Orad’s DVG :Orad’s DVG : solutions for scalable solutions for scalable graphics clustersgraphics clusters

April 19, 2023April 19, 2023

images courtesy MPI, Barcoimages courtesy MPI, Barco images courtesy MPI, Barcoimages courtesy MPI, Barco

2

ORAD HI TEC SYSTEMSORAD HI TEC SYSTEMS

Incorporated 1993Incorporated 1993

Leader in 3D viz’ for the television industry, 2 emmy awards...Leader in 3D viz’ for the television industry, 2 emmy awards...

Traded at Frankfurt prime-standard stock exchange, worldwide Traded at Frankfurt prime-standard stock exchange, worldwide offices. Also traded at AIM stock exchange in London. 18M$ offices. Also traded at AIM stock exchange in London. 18M$ revenue, 9M$ cash assets.revenue, 9M$ cash assets.

Long experience in use and customization of 3D real-time Long experience in use and customization of 3D real-time visualisation platforms ( Onyx, PC…) Orad is a user as well as a visualisation platforms ( Onyx, PC…) Orad is a user as well as a supplier of HW.supplier of HW.

Partnering with Intel ( one of Orad’s shareholder ) Nvidia… Partnering with Intel ( one of Orad’s shareholder ) Nvidia… leaders in processing and chip designleaders in processing and chip design

Successful diversification in HW design 2 years ago : now the Successful diversification in HW design 2 years ago : now the largest installed base in scalable graphics systems with DVG.largest installed base in scalable graphics systems with DVG.

3

Orad’s DVGOrad’s DVG

The world’s most powerful visualisation cluster composed of MULTIPLE COTS rendering units tied together to produce a single output channel

The world’s most powerful visualisation cluster composed of MULTIPLE COTS rendering units tied together to produce a single output channel

Use any number of DVG nodes to address ANY level

of scene complexity

Use any number of DVG nodes to address ANY level

of scene complexity

4

RationaleRationale

PC “high end 3d game cards” have PC “high end 3d game cards” have advanced tremendously in the past few advanced tremendously in the past few years, reached and exceeded sgi’s years, reached and exceeded sgi’s performanceperformance

Single chip design Single chip design price/performance price/performance ratioratio

Current situation:Current situation:

Requirement for photorealism and good Requirement for photorealism and good image quality was unachievable by game image quality was unachievable by game cards (performance and quality) nor by sgi cards (performance and quality) nor by sgi (performance and price)(performance and price)

5

Scalable architectureScalable architecture

e.g. : compositing 3 cards to one channel in HP workstation

6

The DVG architectureThe DVG architecture

PC cluster solution for PC cluster solution for scalable scalable performance on multiple or performance on multiple or SINGLE displays.SINGLE displays.

M nodes drive N displays, M & N are any numbers, M>NM nodes drive N displays, M & N are any numbers, M>N

At the heart of the DVG technology is the DVG board, a highly At the heart of the DVG technology is the DVG board, a highly programmable (FPGA) PCI card.programmable (FPGA) PCI card.

=> achieving genlocking AND combining.=> achieving genlocking AND combining.

The DVG board The DVG board directly interactsdirectly interacts with the graphical board with the graphical board allowing allowing fully linear scalingfully linear scaling of performance of performance

A A dedicated patent-pending pixel bus “DVG-bus” interconnects dedicated patent-pending pixel bus “DVG-bus” interconnects the DVG units so the DVG units so nono graphics , CPU or motherboard resources graphics , CPU or motherboard resources are sacrificedare sacrificed

DVG board occupies space of 1 to 2 PCI slots depending on DVG board occupies space of 1 to 2 PCI slots depending on version.version.

7

DVG ’s « U-sync » , multi OS supportDVG ’s « U-sync » , multi OS support

U-Sync ( patent pending ) : achieves universal U-Sync ( patent pending ) : achieves universal synchronisation of ANY off-the-shelf board. ATI and synchronisation of ANY off-the-shelf board. ATI and Nvidia. Boards certified today are NV35 and R350 Nvidia. Boards certified today are NV35 and R350 based.based.

Multi OS support : software drivers for DVG in Multi OS support : software drivers for DVG in Windows 2000, or Linux, both proven on the field.Windows 2000, or Linux, both proven on the field.

