Some models have an additional index M - it means the presence of a more advanced memory controller that supports up to 2 TB of RAM. All others are limited to “just” one terabyte. With younger models, DDR4-2666 memory is used, while older ones can work with DDR4-2933. As for the socket, it is the same as Scalable - LGA 3647.
Among the functional innovations, I would like to mention Intel Deep Learning Boost - an additional set of instructions AVX-512, which accelerates machine learning and Deep Learning due to vector operations of scalar product 8- bit and 16-bit values with 32-bit summation.
Numerical operational characteristics are given in the table. Note that compared with the previous generation, the choice of models with more than 20 cores has increased - previously there were only one, now 4. Thanks to improvements in the manufacturing process, we managed to lower TDP: earlier it was 255 W for 28 cores, now it is 205 W. So, within the 14 nm framework, there is still something to do.
CPU | Cores / threads | Frequency GHz | TDP, W | Cache, MB | Price, $ |
---|---|---|---|---|---|
Xeon W-3275M | 28/56 | 2.5-4.6 | 205 | 38.5 | 7453 |
Xeon W-3265M | 24/48 | 2.7-4.6 | 205 | 33 | 6353 |
Xeon W-3245M | 16/32 | 3.2-4.6 | 205 | 22 | 5002 |
Xeon W-3275 | 28/56 | 2.5-4.6 | 205 | 38.5 | 4449 |
Xeon W-3265 | 24/48 | 2.7-4.6 | 205 | 33 | 3349 |
Xeon W-3245 | 16/32 | 3.2-4.6 | 205 | 22 | 1999 |
Xeon W-3235 | 12/24 | 3.2-4.5 | 180 | 16.5 | 1398 |
Xeon W-3225 | 8/16 | 3.7-4.4 | 160 | eleven | 1199 |
Xeon W-3223 | 8/16 | 3,5-4,2 | 160 | eleven | 749 |