Link to home
Start Free TrialLog in
Avatar of Ex0 SyS
Ex0 SySFlag for Switzerland

asked on

PLS We need very URGENT Specific Help to make the good choice for vHPC Arbitrary Precision Arithmetic Modular Multicalculator CPU, Co-CPU, GPU Computing Unit Server (Hardware/Software)...

Dear Admins, Dear Mods, Dear Experts,

We are a Swiss compagny Inovating in IT / Mathematical / Scientist Tchnologies and today we really need to go on next production step that requires very modular high precision computing resources hardware, lilbraries and frameworks.

Actually we need to be able to do very high arbitrary arithmetic precision calculation (ex. 1'000'000'000 to 1'000'000'000'000'000'000'000'000'000 Up +++ Bits Numbers), Very High GPU Computing Tasks, 3D Rendering, Virtualisation Simulations an ALL of that with the Benefit of the Lastest Calculation Hardware Technologies on the Market like "Double-Internal Xeon CPU v4", "Xeon Phi" and "Nvidia Tesla" PCIE 16x v3 Computing Units.

One very frustrating thing is that "I personnaly" need to continue using the very well know and such common basic tool called Microsft Excel in parallel of our Dev Team (I'm the Creator not the Genius Dev and I need to continue using Excel to provide my Prooves-Of-Concepts Templates because I have very high skills with this software)...

Actually I can use some free Excel addons like "XNUMBERS" or "MATRIX" or "Mark McIlroy Arbitratry Precision Number Calculations" or "xlPrecision (just bought the 2,147,483,648SD Research Edition)" or "NAG Library Excel Addin" or "QuantLib" and they all seems to be nice but are they Powerfull enough, because actually I don't really know if all that kind of solutions are limited by my (the) hardware or by their own internal limits or maybe Microft Excel's own's internal limits (Do you have any recommendations) ?!

Do you know if there is any kind of "Wolfram Alpha" or "Any Other" Solutions that are able to integrate High Arrbitrary Precision Computing Technology in an easy way to be used with common Applications such as Microsoft Excel (A real need for me) or any common development framworks such as C/++#/Java/Python/... ?

Is it possible to use any kind of "Wolfram Alpha" or "Any Other" High Arrbitrary Precision Computing Technology Solutions to be setup on a High Performance Computing Server (HPC) such as "Microsoft HPC Server 2008 R2" or any "Linux/UNIX Server Release" to be able to use "HPC Services on any Workstations" to Compute Tasks in Live over HPC Services ?

Today we are building a Powerfull Virtulized HPC Server (vHPC) based on low level hypervisor server like ESXi to perform many different tasks, the most important one is the high arbitrary arithmetic precision calculation and also Very-High GPU Computing and 3D Rendering.

We want to use 2 Integrated Xeon CPU E5-2687Wv4 plus 1 Nvidia Graphic Card GeForce 1080 with also 2 Co-Processor Xeon Phi (1 today 1 in a close futur) and 1 more Nvidia Tesla Unit in a medium far future, all of that on a Single Modular Machine with a Huge 2000W Power Supply and 128 GB DDR4-2400, what do you think about this kind of Hardware Configuration ?

It seems that the best approach is a "Virtual High Performance HPC Clustering Server" to perfrorm calculations using HPC Services on Workstations to experience the best hardware resources by increasing the exploitation of the Physical Server Hardware on many multiple virtual clustered (vHPC) machines with intermediate power each rather than direct access to all the material resources at full power on a single physical unit. (Apparently the  Johns Hopkins Applied Physics University has opted for this kind of HPC Virtualization Technology "vHPC", so I think that this choice should be appropriate given the similar nature of the calculations we have to deal with)

In this context I think we should use low Hypervisors Systems to operate and actually allocate hardware resources to virtual machines and in this area there are only "VMware vSphere ESXi Server Center" that is capable to do so simply, what do you think about that?

All the material that we must choose must be compatible with "VMware vSphere ESXi Server Center," I think especially in the Hypervisors systems installation on HDD SAMSUNG 950 M.2, operational data on a cluster SAMSUNG SSD 850 RAID1 and the management thereof in the infrastructure of the VMware hypervisor (drivers and others), once again do you think it might be a good solution or not ?

Finaly do you have any other good recommendations for the best way for us to get out the concepting phase to go to the production level with the best Hardware and Software technologies and what is the best Framework to be easily able to compute VERY BIG NUMBERS ?