=> No dependance on any vendor, the client => No dependance on any vendor, the client system will upgrade in sync with mass system will upgrade in sync with mass market evolution. Client will decide system market evolution. Client will decide system ressources according to his own present ressources according to his own present preferencespreferences

8

The DVG supports several different combining modes:The DVG supports several different combining modes:

- Sample division (for Anti-Aliasing)Sample division (for Anti-Aliasing)

- Time divisionTime division

- Image divisionImage division

- Eye division etc...Eye division etc...

Composition is done on board, no need to move pixels back into memory Composition is done on board, no need to move pixels back into memory with associated overhead in performance and # of CPUs.with associated overhead in performance and # of CPUs.

Various chaining modes can be mixedVarious chaining modes can be mixed

Modes can be switched with no change of wiringModes can be switched with no change of wiring

Combining configuration can be chosen according to the critical Combining configuration can be chosen according to the critical system/data-base resourcesystem/data-base resource

Combining ( a.k.a. “chaining” )Combining ( a.k.a. “chaining” )

9

More on combining : Anti-aliasingMore on combining : Anti-aliasing

Anti-aliasing is critical for image qualityAnti-aliasing is critical for image quality

Both Nvidia and ATI based DVGs can do 2-sample anti-aliasing with Both Nvidia and ATI based DVGs can do 2-sample anti-aliasing with no performance lossno performance loss

For comparison, Onyx2 can do 4-sample anti-aliasing, 8 or moreFor comparison, Onyx2 can do 4-sample anti-aliasing, 8 or more

One DVG rendering unit can implement 4-sample (also 6 samples in One DVG rendering unit can implement 4-sample (also 6 samples in the case of Radeon 9700 and above ) anti-aliasing but with a certain the case of Radeon 9700 and above ) anti-aliasing but with a certain performance cost ( fill rate )performance cost ( fill rate )

Multiple DVG rendering units connected together (see below) can do Multiple DVG rendering units connected together (see below) can do up to 16-sample anti-aliasingup to 16-sample anti-aliasing

10

More on combining : Sample chaining More on combining : Sample chaining (for Anti-aliasing)(for Anti-aliasing)

The purpose of this mode is to generate higher than 2-sample anti-The purpose of this mode is to generate higher than 2-sample anti-aliasing qualityaliasing quality

4-sample anti-aliasing can be performed by two units at twice the 4-sample anti-aliasing can be performed by two units at twice the speed of a single unitspeed of a single unit

8-sample anti-aliasing can also be performed by 8-sample anti-aliasing can also be performed by twotwo units, at lower units, at lower speedspeed

For higher quality and/or higher performance, three or four units can For higher quality and/or higher performance, three or four units can participate in the chainparticipate in the chain

This mode does This mode does notnot add delay add delay

11

More on combining : More on combining : Time-division chainingTime-division chaining

This mode scales performance linearly with the number of units This mode scales performance linearly with the number of units in the chain, without any limitin the chain, without any limit

With N units in the chain and 60Hz clock each node generates With N units in the chain and 60Hz clock each node generates 60/N images per second.60/N images per second.

Therefore each rendering unit has N field times to generate its Therefore each rendering unit has N field times to generate its graphicsgraphics

So it can render graphics N times more complex than a single So it can render graphics N times more complex than a single unit couldunit could

The final output of course is still a 60 images per second The final output of course is still a 60 images per second displaydisplay

This mode generates latency of N-1 frame timesThis mode generates latency of N-1 frame times

12

Time division chaining, cont’dTime division chaining, cont’d

In systems such as flight simulators, too big latency is undesired, In systems such as flight simulators, too big latency is undesired, so the length of time division chain must be limited. However, it can so the length of time division chain must be limited. However, it can be mixed with sample division chain, which is also increasing the be mixed with sample division chain, which is also increasing the performance. performance.

But if time division is used to increase the frame rate, then the But if time division is used to increase the frame rate, then the effective latency effective latency does notdoes not increase. Example: increase. Example:

– Model runs at 5Hz on 60Hz display. The natural latency (or Model runs at 5Hz on 60Hz display. The natural latency (or response) is 60 / 5 = response) is 60 / 5 = 1212 frames. frames.

– If the performance is scaled up to 20Hz by using time division of If the performance is scaled up to 20Hz by using time division of 4 units, then the natural latency is 60 / 20 = 3 frames.4 units, then the natural latency is 60 / 20 = 3 frames.

