Intel Xeon E5-2690 and E5-2660 8-core Sandy Bridge-EP Review

[siopao1984]

---

Introduction / Quick RecapPrior reading before our quick recap: [to be updated] by Nebojsa Novakovic

The Xeon E5 (8 core/16 thread) silicon live in living colour - with a 20MB L2, die size of 416mm and a transistor count of ~2.263 billionOverview of the Xeon E5 - basically all the improvements of the Sandy Bridge microarchitecture like AVX and Turbo Boost 2.0 brought to the server/workstation platformBlock Diagram - with 2 QPI links, Quad Channel LRDIMM support and 40 PCIe 3.0 lanes per socket, the Xeon E5 should be a no brainer for high performance workloadsPartitioning of the 32nm planar double-gate transistors Sandy Bridge-EP dieXeon E5-2600 SKUs ranging from low power quad core variants to the full fledged 8-core 150W monstersIntel Integrated I/O in a nutshell - integrating traditional north bridge into the CPU and providing DMA access to storage and network controllers for reduced latenciesUnlike earlier Xeons and SNB/SNB-E, we're sad to report that multiplier adjustment is disabled (BIOS values set to read only) and base clock straps are non-existent (frequency only goes up to ~108MHz) across the current stepping. We sure hope Chipzilla changes their mind (since traditionally it has never been obstructed, albeit not officially supported) or maniac SR-X overclockers/high frequency traders are going to be turned away.Quad channel IMC in the Xeon E5 also supports DDR3 LRDIMM (evolution of FB-DIMM but pin compatible with regular UDIMM) that allows densities up to 768GB (24 x 32GB modules)

Read more: Intel Xeon E5-2690 and E5-2660 8-core Sandy Bridge-EP Review by VR-Zone.com

[siopao1984]

Review Kit

For our review, we were provided with a pair of Xeon E5-2690 (2.9GHz, 135W, C1 Stepping) and E5-2660 (2.2GHz, 95W, C2 Stepping), 64GB of Samsung DDR3 LRDIMM and a Intel 2U barebone system





Intel S2600GZ "Grizzly Pass" Server motherboard - an orgy of 24 memory slots, SAS RAID and 10gE networking





The R200 "Big Horn Peak" rackmount is also impressive, with (up to 24 x 2.5inch) SAS hotswap and redundant 80+ platinium power all packed in a 2U form factor





Intel SSD 710 SSD that came with it - this 25nm MLC goes for long term enterprise reliability rather than sheer performance, hence the less than impressive numbers relative to other modern offerings





Test Setup

For a relevant unprejudiced comparison, we pitted today's review subjects against a Xeon X5690 (the flagship Westmere-EP) and an AMD Opteron 'Interlagos' 6274, all running Windows 2008 R2 with latest patches. Just for kicks and shock value, we've included our favourite watercooled Core i7-3960X running at 4.78GHz in our lineup. We've also disabled all power management features to get the max performance out of every setup in this lineup.



System 1 (Dual Xeon E5-2690)

2 x Intel Xeon E5-2690 2.9GHz 8-core C1 Stepping with Turbo/HT on

32GB Samsung LR-DIMM @ 11-11-11-28 1T (Quad Channel)



System 2 (Dual Xeon E5-2690)

2 x Intel Xeon E5-2660 2.2GHz 8-core C2 Stepping with Turbo/HT on

32GB Samsung LR-DIMM @ 11-11-11-28 1T (Quad Channel)



System 3 (Dual Xeon X5690)

2 x Intel Xeon X5690 3.46GHz 6-core with Turbo/HT on

EVGA SR-2 W555

24GB G.Skill Value Series DDR3 unregistered @ 9-9-9-24 1333MHz 1T (Triple Channel)



System 4 (Dual AMD Interlagos 6274)

2 x AMD Opteron 'Interlagos' 6274 with Core Scheduling Patch

32GB Samsung ECC Registered DDR3 @ 9-9-9-24 1333MHz (Quad Channel)



System 5 (Single Intel Core i7-3960X @ 4.68GHz)

Intel Core i7-3960X @ 4.68GHz (104MHz x 46)

16GB G.Skill RipJawsZ 2133 Unregistered DDR3 @ 9-12-10-28 1T 2220MHz (Quad Channel)

[siopao1984]

Benchmarks

The results are fairly conclusive/obvious - so we will just put a short description of the benchmark tests and stressors.

Cinebench R11.5 (MAXON: CINEBENCH)

Based on CINEMA 4D
Shows near linear scaling with CPU clock speed.
Performs a fair amount of floating point operations.






