Do E-class CPUs down clock when AVX in the execution stack? if it does Why?
The Core i7 processors that are referred to as "Haswell-E" and "Broadwell-E" are minor variants of the Xeon E5 v3 "Haswell-EP" and Xeon E5 v4 "Broadwell-EP" processors. These have lower "maximum Turbo" frequencies for each core count when 256-bit registers are being used.
For the Xeon E5 v3 ("Haswell-EP") processors, the maximum Turbo frequencies with and without the use of 256-bit registers are documented in Tables 2 and 3 of "Intel Xeon Processor E5 v3 Product Family: Processor Specification Update" (Intel document 330785-010, February 2016). The values in Table 2 (without 256-bit instructions) are the same values that one can read from the hardware registers (MSR_TURBO_RATIO_LIMIT and MSR_TURBO_RATIO_LIMIT1) described in Chapter 35 of Volume 3 of the Intel Architectures SW Developer's Manual.
The "specification update" for the Xeon E5 v4 ("Broadwell-EP") does not contain a table of maximum Turbo frequencies with 256-bit registers in use.
I have not looked for this information for the Core i7 parts derived from the Haswell-EP and Broadwell-EP processors. The corresponding "specification update" documents would be a good place to start. Some of the Haswell-E based Core i7 parts may have the same core count, power envelope, and nominal frequency as a Xeon E5-16xx processor, in which case the values in the Xeon E5 v3 "specification update" might apply.
Certain AVX workloads may run at lower peak turbo frequencies, or drop below the Non-AVX Base Frequency of the SKU. This type of behavior is due to power, thermal, and electrical constraints.
A white paper was released for E5v3 (Haswell) describing some of these behaviors:
We are currently developing a plan to update our documentation to support AVX-512 and the E5v5 (Skylake) product line. The concepts described in the Haswell paper remain largely the same.
Tanks for that information. What is AVX offset for? Is there a degradation problem with running a Processor at TJmax.