– We need to add 4 - 1 = 3 frames of latency introduced by the We need to add 4 - 1 = 3 frames of latency introduced by the chain. But they are 20Hz, not 60Hz frames, so we need to chain. But they are 20Hz, not 60Hz frames, so we need to multiply them by 3. Thus, combined latency is 3 * 3 + 3 = multiply them by 3. Thus, combined latency is 3 * 3 + 3 = 1212 frames.frames.

13

More on combining :More on combining : Imaging division chaining Imaging division chaining

Each rendering unit is “responsible” for a sub-region of the Each rendering unit is “responsible” for a sub-region of the final output imagefinal output image

This mode is better suited for applications where performance This mode is better suited for applications where performance is limited by pixel fill such as high resolution formats is limited by pixel fill such as high resolution formats

NoNo added latency added latency

Special dynamic load balance algorithms:Special dynamic load balance algorithms:

““Quadrant division” with a dynamic rendering time criterionQuadrant division” with a dynamic rendering time criterion

““Interleaved division”Interleaved division”

The “Quadrant division” method can also improve the The “Quadrant division” method can also improve the geometry processing using cullinggeometry processing using culling

14

Imaging division chaining cont’d Imaging division chaining cont’d

Image division effectively increases available texture memory, Image division effectively increases available texture memory, even more so as it is combined with AA chaining :even more so as it is combined with AA chaining :

– eg : a 1600x1200 renderer at 6 samples uses 95Mb of texture eg : a 1600x1200 renderer at 6 samples uses 95Mb of texture memory for buffering, remaining only memory for buffering, remaining only 33Mb 33Mb ( assuming 128 Mb ( assuming 128 Mb total )total )

While 1024 x 768 rendering screen at 2 samples uses 15Mb of While 1024 x 768 rendering screen at 2 samples uses 15Mb of texture leaving texture leaving 133 Mb available !133 Mb available !

=> Image division and anti-aliasing chaining => Image division and anti-aliasing chaining increase available texture memory for increase available texture memory for

graphic texturesgraphic textures

15

Split image division chainingSplit image division chaining

Each unit renders fragment of the scene. Combiner creates output Each unit renders fragment of the scene. Combiner creates output image bigger than input components.image bigger than input components.

Both vertical and horizontal image division is allowed Both vertical and horizontal image division is allowed

Application can use view culling to gain geometry rateApplication can use view culling to gain geometry rate

Static load-balanced gain on pixel fill rateStatic load-balanced gain on pixel fill rate

++ ++ ==

16

Interleaved image division chainingInterleaved image division chaining

Each unit renders full scene, but in window “squeezed” horizontallyEach unit renders full scene, but in window “squeezed” horizontally

Each window has projection matrix shifted by a subpixel. Combiner Each window has projection matrix shifted by a subpixel. Combiner interleaves pixels to produce output image with higher resolution.interleaves pixels to produce output image with higher resolution.

No gain on geometry rate but dynamic load-balanced gain on pixel No gain on geometry rate but dynamic load-balanced gain on pixel fill ratefill rate

Cannot use antialiasing of nVidia and ATi until programmable Cannot use antialiasing of nVidia and ATi until programmable sample locations are available.sample locations are available.

++ ++ ==

17

Dynamic image division chainingDynamic image division chaining

Each unit renders fragment of the scene in viewport smaller than full Each unit renders fragment of the scene in viewport smaller than full window. The rest of window is filled with black. Combiner adds window. The rest of window is filled with black. Combiner adds images.images.

Application can use view culling to gain geometry rate.Application can use view culling to gain geometry rate.

Viewports can be resized on the fly, so application can do dynamic Viewports can be resized on the fly, so application can do dynamic load balancing for pixel fill rate.load balancing for pixel fill rate.

Overhead time (e.g. ‘swapbuffers’) is bigger (because each unit Overhead time (e.g. ‘swapbuffers’) is bigger (because each unit renders in full window).renders in full window).

++ ==

18

More on combining : More on combining : Eye division chainingEye division chaining

For active stereo ( 96,110,120hz ), left/right eye division : linear For active stereo ( 96,110,120hz ), left/right eye division : linear performance increaseperformance increase

The first chosen method in CAVE, VRs, as it is one of the most The first chosen method in CAVE, VRs, as it is one of the most efficient.efficient.

