COMPUTERBLADE WITH OFFSET NETWORK CARDS

Description

[Technical field]

[0001] The present invention relates to the field of supercomputers and more particularly concerns a liquid-cooled computing blade for a supercomputer as well as a supercomputer comprising at least one such computing blade. [State of the prior art]

[0002] An HPC (High Performance Computing) computer, or supercomputer, comprises a plurality of computing blades in the form of modules inserted into a bay. Each computing blade comprises a motherboard comprising a plurality of processors each configured to process data. Each computing blade also comprises high-speed network cards, for example of the Ethernet ®< mezzanine type (OCP3 card), to enable communication between the computing blades and the use of the power of several processors to execute the same task. To be functional, each of these network cards must be connected to the motherboard.

[0003] In air-cooled HPC computers, the classic connection system is card-to-card, i.e. the network card is guided in a mechanical chassis to the motherboard connector and insertion is done manually or using an insertion lever.

[0004] In liquid-cooled HPC computers, the motherboard is mounted on one side of the cold plate, through which heat sinks are mounted to cool the processors. However, such a configuration can make the insertion and card-to-card connection of network cards with the motherboard particularly complex, particularly due to the lack of space around the motherboard. In addition, each network card must have a thermal interface with the cold plate through which the coolant circulates in order to cool the said network card. The traditional card-to-card connection is therefore not satisfactory for this type of computing blade.

[0005] Document EP3500079A1 relates to a liquid cooling system for an electronic card that may be inefficient or unsuitable. Document US2022229798A1 describes an information processing system with a floating paddle card that may be inefficient or unsuitable.

[0006] There is therefore a need for a simple and effective solution to at least partially overcome these drawbacks. [Statement of the invention]

[0007] To this end, the invention firstly relates to a computing blade for a multi-computing blade supercomputer, said computing blade comprising a motherboard, comprising a plurality of processors mounted on a support, and a cold plate, sized to cover the support of the processors and comprising a plurality of heat sinks, a cooling circuit comprising channels inside which circulates a so-called "cold" heat transfer fluid, intended to supply the heat sinks, and an evacuation circuit comprising channels inside which circulates a so-called "hot" heat transfer fluid having heated up through the heat sinks, each heat sink being mounted in line with a processor in order to cool it by heat exchange, the computing blade comprising as many network cards as there are processors, each network card being associated with a processor,each network card comprising a plug-in end configured to receive a connector, each network card being configured to allow data communication between said processor and another processor of the computing blade or another computing blade, the computing blade being remarkable in that each network card is mounted on the cold plate while being connected to the motherboard via a connection system, said system comprising a first connector, connected to the plug-in end of the network card, and a second connector, connected on the one hand to the first connector via a cable and on the other hand to the motherboard.,

[0008] The liquid-cooled computing blade according to the invention allows the electrical connection of the network cards to the motherboard via the connection systems. The connection system of a network card thus allows said network card to be connected to the motherboard remotely via a cable so that the network cards can be remote from the motherboard while ensuring their cooling by the cold plate. The network cards can thus be arranged on the opposite side of the cold plate from the motherboard, in a location with sufficient space. The computing blade according to the invention therefore makes it possible to optimize the space available on both sides of the cold plate, in particular by making the placement of the network cards independent of the placement of the motherboard thanks to the remote connection system between the motherboard and the network cards.The cable and connector connection system also allows for easy connection and disconnection of network cards from the motherboard. The remote location of the network cards from the motherboard can also allow for easy replacement of a network card for maintenance purposes when the network cards are mounted in an easily accessible location on the compute blade.

[0009] Preferably, the cable of the connection system is flexible to facilitate positioning and connection of the first connector and the second connector.

[0010] More preferably, the second connector is connected to the motherboard through the cold plate in order to access the motherboard by the shortest path and thus optimize the space occupied by the blade and its components. Alternatively or in addition, all or part of the second connectors could be connected to the motherboard by cables bypassing the cold plate by one or more of its edges.

[0011] Advantageously, each network card is mounted on the cold plate by means of a support part, in particular so as to allow the use of standard network cards sold commercially.

[0012] In one embodiment, the support part comprises a body and a device for holding the first connector, mounted on said body, in order to maintain the connector in an optimized connected position and thus prevent it from moving during transport or during use of the computing blade.

[0013] Preferably, the holding device comprises a support blade of the first connector and a plate for holding the first connector on said support blade.

[0014] Advantageously, the support blade is made of metal, preferably aluminum, in order to withstand high temperatures to which the components internal to the computing blade may be subjected while being inexpensive.

[0015] Advantageously, the holding plate is made of metal, preferably steel, in order to withstand high temperatures to which the internal components of the calculation blade may be subjected while being inexpensive.

[0016] In one embodiment of the invention, the holding plate is fixed to the support blade using at least one screw, preferably using two screws, in order to effectively lock the first connector in the connected position in the holding device.

[0017] Preferably, the at least one screw is captive or unlosable.

