Virgin MVR-02 Cosworth

[beelsebob]

The MVR-03 was designed for the first part of its early life in the new (former) Wirth Research CFD facility that was the most powerful untill Renault updatad theirs a few months back.

Arent all sims and computer models limited to how powerful they can be?

No, an better supercomputer of course can generate better results because more information can be inserted. Of course tihs is still dependant of the software running on these supercomputers. But the point is -> more computing power -> better outcoming results. Then I dont take software errors etc. in caution.

The point JET was making I believe is that the FIA limit the amount of computing horsepower you can use in a year, if you build a more powerful supercomputer, you get less time using it.

Of course teams will all probably be cleverly designing their supercomputers to do poorly at the FIA's benchmark, but well at actually producing usable results, but that's rather beside the point.

[siwillems]

Arent all sims and computer models limited to how powerful they can be?

No, an better supercomputer of course can generate better results because more information can be inserted. Of course tihs is still dependant of the software running on these supercomputers. But the point is -> more computing power -> better outcoming results. Then I dont take software errors etc. in caution.
The point JET was making I believe is that the FIA limit the amount of computing horsepower you can use in a year, if you build a more powerful supercomputer, you get less time using it.

Of course teams will all probably be cleverly designing their supercomputers to do poorly at the FIA's benchmark, but well at actually producing usable results, but that's rather beside the point.

I think there limited to the amount of FLOPS they able to run at or another similiar benchmarkable figure, though I remember vaguely wirth believed he was using the maximum he was able to use with the old system and commented he wasn't sure were the absolute maximum would be drawn by the FIA untill they had an audit - that i believe is when they ramped the sytem up with teir upgrade. I wouldn't be suprised to hear of talk about upgading again if Wirth was around, Oh well lets see what will happen shall we?

[siwillems]

http://translate.google.co.uk/translate ... 85.html%2F

[bjpower]

Its almost impossible to "power" of a computer system. too many variables and dependencies on the type of work the system is doing.

[beelsebob]

Its almost impossible to "power" of a computer system. too many variables and dependencies on the type of work the system is doing.

Yes, which is why you benchmark it at some uniformly known task. It sounds like what the FIA are doing is checking the maximum number of floating point operations the system can do in a second, which is probably a reasonable estimation given that CFD is all about floating point vector manipulation. I wonder though if you could get around it by building a system particularly suited to integer work (perhaps even with no dedicated floating point units), and then used high precision fixed-point vectors for your CFD work – would this allow you to completely circumvent the FIA regs?

[ESPImperium]

The MVR-03 was designed for the first part of its early life in the new (former) Wirth Research CFD facility that was the most powerful untill Renault updatad theirs a few months back.

Arent all sims and computer models limited to how powerful they can be?

No, an better supercomputer of course can generate better results because more information can be inserted. Of course tihs is still dependant of the software running on these supercomputers. But the point is -> more computing power -> better outcoming results. Then I dont take software errors etc. in caution.

The upgrade that Wirth did was to eliveate the ammount of bottle-necks that he had in his hardware, the lower ammount of hardware bottlenecks you have the more efficent your stuff works. Whilst he was lowering the bottle-necks he also upgraded his memory and his processors to high efficency memory and processors. This is where he found his extra 60% plus computing power that made his system the best in F1.

The thing is that the ammount of calculations that Virgin can do with it is now drastically lower due to them using a wind tunnel as they FIA have a equivelency formula that is like this: 45% CFD/30% Tunnel/20% Simulator (If they have one)5% Straightline Testing. Virgin are pretty much on that like mots of the teams out there now, however theese figures can be adjusted to the ammount of straightline days you do as a days straightline can be traded for a days full scale in the tunnel or so many extra calculations in CFD.

Many teams are now upgrading their facilities every 2 years on average with continual improvement in CFD software ongoing, whitch is the next grey area, and the RRA is in focus at present with the alleged Red Bull overspend. One thing that has been talked about is allocating credits for CFD usage, Tunnel usage and Straight-line and Simulator usage, however id like to see straigtline days changed for testing days arround a full track twice a year with all teams attending one three day test or a couple two day tests with young drivers on board.

But i think the RRA will be sorted out and clarified within the next 6 to 8 months whitch will make interesting reading if it ever sees the cold light of day for all to read in a totally transparent manner.

[bjpower]

Its almost impossible to "power" of a computer system. too many variables and dependencies on the type of work the system is doing.