No added delay No added delay

19

More on combining : More on combining : Scene division chainingScene division chaining

The 3d objects in the scene are split between the different The 3d objects in the scene are split between the different rendering unitsrendering units

Each pixel’s distance to the camera (its Z value) is transmitted Each pixel’s distance to the camera (its Z value) is transmitted between rendering unitsbetween rendering units

Based on this information, the DVG boards in the chain Based on this information, the DVG boards in the chain correctly composite the final imagecorrectly composite the final image

Performance scales linearly with the number of rendering unitsPerformance scales linearly with the number of rendering units

Differently from the other methods, this method not only Differently from the other methods, this method not only increases total performance, but also effectively increases increases total performance, but also effectively increases graphics resources such as texture memory, beyond the graphics resources such as texture memory, beyond the capability of a single rendering unit.capability of a single rendering unit.

No added delayNo added delay

20

More on combining: More on combining: Volume rendering divisionVolume rendering division

Many visualization applications, Many visualization applications, primarily in oil & gas and medical primarily in oil & gas and medical imaging are based on volume, imaging are based on volume, not polygonial renderingnot polygonial rendering

The process can be The process can be implemented on COTS graphics implemented on COTS graphics hardwarehardware

Orad provides the market’s only Orad provides the market’s only cluster-based solution for volume cluster-based solution for volume rendering applications rendering applications

The cluster linearly increases The cluster linearly increases performanceperformance

no added delayno added delay

21

DVG FormatsDVG Formats

The DVG supports all formats up to output pixel frequencyThe DVG supports all formats up to output pixel frequency

of of 300 MHz (!!!),300 MHz (!!!), the Onyx is limited to 220 MHzthe Onyx is limited to 220 MHz

Supported VESA and other standards:Supported VESA and other standards:

All SDI and HD video formatsAll SDI and HD video formats

All 800x600, 1024x768, 1280x960, 1280x1024 formatsAll 800x600, 1024x768, 1280x960, 1280x1024 formats

1280x1024 120Hz (stereo)1280x1024 120Hz (stereo)

1600x1200 (60-85Hz)1600x1200 (60-85Hz)

1792x1344 (60,75Hz)1792x1344 (60,75Hz)

1856x1392 (60, 75Hz)1856x1392 (60, 75Hz)

1920x1440 (60, 75Hz)1920x1440 (60, 75Hz)

2048x1536 60Hz2048x1536 60Hz

22

DVG upgradability - exampleDVG upgradability - example

DVG has gone through many generations of nVidia graphics h/w DVG has gone through many generations of nVidia graphics h/w since its inception:since its inception:

GeForce2GeForce2GeForce3GeForce3GeForce3 Ti500GeForce3 Ti500GeForce4 Ti4600-> FX GeForce4 Ti4600-> FX 5900 -> Quadro FX30005900 -> Quadro FX3000

Graphics processor clock increased by 50%, memory speed by Graphics processor clock increased by 50%, memory speed by 60%, geometry processing speed more than doubled, effective tex 60%, geometry processing speed more than doubled, effective tex memory grew from 64MB to ~2GB!!!memory grew from 64MB to ~2GB!!!

Sgi has gone from Onyx 2 to Onyx3000 (irrelevant for graphics Sgi has gone from Onyx 2 to Onyx3000 (irrelevant for graphics performance) and from Infinite Reality to ir2 to ir3 and now ir4performance) and from Infinite Reality to ir2 to ir3 and now ir4

Ir2 over ir increased performance by ~20%, ir3 increased Ir2 over ir increased performance by ~20%, ir3 increased onlyonly texture memory to 256MB, ir4 again – only texture memory to 1GB texture memory to 256MB, ir4 again – only texture memory to 1GB and fill rate. The strategy of proprietary graphics was eventually and fill rate. The strategy of proprietary graphics was eventually abandonned.abandonned.

=> DVG is “future proof”=> DVG is “future proof”

23

Software certification and portingSoftware certification and porting

Easy porting of generic (Vega Prime, Performer, Amira, Opticore Opus, Easy porting of generic (Vega Prime, Performer, Amira, Opticore Opus, Virtools…) and customized rendering applications, authoring, assembly and data Virtools…) and customized rendering applications, authoring, assembly and data base manipulation toolsbase manipulation tools

2 methods for porting : 2 methods for porting :

DVGlibDVGlib – a library used for the DVG system administration providing DVG specific – a library used for the DVG system administration providing DVG specific functionality like setting up chain configuration, format, as well as synchronized swap functionality like setting up chain configuration, format, as well as synchronized swap buffer.buffer.

