SiSoftware Sandra 2009 Interview with Adrian Silasi

SiSoft this week will release a new version of their popular application, Sandra. New in Sandra 2009, a  handy system tool and analyzer, is GPU and in specific, a set Generic purpose GPU and GPU render tests. Next to that you'll also spot a test to measure framebuffer bandwidth. Suffice that say, that's our cup of thee and therefore we wanted to fire off a small Q&A session at the lead programmer of SANDRA and Hardware Guru himself, Adrian Silasi. Topics involve SANDRA 2009, GPUs and compute API's like CUDA and CAL.

[Guru3D] Hello Adrian, can you tell us a tad about yourself, what do you do. How long have you been part of the Sandra project and how many people are actually working on the project.
 

[Adrian]  My name is Adrian Silasi. I first became interested in programming and more specifically the inner workings of my computer whilst studying Electrical & Electronic Engineering at University. It soon became apparent that I could not find the information I was looking for at a price I could afford. 14 years later the quest for greater understanding of the inner workings of computers, as well as which technologies would bring the best improvements, continues.

We are a small core team drawing on the expertise of contributors from around the World, much like most on-line magazines/review sites.

[Guru3D] Can you tell us a little bit on what Sandra is and what it does, why should people test/buy this software.

[Adrian]  Sandra is a suite of modules that collectively allow the user (home, enthusiast or technical) to analyse the performance of their computer as well as to gain an insight into what "makes it tick". We also produce commercial versions, such as the Engineer and Enterprise, that allow support staff to inspect remote computers from their own desk. These commercial versions contain additional modules for the enthusiast and business user which we believe is good value when you consider the current cost of this kind of Enterprise-level network-wide software.

[Guru3D] With all the hardware changes and the technology industry developing/evolving at an excruciating fast rate. Analyzing hardware ourselves, the pace can drive us crazy. How do you keep up with the new technologies. Trial and error, or do you guys get low-level industry support in terms of hardware and architecture ? How hard is it to keep up with everything.
 

[Adrian]  This remains the toughest challenge; we benefit from the fact that we have been operating for over a decade and therefore have strong links with the industry which allow us to receive low-level support. Otherwise it would not be possible to support new technologies before release (or soon after). Just about all the information we receive is under NDA which prevents us from sharing or discussing any of it.

While a lot of reviewers and enthusiasts are forward looking, we have to continually balance supporting those at the cutting edge (who use GPGPU, hardware assisted cryptography, Windows Server 2008, etc.) and those who do not have the very latest kit but would like to use our software.

With a field so vast, we try not to be "everything for everyone" but know our specific niches.

[Guru3D] The release of Sandra 2009 is just around the corner and SiSoftware making a new approach to the software. For the first time ever we notice GPU benchmarks in Sandra. Why was this decision made ?
 

[Adrian]  We believe the industry is seeing a shift from the model where heavy computational workload is processed on a traditional CPU to a model that uses the (GP)GPU. While in the past the GPU was only used for games and graphics, programmable shaders have changed this perception. With the release of GPGPU programming languages it is now possible to compile most algorithms to run on a (GP)GPU without the developer knowing anything about programmable shaders.

The CPU is in no immediate danger of disappearing; even when running heavy (GP)GPU workloads, a fast (multi-core) CPU to feed all the (GP)GPUs is still needed.

Next page please.