Yes, which is why you benchmark it at some uniformly known task. It sounds like what the FIA are doing is checking the maximum number of floating point operations the system can do in a second, which is probably a reasonable estimation given that CFD is all about floating point vector manipulation. I wonder though if you could get around it by building a system particularly suited to integer work (perhaps even with no dedicated floating point units), and then used high precision fixed-point vectors for your CFD work – would this allow you to completely circumvent the FIA regs?

It really looks like one of them unenforceable rules. there are so many ways around it. offload the floating point to unused pcs in the office - similar to setti etc.
It would also make the systems more expensive. you would have as many sub processors as possible, toe cards etc. you would probably end up using amd apu's so you get the graphics chip for the floating point and a few x86 cores as well.
or you could offload everything to a cloud provider and claim you don't know anything about the hardware.

[MIKEY_!]

Yes computer farms are probably the way to go.

[beelsebob]

Its almost impossible to "power" of a computer system. too many variables and dependencies on the type of work the system is doing.

Yes, which is why you benchmark it at some uniformly known task. It sounds like what the FIA are doing is checking the maximum number of floating point operations the system can do in a second, which is probably a reasonable estimation given that CFD is all about floating point vector manipulation. I wonder though if you could get around it by building a system particularly suited to integer work (perhaps even with no dedicated floating point units), and then used high precision fixed-point vectors for your CFD work – would this allow you to completely circumvent the FIA regs?

It really looks like one of them unenforceable rules. there are so many ways around it. offload the floating point to unused pcs in the office - similar to setti etc.
It would also make the systems more expensive. you would have as many sub processors as possible, toe cards etc. you would probably end up using amd apu's so you get the graphics chip for the floating point and a few x86 cores as well.
or you could offload everything to a cloud provider and claim you don't know anything about the hardware.

I'm certain that this would not get round the issue – when talking about "a supercomputer", what is meant is actually a cluster of much smaller computers, not just in the case you describe, but all cases – the FIA would treat the whole office as "a supercomputer", and would include the FP performance of all the other machines.

[TzeiTzei]

Isn't RRA a FOTA thing? FIA have nothing to do with it.

[gridwalker]

BJPower mentioned offloading calculations to other PCs, like "setti" ... I assume you mean the SETI@home volunteer computing project?

I'm a keen advocate of volunteer computing networks; I currently volunteer my spare CPU clock-cycles for Rosetta@home (analysing protein structures for medicinal use) and LHC@home (hunting for the Higgs boson) but I wouldn't recommend such a system for CFD applications for a number of technical reasons. I won't bore you with too many details, so I will try and summarise.

Firstly; you cannot perform time critical operations using the process, as there is no guarantee that the volunteer processing any particular chunk of data will even turn their computer on during a set period. You could be waiting weeks for a critical calculation whilst a volunteer is on holiday, unless you are willing to allow multiple copies of the data set to be transmitted to ensure redundant capacity.

Secondly; calculations need to be data-light but processor heavy to ensure the efficiency of the network. Therefore it would be impossible to transfer complex aerodynamic models through to volunteers for analysis. The data sets submitted for volunteer computing need to be carefully selected in order to ensure that the complexity of the IT Infrastructure required is cost effective in terms of bandwith required offset by processing power gained in return.

Thirdly; As mentioned before, you need to send out multiple copies of any particular data set in order to guarantee a complete set of results. Without multiple redundancies in place, data could be lost due to an error or altered by competitors. For a smaller team, it would be difficult to secure sufficient volunteers to ensure a large enough network that can guarantee both swift processing and an accurate set of results.

If there was a way that I could volunteer my computer to an F1 team then I would probably do it, but I do not see how these practical considerations can be overcome in the near future.

[raymondu999]

Gridwalker, he means offloading it to other computers in the office rather than volunteers

[gridwalker]

That would still be easy enough to audit then; too much cost, not enough benefit IMHO

[kalinka]

Also if you offload it on many office clients, there is always an increased possibility of industrial espionage. Large, complicated networks are always much more vulnerable. Probably you don't want any of those clients to be connected on internet, what would then makes other jobs very difficult. As for audit, it's not quite that simple as it seems. You can bring in quite easily a few powerful servers just for the weekend, and store them elswhere...FIA really need an auditor who actually has a workplace at every team if they want to be 100% sure. I'd say scrap wind tunnels completely,and enable CFD nonrestricted. Wind tunnel upgrades are much more expensive than CFD upgrades + small teams probably never can build own wind tunnel....

[MIKEY_!]

actually, good point bye bye computer farms!