DVG wraperDVG wraper – intercepts OpenGL calls in order to configure rendering for compositing – intercepts OpenGL calls in order to configure rendering for compositing

Successful porting processes have already been implemented by customers like Successful porting processes have already been implemented by customers like British Aerospace, Lockheed Martin and others (typical duration – few hours if British Aerospace, Lockheed Martin and others (typical duration – few hours if application support distributed rendering )application support distributed rendering )

The DVG supports both Windows and Linux based rendering codesThe DVG supports both Windows and Linux based rendering codes

Orad offers porting assistance services by a dedicated teamOrad offers porting assistance services by a dedicated team

The OpenGL driver is the unmodified driver supplied by the graphics card maker The OpenGL driver is the unmodified driver supplied by the graphics card maker and thus provides the most up-to-date extensions and optimizations as they are and thus provides the most up-to-date extensions and optimizations as they are releasedreleased

24

Certified Libraries/applicationsCertified Libraries/applications

Open SGOpen SG

Sgi’s PerformerSgi’s Performer

VRCO’s CAVELibVRCO’s CAVELib

MPI’s Vega PrimeMPI’s Vega Prime

BlueBerry3DBlueBerry3D

Princess Interactive’s SentaxPrincess Interactive’s Sentax

Opticore’s WebViewer and OpusOpticore’s WebViewer and Opus

VirtoolsVirtools

SDS Flight SimulatorSDS Flight Simulator

3D Pipeline3D Pipeline

BAE’s own simulation applicationsBAE’s own simulation applications

Lockheed Martin’s own simulation Lockheed Martin’s own simulation applicationsapplications

Heinz Nixdorf Augmented Reality Heinz Nixdorf Augmented Reality applicationapplication

Lightning (Fraunhofer inst.)Lightning (Fraunhofer inst.)

Covise (Vircinity)Covise (Vircinity)

Modviz renderizerModviz renderizer

TGS AmiraTGS Amira

VR COM VD2VR COM VD2

Thales SpacemagicThales Spacemagic

Comet DodamComet Dodam

DaimlerChrysler: dbviewDaimlerChrysler: dbview

ICIDO: ido (in progress)ICIDO: ido (in progress)

... among a growing list ( see web site ... among a growing list ( see web site ))

25

More DVG featuresMore DVG features nVidia shaders:nVidia shaders:

Pixel lightingPixel lighting

Bump lightingBump lighting

Environmental mappingEnvironmental mapping

RefractionRefraction

Hardware based image post – processing:Hardware based image post – processing:

NVG / Flir “look”NVG / Flir “look”

Chromakey ( for augmented reality…)Chromakey ( for augmented reality…)

Multiple video insertions (mapped on a polygon)- OptionalMultiple video insertions (mapped on a polygon)- Optional

Instructor’s videoInstructor’s video

Cockpit monitorsCockpit monitors

Collaborative sessionCollaborative session

Photorealism

Simulations special effects

Photorealism

Simulations special effects

26

More DVG featuresMore DVG features

VISUALISATION CLUSTER MANAGEMENT (“VCM”):VISUALISATION CLUSTER MANAGEMENT (“VCM”):

Encompass all the HW and SW technlogies designed by Orad for Encompass all the HW and SW technlogies designed by Orad for visualisation cluster management :visualisation cluster management :

=> Multiple renderers can be shared by a group of users over a => Multiple renderers can be shared by a group of users over a network. Depending on the day’s use, ressources are allocated network. Depending on the day’s use, ressources are allocated across the network to each user ( eg : a 4 channel with 2 renderer across the network to each user ( eg : a 4 channel with 2 renderer on each becomes a 2 channel with 4 renderer on each )on each becomes a 2 channel with 4 renderer on each )

=> Cluster Permanent Availability “CPA” software : any broken => Cluster Permanent Availability “CPA” software : any broken renderer in a channel can automatically be by-passed so that for renderer in a channel can automatically be by-passed so that for instance the channel has lower AA but continues working instance the channel has lower AA but continues working

VCN is managed by the DVG service software which also Handles VCN is managed by the DVG service software which also Handles the reconfiguration of combining depending on application, as the reconfiguration of combining depending on application, as well as selection of formats. Manages VCN from a logical/SW well as selection of formats. Manages VCN from a logical/SW standpointstandpoint

27

DVG Key differenciatorsDVG Key differenciators

• Best price/performanceBest price/performance

• Easily and quickly upgradable with new generation boards = > “futureproof” Easily and quickly upgradable with new generation boards = > “futureproof” “board&vendor-independant” : the safest investment“board&vendor-independant” : the safest investment

• inexpensive support thanks to a majority of COTS components inexpensive support thanks to a majority of COTS components

•*The most experienced supplier in COTS scalable graphics with more than 3 years *The most experienced supplier in COTS scalable graphics with more than 3 years of R&D...of R&D...

• Compositing flexibility ( 8 modes and variants )according to system’s critical Compositing flexibility ( 8 modes and variants )according to system’s critical resource, teamed with the most advanced cluster management ( VCN )resource, teamed with the most advanced cluster management ( VCN )

• The highest OPENGL/DirectX compatibility ( unmodified drivers )The highest OPENGL/DirectX compatibility ( unmodified drivers )

• Adressing special issues related to combining : accurate data-lock, interface-Adressing special issues related to combining : accurate data-lock, interface-sync, “wait-for”…sync, “wait-for”…

•The largest installed base of PC based scalable graphics systems ( 500 units The largest installed base of PC based scalable graphics systems ( 500 units worldwide. Incl. Bae, Northrop Grumman, Telefonica, US navy research labs, worldwide. Incl. Bae, Northrop Grumman, Telefonica, US navy research labs, Peugeot... )Peugeot... )

•A cross-market supplier, amortizing R&D accross many verticals so each can A cross-market supplier, amortizing R&D accross many verticals so each can benefit of larger global revenues ( industry, research, military…) No vertically-benefit of larger global revenues ( industry, research, military…) No vertically-focused software development => Does not compete with its own clientsfocused software development => Does not compete with its own clients

28

DVG applicationsDVG applications

CAVEs, Workbench, Flight CAVEs, Workbench, Flight simulators…simulators…

Civil and military simulationsCivil and military simulations

Mission planningMission planning

Urban planningUrban planning

Car designCar design

Car driving simulationCar driving simulation

Interactive walk throughInteractive walk through

Theme parksTheme parks

Architectural designArchitectural design

Scientific/Medical visualizationScientific/Medical visualization

Collaborative EngineeringCollaborative Engineering

Museums, Planetariums and Museums, Planetariums and Cultural CentersCultural Centers

Hazard Perception / Disaster Hazard Perception / Disaster ManagementManagement

Oil & Gas explorationsOil & Gas explorations

Homeland SecurityHomeland Security

Augmented realityAugmented reality

29

High profile customersHigh profile customers

British AerospaceBritish Aerospace

Heinz Nixdorf Inst.Heinz Nixdorf Inst.

PeugeotPeugeot

Old Dominion UniversityOld Dominion University

Naval Research LabNaval Research Lab

Lockheed MartinLockheed Martin

Northrop GrumannNorthrop Grumann

Laing’O RourkeLaing’O Rourke

Chinese AFChinese AF

30

ANNEX : miscellaneous supporting ANNEX : miscellaneous supporting examplesexamples

31

DVG applications: Augmented RealityDVG applications: Augmented Reality(Cluster at Paderborn University)(Cluster at Paderborn University)

* Courtesy Paderborn University* Courtesy Paderborn University

32

DVG applications: Augmented RealityDVG applications: Augmented Reality

* Courtesy Paderborn University* Courtesy Paderborn University

33

Form-factors : DVG VR-X (1/2)Form-factors : DVG VR-X (1/2)

Screen shots from an integrationScreen shots from an integrationin HP’s XW8000 workstation, Orad’s in HP’s XW8000 workstation, Orad’s privileged partner for workstationsprivileged partner for workstations

34

Form-factors : DVG10 VR (2/2)Form-factors : DVG10 VR (2/2)

Orad DVG10 VR backpanel. Orad’s Orad DVG10 VR backpanel. Orad’s offering for integrated rack-mounted offering for integrated rack-mounted PCs.PCs.