CUBE High Value Performance Computer FAQ

What does HVPC stand for?

High Value Performance Computing.  This is used to make the strong point that CUBE systems are not High Performance Computing (HPC) systems.  They are a compromise between performance and price.  They do deliver greater performance than a simple system, but they give up some very high end capabilities to make their purchase reasonable.

Why does PADT sell CUBE HVPC Systems instead of general purpose computer systems or HPC systems?

There are plenty of people out there who do a great job selling nice general purpose and HPC systems.  We found that in our own business we just could not justify the cost of an HPC system to run our simulations on.  Technology changed to fast and we never seemed to get our money’s worth out of the investments. For years we have been building our own systems making sure to pick components that are fast, but not the fastest to help cut back on costs.  Sometimes we have bought older systems and refurbished them. We shared our practices with some of our customers and they asked PADT if they would be able to  buy a similar systems from us. 

For a long time we resisted this because the price difference between our “value” systems and standard systems was not enough.  With the introduction of new architecture from AMD and very cool motherboards from Supermicro, we were able to build systems that were 1/3 to ½ the cost of “name brand” systems. We felt there was enough value there to offer them as an item our customers could purchase.

Do you offer INTEL® based CUBE HVPC Systems?

Yes! Since we first started offering CUBE HVPC Systems in 2010, we have had numerous requests for to configure FEA machines based on the INTEL CPU. Customer feedback showed that our prices were still far below the pricing of name brand hardware providers and they were happy to purchase these machines from us. We sold our first INTEL® based CPU Systems in late 2011.

All of your literature talks about CFD, FEA, and Explicit Dynamics from ANSYS, Inc., can you use CUBE HPC systems to run other tools?

You can run anything on these systems that use a x86 architecture with Windows or Linux operating systems.  We have created these systems to be a balance between performance and cost for people running ANSYS, Inc. CFD, FEA and Explicit Dynamics.  Keep in mind your performance may not be as good and it may not even run.  We would be more than happy to try things out and see what the actual performance is.

Just how much slower are CUBE HVPC Systems when compared to top of the HPC systems?

This is of course very problem dependent, and system dependent. However, since we began this journey in 2010 we have compiled a ton of ANSYS benchmark data using our systems.  Here are a couple of links to useful and relevant benchmark data for 2012 and beyond!

ANSYS R14 with Nvidia TESLA C2075 GPU Benchmarks:

http://www.padtinc.com/blog/post/2012/04/19/to-gpu-or-not-gpu-with-ansys.aspx

ANSYS Benchmarks with PADT benchmark #7

http://www.padtinc.com/support/benchmark/bm07/default.asp

Can I swap out the CPU's?

Yes, both AMD and INTEL® have designed their socket configurations for future upgradability.  All of our current systems work with the new AMD interlagos processors. Updated for 2012!

Can I use Windows HPC on a micro-cluster or compute server?

You can use it… but we wouldn’t recommend it. We do not load it onto a system that uses Infiniband.  In our benchmarks we have found that it runs about 10-20% slower.  In addition, Windows HPC uses more system resources then Linux.

My IT person says I should use RAID5 and not RAID0, what is up with that?

Your IT person is right for most every usage of disk drives.  When you are doing numerical computation, the files are temporary and huge.  RAID5 slows things down a ton.  Remind them that this is a batch server machine and you don’t need redundant or fault tolerant disk arrays.  You just need speed!

Why don’t you use [Hardware Name] for the [CUBE System Name]?

We looked around and put together systems that fit our definition of High Value Performance Computing.  There may be other options, and we encourage you to suggest them.  To keep costs down we are sticking with what we picked and will adjust as necessary.

What type of support do you offer?

8 hours of phone support and one year parts only warranty is what is included with the purchase of a CUBE.  This helps us to keep the cost down.  If you do need service beyond what comes with the system you can call PADT and we would be happy to provide you with a quote for whatever your additional needs are.

Don’t these systems get really hot?

No, especially if you consider how many cores they have.  AMD’s in the past were very hot; we popped a couple during the summers here in Arizona.  These current chips are simply air cooled and so far have stayed at a reasonable temperature.

Intel Xeon e5-2690 Temperature  & Power usage video:

Don’t these systems pull a lot of power?

Not really.  They are very efficient and even the 128 Core system pulls very little power.  If you need specifics, please contact PADT and we will share with you our estimate for your system configuration.

Intel Xeon e5-2690 Temperature  & Power usage video:

http://www.youtube.com/watch?v=cxeuFq4Aan4&list=UUkT94L2_i9Oc6xhb_7Wu6lQ&index=1&feature=plcp

AMD Opteron Power Usage Video:

http://www.youtube.com/watch?v=pbykX4jAM6w&feature=plcp

Do you offer advice or assistance in picking the right system for my specific needs?

Helping users find the right system is the primary goal of our team.  There are a ton of options out there and we have had customers literally spend a half of a year looking for the “perfect” system.  Once they get it, it is out of date.   We will spend a small amount of time (10-15 minutes) with you and help configure the “good enough” system. 

I don’t see the exact system I want. Can I specify a custom system?

We encourage it.  The systems we present should fit the needs for most users, but if you need something special, let us know.

How often do you look at the market and readjust your system configurations to the new “sweet spot?”

We watch it all the time but will redo our configurations (and prices) once every 3 months.

Do you offer any accessories that you would recommend?