Euler3D (CASE Lab Research: Euler3d Benchmark)

AGARD 445.6 aeroelastic test wing (computational fluid dynamics)
Does not scale very well with adding processors.
Does not scale linearly with processor clock speed.
Floating point intensive workload






UCBench 2011 (UnRAR-crack benchmark)

RAR password cracker
AES/SHA-1 units for decryption and support for SSE4/XOP instruction sets
Scales with clock speed and cores




wPrime 2.07 1024M (Measure Computer Performance | wPrime Multithreaded Benchmark)

Recursive call of Newton's method for estimating functions
Scales with core count, clock speed and memory latency
FP heavy

[siopao1984]

Benchmarks (AIDA 64 v2.20)

Tests descriptions taken from Aida64 » Benchmark guide

CPU Queens: This simple integer benchmark focuses on the branch prediction capabilities and the misprediction penalties of the CPU. It finds the solutions for the classic “Queens problem” on a 10 by 10 sized chessboard





CPU PhotoWorxx: This integer benchmark performs different common tasks used during digital photo processing





CPU ZLib: This integer benchmark measures combined CPU and memory subsystem performance through the public ZLib compression library





CPU AES: This integer benchmark measures CPU performance using AES (a.k.a. Rijndael) data encryption. It utilizes Vincent Rijmen, Antoon Bosselaers and Paulo Barreto’s public domain C code in ECB mode





CPU Hash: This integer benchmark measures CPU performance using the SHA1 hashing algorithm defined in the Federal Information Processing Standards Publication 180-3





FPU VP8: This benchmark measures video compression performance using the Google VP8 (WebM) video codec





FPU Julia: This benchmark measures the single precision (also known as 32-bit) floating-point performance through the computation of several frames of the popular “Julia” fractal





FPU Mandel: This benchmark measures the double precision (also known as 64-bit) floating-point performance through the computation of several frames of the popular “Mandelbrot” fractal





FPU SinJulia: This benchmark measures the extended precision (also known as 80-bit) floating-point performance through the computation of a single frame of a modified “Julia” fractal

[siopao1984]

Benchmarks (Sandra 2012 SP2)

Tests descriptions taken from SiSoftware Zone

Processor Arithmetic (Dhrystone): The benchmark is designed to contain a representative sample of types of (integer) operations, mostly numerical, used by applications.





Processor Arithmetic (Whetstone): The Whetstone benchmark is widely used in the computer industry as a measure of FPU or Co-Processor performance. Floating-point arithmetic is most significant in programs that require a Co-Processor. These are mostly scientific, engineering, statistical and computer-aided design programs.





Processor Multi-Media: This benchmark generates a picture (640x480) of the well-known Mandelbrot fractal, using 255 iterations for each data pixel, in 32 colours. It is a real-life benchmark rather than a synthetic benchmark, designed to show the improvements MMX/Enhanced, 3DNow!/Enhanced, SSE(2) bring to such an algorithm.







Processor Cyptography: A set of benchmarks designed to measure cryptographic performance in the most popular algorithms today: AES for encryption/decryption (AES256/AES12 and SHA for signing (SHA256/SHA1). They allow us to show the importance of hardware accelerated support in the latest processors (CPUs) on the market today.

[siopao1984]

Conclusion

What can we say? Performance wise, the new Sandy Bridge-EP Xeon E5 that we've reviewed today comprehensively beats the previous Westmere-EP Xeon 5600 generation, and annihilates/bulldozers the competition in just about every synthetic test. This doesn't come as a surprise anyway, as we've all seen the power efficiency and IPC improvements of the desktop Sandy Bridge variants (Core i7 3960X is just a reduced 6-cores-out-of-8 version of this same die) throughout 2011 over the older Nehalem/Westmere parts. We do hope AMD responds soon with a strong 'Piledriver' based Opteron 'Abu Dhabi' CPU offering, or else we might have to wait a long time again for Ivy Bridge/Haswell equivalents to show up, literally impeding the progress of the human race.

In a followup piece later this month, we will explore:

Linpack HPC results
Memory Bandwidth/Latency Scaling
Hardware Accelerated Virtualization (with the C2 stepping)
Realistic Web and Database workloads
Science!
Performance/Watt and Performance/$
Other dual socket motherboards from AIB vendors like ASUS, EVGA, Gigabyte and Supermicro
Overclocking (if we do find a way)






Heres a teaser of what we're publishing later today (8-core microATX porn):

[eXpLoDe]

ayayay... overwhelming performance. supppeeerrrr imba kaayo pang server.