Menu Content/Inhalt
Home arrow Server & Workstation arrow Transition from DXT to DXR Platform in 2012

Transition from DXT to DXR Platform in 2012 Print
April 2012

Another good thing is coming to an end in April 2012.  After serving our customers for 34 months, we are transitioning the production of new Dual Socket server systems from the 8th to the 9th generation.  This is a process called creative destruction.  This refers to new systems to be made in Compucon and not to systems in service at client sites.  Systems in service will live far longer than 34 months and many will live to 100 months if they are well looked after. 

Good things give way to better things.  This article explains how the 9th generation Dual Socket technology is better.  Let us tune to 2 Intel names for correlation- Tylersburg stands for the 8th generation and Romley stands for the 9th generation.  To further assist this appraisal process for people familiar with PC technologies, 8th generation server technology is associated with X58 desktop technology and 9th generation is associated with X79 which was introduced a few months ago. 

There are 4 major improvements: (a) The new series of CPU has up to 8 cores versus 6 previously and is faster, (b) The new CPU can talk with more memory via 4 channels simultaneously as compared to 3 channels previously, (c) The new CPU includes PCIe controllers of the 3rd generation versus 2nd generation and North Bridge based previously, and (d) The new platform supports a total of 72 PCIe lanes as compared to 36 lanes in the previous generation.

How much faster?   This obviously depends on the type of applications being considered.  Online literatures have suggested an improvement of 12 to 20 percents based on a similar cost base.  Another major benefit is the availability of up to 16 threads for computing and this will reduce the latency of launching applications such as for online transactional processing (OLTP).  The better performance is also helped by a bigger pipe for memory transfer.  The new series of CPU is denoted by a prefix of E5 followed by 4 digits.  The previous series is E5 followed by 3 digits.

The new standard Compucon motherboard has 16 DIMM slots each capable of 32GB module when the latter is available in the future.  With 8GB modules, we can reach 128GB main memory and this is plenty for most applications. 

PCI Express 3.0 is defined for 1Gbps of data transfer rate per lane and in duplex.  When we aggregate 16 such lanes together, the bandwidth is as large as 16Gbps.  Which peripheral or sub-system needs this level of data transfer rate?  Not many.  A SAS 600 hard disk has an interface bandwidth of 6Gbps but the disk mechanism has not touched this level yet.  10G Ethernet is for 10Gbps and thus it is close.  Video can soak up 16Gbps or more if available and it is the single subsystem that asks for infinite bandwidth.  Would a mix of applications choke a PCIe 3.0 lane?  They will not because PCIe lane is point to point or one lane per device. 

The new standard Compucon motherboard has 6 PCIe slots, 3 with 6 lanes and 3 with 8 lanes.  The total is 72 lanes for 72Gbps of system level data transfer capability.  This is another advancement of capability.  On top of this, the lanes are controlled directly from the CPU and not the North Bridge.  This will cut down latency significantly.

Owing to a higher level of capabilities, Romley based systems called DXR will be more expensive than DXT during the transition period.  We will continue to supply DXT on request, and show DXR as the norm on our system pricelists for May 2012 and thereafter.