We have handbags and necklaces that would look great with your system…  OK, maybe not.  We do have an accessories page (Click Here) that shows a few items we use and that we feel others might like.  Hardware RAID Controllers, Faster CPU's, 15k SAS2 hard drives, Monitors, keyboards, installing some OpenSource tools, etc…

My IT/Corporate buyer says I have to buy from [Company].  Can you help me justify purchasing from you?

Yes, we can try to help.  In the past we have shared with them how the needs of a simulation user are different and how going with a CUBE HVPC system that is outside of their standard domain, is actually a good thing.  When they see the money they are saving it helps in their decision.

I need help convincing my co-workers/boss/IT department that a HVPC approach is a good fit for us, can you help with that?

We would be more than happy to get on a conference call with anyone in your company that questions HVPC. Maybe it is not the typical purchase for your company, but either way we will explain the pro’s and con’s and let you decide.

How do you build Cube HVPC systems? What quality standards and procedures do you follow? Are you ISO certified?  Are your systems UL or CE certified?

We build them in a cubicle next to where the IT guys sit.  We do not follow a standard procedure and we have no certification.  These are inexpensive systems and to keep them that way we have no ISO or quality program for them.  We can share with you a detailed build document that shows everything we did, what parts we used, and where they came from.

I live in [Some country other than the US].  Can I purchase a Cube HVPC System?

Right now we only sell in the US.  There are a lot of legal issues with selling outside of the US so we are going to keep things simple for now.

I’m so cheap I don’t even want to pay PADT your ridiculously low prices to put these systems together.  Will you tell me what parts you use so I can build my own?

Sure, we’ll send you a list of the parts we usually use.

Will you benchmark my problem on a Cube HVPC System?

Maybe, if we have a machine to run on. In order to keep costs down we don’t keep an inventory and what machines we do have get used for by our engineers.  So please do ask and we will let you know.

Can I have a machine to try out at my place?

No, to keep costs down we don’t keep an inventory for sale or demo. We do have a few machines we can move around, but in most cases it is not feasible. 

Our FEA engineers keep pestering me about using the new NVidia TESLA C2075 GPU with ANSYS. Do you have relevant benchmark data that can help me better understand the benefits of GPU computing for the scientific community?

Yes, please use our PADT benchmark page.  However, for 2012 we do have some of the latest benchmark data on the Nvidia GPU’s using the industry standard V14sp-5 HPC Benchmark

ANSYS R14 with Nvidia TESLA C2075 GPU Benchmarks:

http://www.padtinc.com/blog/post/2012/04/19/to-gpu-or-not-gpu-with-ansys.aspx

ANSYS Benchmarks with PADT benchmark #7

http://www.padtinc.com/support/benchmark/bm07/default.asp

Will [Software Package Name] run on a Cube HVPC System?

All ANSYS, Inc. products run.  These are very standard Linux systems and supported it will probably work.  In some cases, we would have to test other software programs for you.

Will you load [Software Package Name] on my Cube HVPC System before you ship it to me?  Even if it is a competitor of ANSYS, Inc?

Yes.  If you send complete instructions, a support number to call if we have problems, and a simple test case,  loading software and testing it is part of the purchase price.

Can I lease a CUBE HVPC System?

We do not offer leases except to our ANSYS, Inc. software customers that we sell software to and support.  It is another one of those things that drives up the price for everyone. If you need to lease we will be happy to work with a third party leasing company.

I only need extra horsepower for a short period of time, can I rent a CUBE HVPC System for a few weeks or by the month?

If we have systems available, we might be able to work out a short term rental.  However, in the end it is often less expensive to just to buy a system.

Can I rent time on one of your CUBE HVPC Systems?

Sorry, but no.  The systems in house are busy and have confidential customer data on them. 

Some common ANSYS benchmark terms.

Rating- Rating is the primary metric used to report performance results of the Fluent Benchmarks. It is defined as the number of benchmarks that can be run on a given machine (in sequence) in a 24 hour period. It is computed by dividing the number of seconds in a day (86400 seconds) by the number of seconds required to run the benchmark. A higher rating means faster performance.

Speedup - Speedup is the ratio of wall-clock time required to complete a given calculation using a single processor to that of the equivalent calculation performed on a concurrent machine. Its value ranges from 0 to the number of processors used for the parallel run. When the speedup is equal to the number of processors used, the speedup is called perfect or linear. Sometimes the speedup will exceed the number of processors. This is referred to as super-linear speedup, and is often caused by the availability and use of larger amounts of "fast" memory (e.g. cache or local memory) compared to the single processor run.

Efficiency - Efficiency is the speedup normalized by the number of processors used, presented as a percentage. It indicates the overall utilization of the CPUs during a parallel calculation. An efficiency of 100% indicates that each CPU is completely occupied with computation during the course of a run, and corresponds to a linear speedup. An efficiency of 60% indicates that each CPU is performing useful computation only 60% of the time. The remaining time is spent waiting for other functions, such as parallel communication or work on other processors, to complete.

Redzone - The redzone is defined as the region of greater than 66% efficiency. It is arbitrarily selected as a region that it is favorable to perform a production run (to limit the amount of time a processor is idle while waiting, for example, for the communication phase of the calculation to complete).

Redzone Rating - The redzone rating is defined as the maximum rating achieved with a parallel run of a given benchmark and platform that occurs within the redzone.

Peak Rating - The peak rating is defined as the maximum rating achieved for a parallel run of a given benchmark and platform with any number of processors.

Back to Top