Dedicated Servers Based on Intel Avoton Processors

This year, Intel introduced the new Atom family of processors in the C2000 family, also known as Avoton. We will discuss the new configurations and their technical features in detail in this article.

Specifications

The 64-bit processors of the Intel Atom C2000 family, also known under the code name Avoton, replaced the previous generation (code name Intel Centerton), which for a number of reasons proved to be of little demand.

New processors are available in a 22-nanometer process. They include from 2 to 8 cores. We offer servers based on the oldest in the line of eight-core Intel Atom C2758 processor (their configurations are described in detail below).

Processors of the Intel Atom C2000 family support the out-of-order execution of 64-bit instructions, which allows to increase productivity. Turbo Boost technology is also supported, with the help of which productivity automatically increases during peak loads. So, the clock frequency of 2.4 GHz with Turbo Boost can be increased to 2.7 GHz.

The power consumption level of the new processors is in the range of 5 - 20 watts.

Like previous models, processors from the Intel Atom C2000 family have 16 PCI Express lanes and support LPC, SPI, SMBus, UART interfaces. The dual-channel DDR controller works with DDRIII-1600 and DDRIIIL-1600 memory.

The new processors are based on the Silvermont microarchitecture, thanks to which it was possible to achieve a significant increase in performance. Consider its features in more detail.

New architecture

The architecture of Intel Atom processors has not undergone major changes for a long time (apart from the transition from 45nm to 32nm process). The advent of the Silvermont microarchitecture was a serious step forward: the processors of the new family are fundamentally different from all previous models of “atoms”.

The new microarchitecture is based on an extraordinary execution of commands, so productivity has increased significantly. At the same time, Intel is still based on the principle of combining certain instructions into a single micro-operation instruction, which increases the processing efficiency of the corresponding x86 commands.

For processors with Saltwell microarchitecture (Intel Centerton processors are based on it), the length of the computing pipeline is 16 steps. Since these processors are based on the paradigm of sequential execution of instructions, micro-operations must go through all stages of the pipeline, even if they do not need access levels to the cache. As a result, a branch with a transition prediction error spends 13 cycles in vain.

In Intel Silvermont architecture, micro-operations can bypass cache access levels and execute immediately when the cache is not used. Therefore, erroneous
prediction consumes only 10 cycles.

Many of the instructions that were slowly executed in the previous architecture were modified to reduce latency and increase throughput. Floating-point calculations have been reduced by a few clock cycles. Pair SIMD calculations are now performed in 4 cycles, and not in 9, as before. The number of instructions per cycle increased by about 1.5 times.

The L2 cache on Intel Silvermont-based processors is tightly integrated with the cores. The cores are combined into modules - pairs with a common L2 cache 1 MB in size. Individual cores, the L2 cache, and the interface between the cores and the cache can be powered independently of each other. The cores in the same module can operate at different frequencies, but by default they work symmetrically.

The modules communicate with each other using the IDI (in-die interface implemented on the chip itself), which has independent read / write channels.

One of the hallmarks of the Silvermont architecture is the abandonment of HyperThreading technology. Improving the performance of many threaded applications is achieved by increasing the number of physical cores.

In order to improve performance in the processors based on Silvermont, the instruction set has been substantially updated. AES-NI and Secure Key encryption acceleration also appeared.

Virtualization hardware support is based on second-generation VT-x technology. A set of Extended Page Tables functions is supported, including addressing the Virtual Processor ID in the TLB buffer and the Unrestricted Guest mode, with which KVM users can access the executable code.

At the level of instructions and functionality, the Intel Atom C2000 family of processors are fully compatible with the Xeon family of processors.

Performance

To give a more visual representation of the performance of new processors, we present a table with the test results.

Testing was performed using the sysbench utility (CPU, Threads, Mutex, Memory columns). The lower the value, the better the performance.

The table also includes the results of calculating the relative performance per core (column “Core 1 Performance”) and per core with frequency normalization (column “Architecture Efficiency”). The higher the value, the better the result.

Model	CPU	Threads	Mutex	Memory	1 core performance	Architecture Efficiency	Kernels	Frequency GHz
2x Xeon E5-2630	37.273	39.992	4.065	67.559	2.24	0.97	12	2.3
2x Xeon E5-2620	42.101	42.878	4.464	73.784	1.98	0.99	12	2
2x Xeon E5620	59.907	29.99	2.832	61.652	2.09	0.87	8	2.4
2x Xeon L5630	71.034	34.059	3.348	68.583	1.76	0.83	8	2.13
Xeon E3-1270	72.565	19.616	1.523	10.8	3.45	1.01	4	3.4
Xeon E3-1230	87.891	26.902	1.622	23.348	2.84	0.89	4	3.2
2x Xeon E5504	94.221	34.645	3.313	77.046	1.33	0.66	8	2
Avoton C-2758	135.624	29.7066	5.25	78.895	0.92	0.38	8	2.4
2x Xeon E5-2603	142.662	33.183	4.575	89.117	0.88	0.49	8	1.8
Core2 Q8300	151.55	31.28	4.345	71.583	1.65	0.66	4	2.5
Core i3-2120	175.782	29.013	1.418	18.201	2.84	0.86	2	3.3
Core2 e8400	252.464	43.443	0.968	26.514	1.98	0.66	2	3
2x Xeon E5130	339.772	45.813	6.838	90.813	0.74	0.37	4	2
Atom atom d525	2159.999	149.234	8.177	210.218	0.23	0.13	2	1.8

The table shows that due to the improvement of the processor architecture, the computational efficiency has increased almost three times. The new “Atoms” have risen to about the level of Xeon processors of old generations, they barely reach the level of Core2 microarchitecture. The performance lag of the architecture from modern Xeon models is compensated by the high frequency of the cores and their number, which favorably affects the final performance.

The test results indicate that the new Atoms have a much wider range of tasks than the previous models. Today it is safe to say that Atom processors are no longer a niche product and are striving for the low-middle segment of the server market. Low cost, low power consumption and fairly ample opportunities undoubtedly make them attractive to a wide range of potential users.

New configurations and their features

We offer the following server configurations based on Intel Avoton processors:

• Intel Atom C2758 2.4GHz, 8GB DDR3, 2 × 1TB SATA - 3000 rubles per month;
• Intel Atom C2758 2.4GHz, 16GB DDR3, 2 × 240GB SSD - 4000 rubles per month.

The new servers will undoubtedly attract many users simply because they are characterized by an excellent price-quality ratio.

The first configuration (Intel Atom C2758 2.4GHz, 8GB DDR3, 2 × 1TB SATA) is well suited for tasks such as serving static web pages, dedicated entry-level hosting, data delivery, etc.

Model c SSD can be used as a front-end server, as well as a caching server. New servers are available for order both in St. Petersburg and in Moscow.

Readers who cannot post comments here are welcome to join us on the blog .