As suspected, Intel has finally unleashed their next generation Xeon E5-2600 V3 processors codenamed Haswell-EP at the IDF 2014 event being held in San Francisco. The Xeon E5-2600 V3 lineup is going to be aimed at the HPC server and workstation market featuring 23 E5-2600 V3 SKUs including the flagship Xeon E5-2699 V3 which boasts 18 Cores, 36 threads, 45 MB of L3 cache and a high price of $4115 US.
Intel Xeon E5-2600 V3 "Haswell-EP" Workstation and Server Processors Unleashed
Today, Intel finally takes the lid off of their Haswell-EP lineup (Xeon E5-2600 V3) which replaces Ivy Bridge-EP (Xeon E5-2600 V2) from last year. Built on the latest Haswell microarchitecture, the Haswell-EP lineup ushers into a new era of high-performance computing on the server and workstation side with specifications and core performance improvements which are just amazing.
Starting off with the architectural details, the Intel Haswell-EP Xeon E5-2600 V3 is configured into several die variants which include HCC (High Core Count), MCC (Medium Core Count) and LCC (Low Core Count models). The LCC variants range from 4-8 cores, configured on a 354mm2 die consisting of 2.6 Billion transistors. The MCC die has 10-12 cores, configured on a 492mm2 die consisting of 3.84 Billion transistors while the HCC die which is reserved for the flagship parts including the Xeon E5-2699 V3 ranges from 14-18 core variants and configured on a 662mm2 consisting of a mind boggling 5.69 Billion transistors. The previous Ivy Bridge-EX flagship had 4.31 Billion transistors configured on a 541mm2 so its easy to tell that the Haswell-EP takes die size and transistor count to a new level. The Xeon E7-2600 V3 "Haswell-EN" processors which are reserved for launch later might have even higher core configurations but their arrival is scheduled around Q4 2014 or Q1 2015.
The HCC variants come in 4 columns with each column sharing a L3 cache module of 2.5 MB. The die shown is of the flagship Xeon E5-2699 V3 variant which has the first four columns equipped with 12 Cores (4 each) and 30 MB of L3 cache (10 each) while the last column has 6 cores and 15 MB of L3 cache which results in a total of 45 MB of L3 cache, 18 Cores, and two threads per core equating to 36 threads. The columns are connected to a PCI-e, QPI and DDR4 memory controller. The architecture is specifically optimized for non unified memory architecture workloads.
Intel's Haswell architecture also brings the latest AVX 2.0 (Advanced Vector Instructions 2.0) which doubles the theoretical FLOPS per clock on each core. Intel's AVX 2.0 allows the processors core to operate faster and adds more optimization from the Turbo Boost technology 2.0 revision featured on the Haswell-EP lineup. Everything from number of active cores, type of workload, power consumption, estimated current and CPU temperatures are taken into consideration when boosting the clock speeds of the Xeon E5-2600 V3 processors. The latest P state design allows each core to operate at its own independent voltages and clock speeds which improves CPU performance for different workloads since each core can be optimized for per core performance based on the task being put ahead. Another update is the QPI (Quick Path Interconnect) speeds which have been updated from 8.0 GT/s to 9.6 GT/s on Haswell-EP. Although some of the Haswell-EP processors still stick with the 8.0 GT/s speeds but the bulk of the lineup features the new standard which enhances bandwidth when using multi socket systems and boards.
The AVX 2.0 also includes the latest FMA 2.0 (Fused Multiply Add) floating point which improves performance based computing. The 256-bit integer vector instructions also increase core performance driving an increase of 90% generation to generation performance delivery starting with original E5-2600 series in 2012.
| Intel Ivy Bridge-EP Xeon E5-2600 V2 | Intel Haswell-E5 Xeon E5-2600 V3 | Intel Broadwell-EP Xeon E5-2600 V4 | |
| Series | Xeon E5-2600 v2 | Xeon E5-2600 v3 | Xeon E5-2600 v4 |
| Max Cores | 12 | 14-18 | 18 |
| Threads | 24 | 28-36 | 36 |
| LLC Cache | 30 MB L3 | 35-45 MB L3 | 45 MB L3 |
| Die SKUs | LCC - 4/6, 256.5mm2, 1.86 Billion MCC - 8/10, 341mm2, 2.89 Billion HCC - 10/12, 541mm2, 4.31 Billion |
LCC - 4/8, 354mm2, 2.6 Billion MCC - 10-12, 492mm2, 3.84 Billion HCC 14-18, 662mm2, 5.96 Billion |
TBA |
| Memory | DDR3-1866 | DDR4-2133 | DDR4-2400 |
| Memory Population | Quad Channel DDR3 | Quad Channel DDR4 (3 LRDIMM) | Quad Channel DDR4 (3DS LRDIMM) |
| Chipset | C610 ‘Wellsburg’ | C610 ‘Wellsburg’ | C610 ‘Wellsburg’ |
| TDP | 130W | 145W (160W Workstation Only) | 145W (160W Workstation Only) |
| Launch | IDF 2013 | IDF 2014 | IDF 2015 |
Intel Broadwell-EP Xeon E5-2600 v4 – The First 14nm Xeon Processors
Following Haswell-EP is Broadwell-EP, Intel next-generation and first 14nm based Xeon processor series featuring their upcoming Broadwell architecture. Broadwell which according to Intel’s Tick-Tock roadmap is a Tick which means its essentially a die shrink (14nm) of Haswell featuring massive changes to the core design.