I know that we have very high specific needs but I really don't have the knowledge to do the best choice to be able to compute very VERY LARGE Numbers (1'000'000'000'000'000'000'000'000'000'000 +++ Bits) with high arbitrary arithmetic precision calculation on Microsoft Excel and common development frameworks, what is the best Computing Solution for it : Virtualization or NOT, "Inte Co-Processor Xeon Phi" and/or "Nividia Multi-GPU Tesla and/or "Internal Dual Intel Xeon CPU v4" or maybe ALL or NONE of them ???

I really hope that you will understand our needs and that you will be able to answer them soon in a positive way.

Thanks in advance and I hope to read you ASAP .. ... .....

Best regards to all readers,

JN



P.S. Sorry for my bad English but I really hope that you will be able to understand my real needs, it's such a difficult task for me to find the right Hardware/Software Combo for such specifics Scientist / Mathematics / iT Engineering Solutions, I'm only a Creator not a Genius...



Here are some useful links about Virtual High Performance Computing (vHPC), Xeon Phi and Nvidia Tesla Computing Units comptibility :

http://www.vmware.com/files/pdf/techpaper/vmware-virtualizing-hpc-technical-computing-with-vsphere.pdf
https://www.virtualizationpractice.com/virtualizing-high-performance-computing-750/
https://labs.vmware.com/academic/publications/performance-evaluation-of-hpc-benchmarks-on-vmwares-esxi-server

http://www.nvidia.com/content/gtc/documents/1091_gtc09.pdf
http://www.nvidia.com/object/io_1254126305481.html
https://technet.microsoft.com/en-us/library/ff793340(v=ws.10).aspx
https://blogs.msdn.microsoft.com/somasegar/2010/09/20/windows-hpc-server-2008-r2-the-next-step-in-high-performance-computing/
https://goparallel.sourceforge.net/using-windows-instead-linux-host-xeon-phi-coprocessor/
https://software.intel.com/en-us/forums/intel-many-integrated-core/topic/392264
https://software.intel.com/en-us/articles/windows-early-enabling-for-intelr-xeon-phitm-coprocessor
http://www8.hp.com/h20195/v2/getpdf.aspx/c04168329.pdf?ver=5
Avatar of Andrew Hancock (VMware vExpert PRO / EE Fellow/British Beekeeper)
Andrew Hancock (VMware vExpert PRO / EE Fellow/British Beekeeper)
Flag of United Kingdom of Great Britain and Northern Ireland image

We've just been involved in the process of building tender information for clients, and inviting Vendors to tender for a HPC.

What's your budget and do you want to do CPU or GPU, or both ?

What overall performance do you want to purchase, and do you want to purchase small and grow the cluster ?
I see you have joined ExEx yesterday and have a fairly complicated question about high performance computing and Excel.  Maybe it would be better split into two or more questions.

Is this all related to Ex0-BAMT and cryptocurrency mining?  

Do you hope to be building, buying, or selling hardware and or software?
Intel has 80 bits of precision in the significant using long double(s)
Excel only has 15 significant digits https://en.wikipedia.org/wiki/Numeric_precision_in_Microsoft_Excel and doesn't use the long double
Avatar of Member_2_231077
Member_2_231077

Presumably we have to read the whole question plus the links before answering?

I'm a bit stuck on the first link, it's premise is that virtualization can improve HPC performance.  This seems doubtful to me since VMware style virtualization is splitting one powerful computer into several smaller logical ones whereas HPC is sticking several small computers together to create a larger logical one. I just can't see why splitting the CPU up into lots of individual time slices and sticking them back together again can improve speed. There is of course the trivial case where the HPC software can't run more than one thread or can't utilize all the RAM but that's a limitation on the HP side, not a positive improvement introduced by virtualization.

Then I reached the bottom of the article and it turned out that the improvement was only 4%, the margin of error for the benchmark software is probably more than that.

Oh, by the way, you can do high precision maths with Excel; you can add double precision to any calculator, you just have to write the macro.
Avatar of Ex0 SyS

ASKER

Hello to all,

First of all I would like to thank everybody that took the time to read and give me a feedback from this (those) very complicated questions.

To be clear, we do not want to sell any kind of hardware solutions, this is really far from our core business that is dedicated to Inovate in Mathematical / Scientist / IT Engineering Tchnologies.

We just need to be able to build (assembly) a powerful machine to do many kind of different tasks, in fact it have to be a Modular Machine, I mean one that will be able to be very evolutive matching our actual needs (today) and future ones without the need to buy a full new hardware configuration each time.

When I talk about a modular machine, It means that we want to be able to add more component in the future, today we need 2 Internal "Xeon E5-2687Wv4 CPU" but we already know that we will need to add minimum one "Co-Processor Xeon Phi" and  one more "Nvidia Tesla GPU Computing" Units related to our future needs and ALL your recommendations.

Our first absolut need today is to be able to compute very larges numbers (1'000'000'000 +++ Bits), for exemple we've just bought "xlPrecision largest version" to be able to do it with more than billions arbitrary precision calculations over Microsoft Excel and of course we also need to do it  with any common dev frameworks without limits (C/++/#/Java/.......)

Here I really don't know what kind of "Library/Framework" to choose to be able to do it (Ex. The GMP Library looks very good but it's mostly focused for Linux so what can we do if we need (have to) continue to use some of our Workstations (Even some Dev's ones) running Microsoft Windows environments ?

In a close future, we will also need to be able to do very High Performance 3D Simulations (Simple ones but on VERY VERY VERY Large maps, imagine something like to simulate tons of Stars Positions or any H2O Particles inside a swimmingpool or something as crazy than that...)

Maybe here would be the real need of High Computing Units like "Xeon Phi" and "Nvidia TElsa" and this is a real quetion for you all Experts :

What would be your recommendations about what kind of hardware would be better to choose for that kind of specific tasks (3D High Computing of course excluding Video Game, Mathematical High Precision Calculations / IT Engineering....) ?

Do someone knows what kind of hardware is actually used by any companies or organisation that needs to do something similar from us ?

After that I noticed that you "andyalder" don't really agree the approach of Virtualizing an HPC Server to increase performances, just to know I've only been inspired by the article about the Johns Hopkins University that made this choice but I really dont have enough knowledge in this domain to be able to well appreciate the situation.

The only thing I know about Virtual solutions is that there is two Categories of Hypervisor, those are running on Hosted Systems like Windows or Linux/UNIX (Ex. Micosoft HyperV or VMware Workstations) and those are Themselves an Operating System that is able to use the Hardware in low-level access that usually means with High Performances (Ex. ESXi, vCenter Server and other VMware (v)'s solutions).

Related to this psecific point, just to be clear with everybody, (vHPC) means "Virtual High Performance Computing" and not something related to "Virtualized HP Helwet Packard Specific Servers", so what would be your best recommendations to be able to do "Live-Heavy-Computing-Calculations" directly from our Workstations (Actualy one Workstation "mine" and soon minimum one more Workstation for the Dev) ?

Maybe a Local-Powerful-Workstation with all Hardware performances dedicated to it or something like Standard-Workstations and a Powerfull "Live-Computing-Server" and in this case what would be the best OS/Software to do it without forgetting my real need to be able to continue using Microoft Excel ?

Honnestly the only thing I know about HPC Computing Server is that they seems to be able to very well use the Hardware of "Xeon Phi Co-Processors" and "Nvidia Tesla" High-Computing-Units and I think it will be useful for our future 3D simulations tasks and here is why I'm talking about "HPC Server Solutions" but I dont know anything else about this technologiy, so please, what is your opinions ?

Finally to give an answer to "d-glitch", yes I'm the creator of the "Ex0-BAMT Project" in Crypocurrencies domain but as you were able to see I had to temporary paused this project, so today my questions are focused on very differents needs but if you are interested we will continue the "Ex0-BAMT Project" in about 1 year and we will need some more active members to join it at the right time (you will be able to contact me with this address : ex0.bamt@gmail.com).

To react about "phoffric" recommendations and to everybody that took the time to read my posts, I have to say sorry for the huge lenght of it (them) and I have to to explain you all why I did it like that without splitting every questions in diffrents posts.

It's only because it is ONE global question about SOME specifics needs that we have actualy have for ONE specific Project that we are working on today and it would be very difficult for me to manage different posts at the same time because I'm really not a blogger or an heavy forum poster and I've to admits that I have very poor skills and experiences in this domain

Just to remember my position, I'm a "Creator" not an "Engineer" or a "IT Programmer" and in this context my habits are usually to give the most possible detailed explainations for every kinds of my analysis, so please try to be indulgent with me about that and If you dont give me the choice I will try to do my best in another way but please do not BAN me for that...

Wow.... it's finally done and it is not such a lite post, so I hope you will try to take the time again to read it and I to provide me your opinions ?

I wish you all my best regards and I hope to read you ASAP .. ... .....

Sincerely,

JN


INFORMATIONS ABOUT THE EVOLUTIVE CHOOSEN HARDWARE (I'M WAITING YOUR RECOMMENDATIONS BEFORE BUYING IT) :

1 x Case - Corsair Carbide Air 540
1 x Power supply - Super Flower Leadex 80 Plus Platinum 8Pack Edition - 2000W
1 x Motherboard (Dual CPU / 3 PCIe 16x v3.0) - Asus Z10PE-D16 WS
2 x Internal CPU's - Intel XEON E5-2687Wv4 3.00 GHz 12 Cores
2 x CPU Fans - Noctua NH-U9S
8 x RAM-Pack - Kingston 16GB DDR4-2400MHz REG ECC CL17 128 GB
1 x Graphic Card (Low GPU Computing) - MSI GeForce GTX 1080 Gaming X 8G
1 x UltraHigh Perfs Media (Install OS's/Hypervors) - Samsung M.2 950 PRO - 512GB
2 x High Perfs (RAID1 Exploitation Datas) - Samsung SSD 850 PRO - 1TB
2 x Medium Perfs (RAID0 Backup) - Seagate Desktop SSHD 3.5" SATA 6Gb/s - 4TB
ASKER CERTIFIED SOLUTION
Avatar of Andrew Hancock (VMware vExpert PRO / EE Fellow/British Beekeeper)
Andrew Hancock (VMware vExpert PRO / EE Fellow/British Beekeeper)
Flag of United Kingdom of Great Britain and Northern Ireland image

Link to home
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
Start Free Trial
Avatar of Ex0 SyS

ASKER

Thank you Andrew for your so fast answer!

Are you always connected, do you take the time to sleep for a while sometime ?

Tesla's GPUs and Xeon Phi Computing would be exactly what we need, but how to be able exploiting their amazing capabilities, of course directly on the server itself, but the most important for me today is to be able to get the benefit of all their High Performances from any Workstations using Microsoft Excel + xlPrecision 2,147,483,648 SD Edition (http://precisioncalc.com/2,147,483,648SD.html)

I know it is a strange need using Excel for that kind of Computing tasks but as I already explained, I'm a Creator, not an IT Engineer or Programmer and the only tool with the one I personnaly have a very high knowledge is Microsoft Excel, I know it's a shame but please don't blame me ;)

Of course for all other parts of this project, we are totally agree with you to do our Computing Tasks directly on the HPC or vHPC Server or even developping directly on it according with the Dev Team.

Could you please recommend me which OS Version to choose as the HPC Server (Linux Release ? / Microsoft Windows HPC Server 2008 R2 ?) without loosing the focus on my necessary need of using Excel.

Finally what do you think about the Johns Hokins University choice to use vHPC Replicated many times in a Virtual Cluster Evironment sharing hardware performances but allowing Multi-Multi-Threads at the same time or it would definitly be better to use one Physical Machine dedicated to use all hardware capability ?

I hope that I gave you enough details about our (my) needs and that you would be able to give me one more of you good answers.

Thank you for your interest and you support and hope to read you again soon .. ... .....

Sincerely,

JN
Red Hat.

Although we use VMware vSphere extensively, we do not use it for HPC.

HPC is on bare metal.
Avatar of Ex0 SyS

ASKER

Hello dear Andrew,

I've already noticed that the Red Hat OS HPC Solution is very recurrent and seems to be very stable, very powerful, very well done, very well all...

However, what's about the way to use Microsoft Excel from our Microsoft Windows Workstations to be able to "Live-Compute" any kinds of "High Precision Opperations" over the "Linux Red Hat-HPC Server" ?

Don't you think that would be better to use the Microsoft HPC Solutions concerning the Excel Compatibility like "Windows Server 2012 R2" including the "HPC Pack 2012 R2" and maybe to setup an on-demand "Microsoft Azure" worker instances (only if it's possible to locally setup an Azure Cloud) ?

Here is a link about how to set up a hybrid high performance computing (HPC) cluster with Microsoft HPC Pack and on-demand Azure worker instances, could you please give me your opinion about it ?

https://azure.microsoft.com/en-us/documentation/articles/cloud-services-setup-hybrid-hpcpack-cluster/

Thank you very much for all of your recommendantions and as usual hope to read you soon...

Sincerely,

JN
SOLUTION
Link to home
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
Start Free Trial
Avatar of Ex0 SyS

ASKER

I understand that and as I already explained, our Dev Team will do exactly the same way.

Ecxel is only a ridiculous but absolute need for me and only for me because this software simply allows me to translate all my personnal technical and logical needs in the fastest, effective and easiest possible way...

I will try HPC Microsoft Solutions first and I will give you a feedback about it if you are interested...

For now I have to make some simulations about the way to build my new environment and I will come back here to give my feedbacks.

Thanks, bye and see you soon .. ... .....

Sincerely,

JN
SOLUTION
Link to home
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
Start Free Trial
Avatar of Ex0 SyS

ASKER

Thanks you all for you helpful recommendations .

I'll give you a feedback ASAP .. ... .....

Regards,

JN