[0018] In one embodiment, the at least one captive or captive screw is mounted at the retaining plate to prevent it from escaping during transport or attachment or removal of the first connector.

[0019] Preferably, the support blade is attached to the body of the support part by being offset by a distance relative to the cold plate in order to adapt the position of the first connector to the plug-in end of the network card.

[0020] Advantageously, the body of the support part is made of a thermal energy conducting material, preferably aluminum, in order to allow efficient cooling of the network card.

[0021] In one embodiment, each network card is attached to an edge of the computing blade, preferably the same edge, by a fixing end, opposite the plug-in end. The network cards can then be easily connected to a communication network via Ethernet ®< type cables.

[0022] One embodiment also relates to a server, in particular a supercomputer, comprising at least one computing blade as presented above. [Description of the drawings]

[0023] Other characteristics and advantages of the invention will become apparent from reading the description which follows. This description is purely illustrative and must be read in conjunction with the appended drawings in which: [ Fig 1 ] There Figure 1 schematically and transparently illustrates an embodiment of the supercomputer according to an embodiment of the invention. Fig 2 ] There Figure 2schematically illustrates a computing blade of the supercomputer of the Figure 1 . [ Fig 3 ] There Figure 3 schematically illustrates a network card mounted on a support piece of the blade of the Figure 2 . [ Fig 4 ] There Figure 4 schematically illustrates the support part and the network card of the Figure 2 , to which the first connector is connected in partial view. [Description of embodiments]

[0024] There Figure 1 illustrates an example of a supercomputer 1 according to an embodiment of the invention. Supercomputer 1

[0025] The supercomputer 1 comprises a bay 5 in which a plurality of computing blades 10 according to the invention are mounted. In this non-limiting example, the computing blades 10 are mounted by being vertically superimposed, for example on guide rails.

[0026] More generally, supercomputer 1 could be a server or any type of computer rack in which it is necessary to mount at least one liquid-cooled multiprocessor computing module. Calculation blade 10

[0027] There Figure 2 illustrates an example of a calculation blade 10 according to the invention.

[0028] The computing blade 10 firstly comprises a motherboard 110 and a cold plate 120. Motherboard 110

[0029] The motherboard 110 comprises a plurality of processors (not visible because they are under the cold plate 120) mounted on a support (also not visible because they are under the cold plate 120). Cold plate 120

[0030] The cold plate 120 is sized to cover the processor support and includes a plurality of heat sinks 122, a cooling circuit 124, an exhaust circuit 126, a so-called "cold" input connector 128 and a so-called "hot" output connector 129.

[0031] The cooling circuit 124 comprises channels inside which circulates a so-called “cold” heat transfer fluid, entering the cooling circuit 124 via the cold inlet connector 128 and which is intended to supply the heat sinks 122.

[0032] Each heat sink 122 is mounted opposite a processor to cool it by heat exchange.

[0033] The evacuation circuit 126 comprises channels inside which circulates a so-called “hot” heat transfer fluid having heated up through the heat sinks 122 and which is evacuated through the hot outlet connector 129.

[0034] The computing blade 10 then comprises a plurality of network cards 130. More precisely, the computing blade 10 comprises as many network cards 130 as there are processors, each network card 130 being associated with a processor.

[0035] Each network card 130 comprises a plug-in end 132 configured to receive a connector and is configured to enable data communication between the processor with which said network card 130 is associated and another processor of the computing blade 10 or of another computing blade 10 (of the supercomputer 1 or of another supercomputer) via said plug-in end 132.

[0036] Each network card 130 is mounted on the cold plate 120 while being connected to the motherboard 110 via a connection system 140.

[0037] The connection system 140 comprises a first connector 142, a second connector 144 and a cable (or a cable harness) 146 electrically connecting the first connector 142 and the second connector 144.

[0038] The first connector 142 is connected to the plug-in end 132 of the network card 130. The second connector 144 is connected on the one hand to the first connector 142 via the cable 146 and on the other hand to the motherboard 110.

[0039] In the example of the Figure 2 , the second connector 144 is connected to the motherboard 110 through the cold plate 120 in order to access the motherboard 110 by a short path and thus optimize the space occupied by the computing blade 10 and its components.

[0040] The cable 146 is flexible so as to be able to conform to the space formed between the heat sinks 122 and to the connection direction of both the first connector 142 (here parallel to the plane along which the cold plate 120 extends) and the second connector 144 (here orthogonal to the plane along which the cold plate 120 extends).

[0041] Each network card 130 is mounted on the cold plate 120 via a support piece 150, as illustrated in the Figure 3 .

[0042] The support part 150 comprises a body 152 and a holding device 154 mounted on said body 152 and configured to hold the first connector 142 when said first connector 142 is connected to the plug-in end 132 in order to prevent the first connector 142 from moving during transport or use of the computing blade 10. In other words, the holding device 154 makes it possible to hold the connector in the connected position on the plug-in end 132 in an optimized manner.

[0043] In reference to the Figure 4 , the holding device 154 comprises a support blade 154L and a holding plate 154P. The support blade 154L receives the first connector 142 while the holding plate 154P blocks the first connector 142 on said support blade 154L.

[0044] Preferably, the support blade 154L is made of metal, preferably aluminum to conduct thermal energy. More preferably, the holding plate 154P is also made of metal, preferably steel.

[0045] Still referring to the Figure 4 , the holding plate 154P is fixed on the support blade 154L using two screws 156 in order to block the first connector 142.

[0046] Each screw 156 is mounted at the holding plate 154P in a captive or captive manner, i.e. said screw is free to rotate to screw onto the support blade 154L but is retained on the holding plate 154P so as to prevent it from escaping during transport and the attachment or removal of the first connector 142.

[0047] The support blade 154L is fixed to the body 152 of the support part 150 and extends parallel to the cold plate 120 while being offset by a distance relative to said cold plate 120.

[0048] The body 152 of the support part 150 is made of a thermal energy conducting material, for example aluminum, in order to allow the cooling of the network card 130 by the thermal energy provided by the heat transfer fluid circulating in the cold plate 120.

[0049] In reference to the Figure 2 , each network card 130 is fixed on an edge of the computing blade 10 by its fixing end, opposite the plug-in end 132. With reference to the Figures 3 and 4 , a metal fixing element 160 makes it possible to fix the network card 130 on the edge of the computing blade 10, for example by screwing or locking.

[0050] The invention makes it possible to optimize the distribution of the components on either side of the cold plate 120, in particular to place the network cards 130 in the most suitable location, in particular on the opposite face of the cold plate 120 relative to the motherboard 110. The invention therefore makes it possible to obtain a computing blade 10 that is compact and easy to maintain. In particular, different types and models of network card can be used since they are no longer constrained by the position of the motherboard in a card-to-card insertion configuration.

Claims

1. Computing blade (10) for a multi-blade supercomputer (1), the computing blade (10) comprising a motherboard (110), which comprises a plurality of processors mounted on a support, and a cold plate (120) sized to cover the support of the processors and comprising a plurality of heat sinks (122), a cooling circuit (124) comprising channels inside which a so-called "cold" heat transfer fluid flows, which is intended to supply the heat sinks (122), and an evacuation circuit (126) comprising channels through which a "hot" heat transfer fluid flows that has warmed up through the heat sinks (122), each heat sink (122) being mounted in line with a processor in order to cool it by heat exchange, the computing blade (10) comprising as many network cards (130) as processors, each network card (130) being associated with a processor, each network card (130) comprising a plug-in end (132) configured to receive a connector (142), each network card (130) being configured to enable data communication between the processor and another processor, the computing blade (10) being characterised in that each network card (130) is mounted on the cold plate (120) and is connected to the motherboard (110) via a connection system (140), the system (140) comprising a first connector (142) which is connected to the plug-in end (132) of the network card (130), and a second connector (144) which is connected on the one hand to the first connector (142) via a cable (146) and on the other hand to the motherboard (110).

2. Computing blade (10) according to claim 1, wherein the second connector (144) is connected to the motherboard (110) via the cold plate (120).

3. Computing blade (10) according to either of the preceding claims, wherein each network card (130) is mounted on the cold plate (120) via a support part (150).

4. Computing blade (10) according to the preceding claim, wherein the support part (150) comprises a body (152) and a holding device (154) for the first connector (142), mounted on the body (152).

5. Computing blade (10) according to the preceding claim, wherein the holding device (154) comprises a support blade (154L) for the first connector (142) and a holding plate (154P) for the first connector (142) on the support blade (154L).

6. Computing blade (10) according to the preceding claim, wherein the support blade (154L) is made of metal, preferably aluminium.

7. Computing blade (10) according to any of claims 5 or 6, wherein the holding plate (154P) is made of metal, preferably steel.

8. Computing blade (10) according to any of claims 5 to 7, wherein the holding plate (154P) is attached to the support blade (154L) by means of at least one screw (156), preferably by means of two screws (156).

9. Computing blade (10) according to the preceding claim, wherein the at least one screw (156) is captive, and mounted on the holding plate (154P).

10. Computing blade (10) according to any of claims 5 to 9, wherein the support blade (154L) is attached to the body (152) of the support part (150) at a distance from the cold plate (120).

11. Computing blade (10) according to any of claims 3 to 10, wherein the body (152) of the support part (150) is made of a thermally conductive material.

12. Computing blade (10) according to the preceding claim, wherein the body (152) of the support part (150) is made of aluminium.

13. Computing blade (10) according to any of the preceding claims, wherein the cable (146) of the connection system (140) is flexible.

14. Computing blade (10) according to any of the preceding claims, wherein each network card (130) is attached to an edge of the computing blade (10) by an attachment end, opposite the plug-in end (132).

15. Supercomputer (1) comprising at least one computing blade (10) according to any of the preceding claims.