First of all, the Xeon E5-2600 V4 flagship processor would feature upto 18 cores and 36 threads on a single socket. That’s 72 threads running simultaneously on a dual socket enterprise setup and from the performance uplift chart , you can note that its going to be a complete and powerful package. Broadwell-EP will also be coupled with 45 MB of Cache (LLC) and will natively support DDR4-2400 MHz memory which is a decent upgrade over Haswell-EP but that puts the total improvement over DDR3 to about 25 percent which is great.
Intel C610 "Wellsburg" Chipset and DDR4 Memory Support:
Another technology update over at the workstation front is the introduction of C612 series chipset and DDR4 memory. The Intel C612 series "Wellsburg" chipset is the updated refreshed of the C610 chipset which adds support for up to 10 SATA 3 (6 GB/s) ports, Discrete/Integrated/Hybrid clocking system, 8 USB 2.0/ 6 USB 3.0 ports, up to 8 PCI-e 2.0 (5 GT/s) lanes (x1/x2/x4), Intel vPRO, AMT, CUPs, MCTP firmware updates, Intel RSTe 4.0 RAID support. The chipset is designed for the 2S market while Haswell-EP processors can drive up to 40 PCI-e lanes (Gen 3.0) and include the latest Fortville network controller (Ethernet Controller XL710 10 GbE/40 GbE).
The DDR4 enhancements and support is another major feature which reduces the voltages while increasing the overall bandwidth of the systems. Intel's Xeon E5-2600 V3 platform can drive up to 384 GB of registered DDR4 and 768 GB of LRDIMMs (using 32 GB DRAM sticks) on 2S platforms but expect more capacity when the frame buffer on DDR4 ram increases to 64 - 128 GB per stick in the future.
Intel Xeon E5-2600 V3 "Haswell-EP" Processors:
| Processor | Cores/ Threads |
Core Clock / Max Turbo | LLC Cache | Quick Path Interconnect (QPI) | TDP (W) | Pricing |
|---|---|---|---|---|---|---|
| E5-2699 v3 | 18 / 36 | 2.3 GHz / 3.6 GHz | 45 MB | 9.6 GT/s | 145 W | $4115 |
| E5-2698 v3 | 16 / 32 | 2.3 GHz / 3.6 GHz | 40 MB | 9.6 GT/s | 135 W | $3226 |
| E5-2697 v3 | 14 / 28 | 2.6 GHz / 3.6 GHz | 35 MB | 9.6 GT/s | 145 W | $2702 |
| E5-2695 v3 | 14 / 28 | 2.3 GHz / 3.3 GHz | 35 MB | 9.6 GT/s | 120 W | $2424 |
| E5-2690 v3 | 12 / 24 | 2.6 GHz / 3.5 GHz | 30 MB | 9.6 GT/s | 135 W | $2090 |
| E5-2687W v3 | 10 / 20 | 3.1 GHz / 3.5 GHz | 25 MB | 9.6 GT/s | 160 W | $2141 |
| E5-2685 v3 | 12 / 12 | 2.6 GHz / 3.3 GHz | 30 MB | 9.6 GT/s | 120 W | $2090 |
| E5-2683 v3 | 14 / 28 | 2.0 GHz / 3.0 GHz | 35 MB | 9.6 GT/s | 120 W | $2090 |
| E5-2680 v3 | 12 / 24 | 2.5 GHz / 3.3 GHz | 30 MB | 9.6 GT/s | 120 W | $1745 |
| E5-2670 v3 | 12 / 24 | 2.3 GHz / 3.1 GHz | 30 MB | 9.6 GT/s | 120 W | $1589 |
| E5-2667 v3 | 8 / 16 | 3.2 GHz / 3.6 GHz | 20 MB | 9.6 GT/s | 135 W | $2057 |
| E5-2660 v3 | 10 / 20 | 2.6 GHz / 3.3 GHz | 25 MB | 9.6 GT/s | 105 W | $1445 |
| E5-2650 v3 | 10 / 20 | 2.3 GHz / 3.0 GHz | 25 MB | 9.6 GT/s | 105 W | $1166 |
| E5-2650L v3 | 12 / 24 | 1.8 GHz / 2.5 GHz | 30 MB | 9.6 GT/s | 65 W | $1329 |
| E5-2643 v3 | 6 / 12 | 3.4 GHz / 3.7 GHz | 20 MB | 9.6 GT/s | 135 W | $1552 |
| E5-2640 v3 | 8 / 16 | 2.6 GHz / 3.4 GHz | 20 MB | 8.0 GT/s | 90 W | $939 |
| E5-2637 v3 | 4 / 8 | 3.5 GHz / 3.7 GHz | 15 MB | 9.6 GT/s | 135 W | $996 |
| E5-2630 v3 | 8 / 16 | 2.4 GHz / 3.2 GHz | 20 MB | 8.0 GT/s | 85 W | $667 |
| E5-2630L v3 | 8 / 16 | 1.8 GHz / 2.9 GHz | 20 MB | 8.0 GT/s | 55 W | $612 |
| E5-2623 v3 | 4 / 8 | 3.0 GHz / 3.5 GHz | 10 MB | 8.0 GT/s | 105 W | $444 |
| E5-2620 v3 | 6 / 12 | 2.4 GHz / 3.2 GHz | 15 MB | 8.0GT/s | 85 W | $412 |
| E5-2609 v3 | 6 / 6 | 1.9 GHz | 15 MB | 6.4 GT/s | 85 W | $306 |
| E5-2603 v3 | 6 / 6 | 1.6 GHz | 15 MB | 6.4 GT/s | 85 W | $213 |
Intel Haswell-EP Xeon E5-2699 V3 Die:
Intel Haswell-EP Xeon E5-2600 V3 Gallery:
Intel Haswell-EP Xeon E5-2600 V3 Slides:
