Description & Topology

The Monolithe cluster is composed of:

• A frontend named the "Monolithe" (monolithe.soc.lip6.fr).
• Multiple compute nodes (see the table below).

The frontend is a DELL OptiPlex 7070 SFF computer based on an Intel Core i7-9700 CPU with 32 GB of RAM. Thus, it is a good candidate for heavy compilations. It is located in the LIP6 132.227.102.0/24 local network and it has a full access to Internet.

Info

Here is the LIP6 Mattermost channel dedicated to the news about the Monolithe. Don't hesitate to join.

Summary Table

Here is a table that summarizes the different compute nodes available in the cluster:

Node Name	Vendor	Board no	Slurm Partition	Release Date	DNS Name	IP Address	MAC Address	SoC Name	SoC Process	CPU	CPU ISA	GPU	RAM	TDP	Idle Power	OS	Kernel Version	Mattermost Channel
Odroid-XU4	Hardkernel	5	xu4	Jul'15	vroum.soc.lip6.fr	132.227.71.61	00:1e:06:30:03:e1	Samsung Exynos-5422	28 nm	4 x ARM Cortex-A15 + 4 x ARM Cortex-A7	ARMv7-A	ARM Mali-T628 MP6	2 GB	~13 W	2.6 W	Linux Ubuntu 22.04.2 LTS	5.04.228	Link
Raspberry Pi 3 Model B	Fondation Raspberry Pi	-	rpi3	Feb'16	rpi3.soc.lip6.fr	132.227.71.40	b8:27:eb:08:1d:51	Broadcom BCM2837	40 nm	4 x ARM Cortex-A53	ARMv8.0-A	VideoCore IV	1 GB	-	-	Linux Ubuntu 22.04.4 LTS	5.15.000	Link
Jetson TX2	Nvidia	3	tx2	Mar'17	tegrax2c.soc.lip6.fr	132.227.71.22	00:04:4b:8d:21:9c	Nvidia Tegra186	16 nm	2 x Nvidia Denver 2 + 4 x ARM Cortex-A57	ARMv8.0-A	Nvidia Pascal / 2 SMX, 256 cores	8 GB	15 W	2.0 W	Linux Ubuntu 18.04.6 LTS	4.09.140	Link
Jetson AGX Xavier	Nvidia	1	xagx	Oct'18	tegraagx.soc.lip6.fr	132.227.71.55	00:04:4b:cb:a8:c8	Nvidia Tegra194	12 nm	8 x Nvidia Carmel	ARMv8.2-A	Nvidia Volta / 8 SMX, 512 cores	16 GB	30 W		Linux Ubuntu 18.04.6 LTS	4.09.201	Link
OptiPlex 7070 SFF	DELL	-	brub	Oct'18	brubeck.soc.lip6.fr	132.227.102.91	e4:54:e8:70:18:eb	Intel Core i5-9500	14 nm	6 x Intel Coffee Lake (Core i5-9500)	x86_64	Intel UHD Graphics 630	16 GB	65 W	-	Linux Ubuntu 18.04.6 LTS	5.04.000	-
Jetson Xavier Nano	Nvidia	2	xnano	Mar'19	jetson-nano1.soc.lip6.fr	132.227.71.57	00:e0:4c:68:01:e9	Nvidia Tegra210 (TM660M-A2)	20 nm	4 x ARM Cortex-A57	ARMv8.0-A	Nvidia Maxwell / 1 SMX, 128 cores	4 GB	10 W	2.5 W	Linux Ubuntu 18.04.6 LTS	4.09.201	Link
Raspberry Pi 4 Model B	Fondation Raspberry Pi	6	rpi4	Jun'19	selfix.soc.lip6.fr	132.227.71.62	dc:a6:32:d9:57:21	Broadcom BCM2711	28 nm	4 x ARM Cortex-A72	ARMv8.0-A	VideoCore VI	8 GB	~7 W	3.3 W	Linux Ubuntu 20.04.6 LTS	5.04.000	Link
Jetson Xavier NX	Nvidia	4	xnx	May'20	tegranx-1.soc.lip6.fr	132.227.71.38	48:b0:2d:2f:70:ed	Nvidia Tegra194 (TE860M-A2)	12 nm	6 x Nvidia Carmel	ARMv8.2-A	Nvidia Volta / 6 SMX, 384 cores	8 GB	15 W		Linux Ubuntu 18.04.6 LTS	4.09.201	Link
Mac Studio M1 Ultra	Apple	-	m1u	Mar'22	m1ultra.soc.lip6.fr	132.227.102.98	9c:76:0e:2e:4c:c2	Apple Silicon M1 Ultra	5 nm	16 x Apple Firestorm + 4 x Apple Icestorm	ARMv8.5-A	Apple 48 cores	64 GB	60 W		Linux Fedora Asahi rel. 39	6.06.003	Link
Jetson Orin NX	Nvidia	-	onx	Feb'23	orinnx.soc.lip6.fr	132.227.71.47	48:b0:2d:d3:d2:80	Nvidia Tegra234	8 nm	8 x ARM Cortex-A78AE	ARMv8.2-A	Nvidia Ampere / 8 SMX, 1024 cores	16 GB	25 W		Linux Ubuntu 20.04.6 LTS	5.10.120	Link
Jetson AGX Orin	Nvidia	8	oagx	Mar'23	orinagx.soc.lip6.fr	132.227.71.46	48:b0:2d:e1:08:dd	Nvidia Tegra234	8 nm	12 x ARM Cortex-A78AE	ARMv8.2-A	Nvidia Ampere / 16 SMX, 2048 cores	64 GB	60 W	9.3 W	Linux Ubuntu 20.04.6 LTS	5.10.120	Link
Jetson Orin Nano	Nvidia	9	onano	Mar'23	orinnano.soc.lip6.fr	132.227.71.23	48:b0:2d:ec:1a:1d	Nvidia Tegra234	8 nm	6 x ARM Cortex-A78AE	ARMv8.2-A	Nvidia Ampere / 8 SMX, 1024 cores	8 GB	15 W		Linux Ubuntu 20.04.6 LTS	5.10.192	Link
Orange Pi 5 Plus	Orange Pi	7	opi5	Jul'23	orangepi5.soc.lip6.fr	132.227.71.60	c0:74:2b:ff:46:15	Rockchip RK3588	8 nm	4 x ARM Cortex-A76 + 4 x ARM Cortex-A55	ARMv8.2-A	ARM Mali-G610 MP4	16 GB	~15 W	2.4 W	Linux Ubuntu 22.04.4 LTS	5.10.160	Link
Raspberry Pi 5 Model B	Fondation Raspberry Pi	-	rpi5	Oct'23	rpi5.soc.lip6.fr	132.227.71.39	2c:cf:67:1f:cf:9a	Broadcom BCM2712	16 nm	4 x ARM Cortex-A76	ARMv8.2-A	VideoCore VII	8 GB	-	-	Linux Ubuntu 23.10	6.05.000	Link
Mercury EM780	Minisforum	-	em780	Jan'24	em780.soc.lip6.fr	132.227.71.63	c8:4d:44:22:a9:42	AMD Ryzen 7 7840U	4 nm	8 x AMD Zen4 ULV	x86_64	AMD Radeon 780M / 12 SMX, 768 cores	32 GB	~50 W	-	Linux Ubuntu 24.04 LTS	6.08.000	Link

Info

The nodes are in the 132.227.71.0/25 local network (from 132.227.71.1 to 132.227.71.126 IP addresses) and they do not have a direct Internet access. However, there is an HTTP/HTTPS proxy. Its DNS name is asim.lip6.fr and the listening port is 3128.

Warning

There are two exceptions in the nodes. Indeed, brub and m1u IP addresses are in the 132.227.102.0/24 local network.

Nodes

Odroid-XU4

Caches

• Cortex-A7
  • L1D private to core: 32 KB
  • L1I private to core: 32 KB
  • L2 shared between the 4 LITTLE cores: 512 KB
• Cortex-A15
  • L1D private to core: 32 KB
  • L1I private to core: 32 KB
  • L2 shared between the 4 big cores: 2 MB

Topology

hwloc: Topology of the Odroid-XU4.

Memory Bandwidth

Cortex-A7, 1 threadCortex-A7, 4 threadsCortex-A15, 1 threadCortex-A15, 4 threadsMixed, 8 threads

Odroid-XU4 bandwidth: Cortex-A7, 1 thread (raw CSV file).

Odroid-XU4 bandwidth: Cortex-A7, 4 threads (raw CSV file).

Odroid-XU4 bandwidth: Cortex-A15, 1 thread (raw CSV file).

Odroid-XU4 bandwidth: Cortex-A15, 4 threads (raw CSV file).

Odroid-XU4 bandwidth: Mixed, 8 threads (raw CSV file).

Raspberry Pi 3 Model B

Caches

• Cortex-A53
  • L1D private to core: 16 KB
  • L1I private to core: 16 KB
  • L2 shared between the 4 cores: 512 KB

Topology

hwloc: Topology of the Raspberry Pi 3 (be careful L1 sizes are wrong).

Memory Bandwidth

Cortex-A53, 1 threadCortex-A53, 2 threadsCortex-A53, 3 threadsCortex-A53, 4 threads

Raspberry Pi 3 bandwidth: Cortex-A53, 1 thread (raw CSV file).

Raspberry Pi 3 bandwidth: Cortex-A53, 2 threads (raw CSV file).

Raspberry Pi 3 bandwidth: Cortex-A53, 3 threads (raw CSV file).

Raspberry Pi 3 bandwidth: Cortex-A53, 4 threads (raw CSV file).

Jetson TX2

Caches

• Cortex-A57
  • L1D private to core: 32 KB
  • L1I private to core: 48 KB
  • L2 shared between the 4 A57 cores: 2 MB
• Denver
  • L1D private to core: 64 KB
  • L1I private to core: 128 KB
  • L2 shared between the 2 Denver cores: 2 MB

Topology

hwloc: Topology of the Jetson TX2.

Memory Bandwidth

Cortex-A57, 1 threadCortex-A57, 4 threadsDenver, 1 threadDenver, 2 threadsMixed, 6 threads

Jetson TX2 bandwidth: Cortex-A57, 1 thread (raw CSV file).

Jetson TX2 bandwidth: Cortex-A57, 4 threads (raw CSV file).

Jetson TX2 bandwidth: Denver, 1 thread (raw CSV file).

Jetson TX2 bandwidth: Denver, 2 threads (raw CSV file).

Jetson TX2 bandwidth: Mixed, 6 threads (raw CSV file).

Jetson AGX Xavier

Caches

• Carmel
  • L1D private to core: 64 KB
  • L1I private to core: 128 KB
  • L2 shared by cluster of 2 Carmel cores: 2 MB
  • L3 shared by all the cores: 4 MB

Topology

hwloc: Topology of the Jetson AGX Xavier.

Memory Bandwidth

Carmel, 1 threadCarmel, 2 threadsCarmel, 4 threadsCarmel, 6 threadsCarmel, 8 threads

Jetson AGX Xavier bandwidth: Carmel, 1 thread (raw CSV file).

Jetson AGX Xavier bandwidth: Carmel, 2 threads (raw CSV file).

Jetson AGX Xavier bandwidth: Carmel, 4 threads (raw CSV file).

Jetson AGX Xavier bandwidth: Carmel, 6 threads (raw CSV file).

Jetson AGX Xavier bandwidth: Carmel, 8 threads (raw CSV file).

OptiPlex 7070 SFF (Brubeck)

Topology

hwloc: Topology of Brubeck.

Jetson Xavier Nano

Caches

• Cortex-A57
  • L1D private to core: 32 KB
  • L1I private to core: 48 KB
  • L2 shared by the 4 cores: 2 MB

Topology

hwloc: Topology of the Jetson Xavier Nano.

Memory Bandwidth

Cortex-A57, 1 threadCortex-A57, 4 threads

Jetson Xavier Nano bandwidth: Cortex-A57, 1 thread (raw CSV file).

Jetson Xavier Nano bandwidth: Cortex-A57, 4 threads (raw CSV file).

Raspberry Pi 4 Model B

Caches

• Cortex-A72
  • L1D private to core: 32 KB
  • L1I private to core: 48 KB
  • L2 shared by the 4 cores: 1 MB

Topology

hwloc: Topology of the Raspberry Pi 4.

Memory Bandwidth

Cortex-A72, 1 threadCortex-A72, 2 threadsCortex-A72, 3 threadsCortex-A72, 4 threads

Raspberry Pi 4 bandwidth: Cortex-A72, 1 thread (raw CSV file).

Raspberry Pi 4 bandwidth: Cortex-A72, 2 threads (raw CSV file).

Raspberry Pi 4 bandwidth: Cortex-A72, 3 threads (raw CSV file).

Raspberry Pi 4 bandwidth: Cortex-A72, 4 threads (raw CSV file).

Jetson Xavier NX

Caches

• Carmel
  • L1D private to core: 64 KB
  • L1I private to core: 128 KB
  • L2 shared by cluster of 2 Carmel cores: 2 MB
  • L3 shared by all the cores: 4 MB

Topology

hwloc: Topology of the Jetson Xavier NX.

Memory Bandwidth

Carmel, 1 threadCarmel, 2 threadsCarmel, 4 threadsCarmel, 6 threads

Jetson Xavier NX bandwidth: Carmel, 1 thread (raw CSV file).

Jetson Xavier NX bandwidth: Carmel, 2 threads (raw CSV file).

Jetson Xavier NX bandwidth: Carmel, 4 threads (raw CSV file).

Jetson Xavier NX bandwidth: Carmel, 6 threads (raw CSV file).

Mac Studio M1 Ultra

Caches

• Icestorm

  • L1D private to core: 64 KB
  • L1I private to core: 128 KB
  • L2 shared by cluster of 2 Icestorm cores: 4 MB
  • L3 shared by all the cores: 96 MB

• Firestorm

  • L1D private to core: 128 KB
  • L1I private to core: 192 KB
  • L2 shared by cluster of 4 Firestorm cores: 12 MB
  • L3 shared by all the cores: 96 MB

Topology

hwloc: Topology of the M1 Ultra (be carefull this diagram is a little bit wrong because the L3 cache is not represented).

Memory Bandwidth on macOS

Icestorm, 1 threadIcestorm, 2 threadsIcestorm, 4 threadsFirestorm, 1 threadFirestorm, 4 threadsFirestorm, 8 threadsFirestorm, 16 threadsMixed, 20 threads

Mac Studio M1 Ultra bandwidth: Icestorm, 1 thread (raw CSV file).

Mac Studio M1 Ultra bandwidth: Icestorm, 2 threads (raw CSV file).

Mac Studio M1 Ultra bandwidth: Icestorm, 4 threads (raw CSV file).

Mac Studio M1 Ultra bandwidth: Firestorm, 1 thread (raw CSV file).

Mac Studio M1 Ultra bandwidth: Firestorm, 4 threads (raw CSV file).

Mac Studio M1 Ultra bandwidth: Firestorm, 8 threads (raw CSV file).

Mac Studio M1 Ultra bandwidth: Firestorm, 16 threads (raw CSV file).

Mac Studio M1 Ultra bandwidth: Mixed, 20 threads (raw CSV file).

Memory Bandwidth on Linux

Icestorm, 1 threadIcestorm, 2 threadsIcestorm, 4 threadsFirestorm, 1 threadFirestorm, 4 threadsFirestorm, 8 threadsFirestorm, 16 threadsMixed, 20 threads

Mac Studio M1 Ultra bandwidth: Icestorm, 1 thread (raw CSV file).

Mac Studio M1 Ultra bandwidth: Icestorm, 2 threads (raw CSV file).

Mac Studio M1 Ultra bandwidth: Icestorm, 4 threads (raw CSV file).

Mac Studio M1 Ultra bandwidth: Firestorm, 1 thread (raw CSV file).

Mac Studio M1 Ultra bandwidth: Firestorm, 4 threads (raw CSV file).

Mac Studio M1 Ultra bandwidth: Firestorm, 8 threads (raw CSV file).

Mac Studio M1 Ultra bandwidth: Firestorm, 16 threads (raw CSV file).

Mac Studio M1 Ultra bandwidth: Mixed, 20 threads (raw CSV file).

Jetson Orin NX

Caches

• Cortex-A78AE
  • L1D private to core: 64 KB
  • L1I private to core: 64 KB
  • L2 private to core: 256 KB
  • L3 shared by cluster of 4 cores: 2 MB

Topology

hwloc: Topology of the Jetson Orin NX.

Memory Bandwidth

Cortex-A78AE, 1 threadCortex-A78AE, 2 threadsCortex-A78AE, 4 threadsCortex-A78AE, 8 threads

Jetson Orin NX bandwidth: Cortex-A78AE, 1 thread (raw CSV file).

Jetson Orin NX bandwidth: Cortex-A78AE, 2 threads (raw CSV file).

Jetson Orin NX bandwidth: Cortex-A78AE, 4 threads (raw CSV file).

Jetson Orin NX bandwidth: Cortex-A78AE, 8 threads (raw CSV file).

Jetson AGX Orin

Caches

• Cortex-A78AE
  • L1D private to core: 64 KB
  • L1I private to core: 64 KB
  • L2 private to core: 256 KB
  • L3 shared by cluster of 4 cores: 2 MB

Topology

hwloc: Topology of the Jetson AGX Orin.

Memory Bandwidth

Cortex-A78AE, 1 threadCortex-A78AE, 2 threadsCortex-A78AE, 4 threadsCortex-A78AE, 8 threadsCortex-A78AE, 12 threads

Jetson AGX Orin bandwidth: Cortex-A78AE, 1 thread (raw CSV file).

Jetson AGX Orin bandwidth: Cortex-A78AE, 2 threads (raw CSV file).

Jetson AGX Orin bandwidth: Cortex-A78AE, 4 threads (raw CSV file).

Jetson AGX Orin bandwidth: Cortex-A78AE, 8 threads (raw CSV file).

Jetson AGX Orin bandwidth: Cortex-A78AE, 12 threads (raw CSV file).

Jetson Orin Nano

Caches

• Cortex-A78AE
  • L1D private to core: 64 KB
  • L1I private to core: 64 KB
  • L2 private to core: 256 KB
  • L3 shared by cluster of 4 cores: 2 MB

Topology

hwloc: Topology of the Jetson Orin Nano.

Memory Bandwidth

Cortex-A78AE, 1 threadCortex-A78AE, 2 threadsCortex-A78AE, 4 threadsCortex-A78AE, 6 threads

Jetson Orin Nano bandwidth: Cortex-A78AE, 1 thread (raw CSV file).

Jetson Orin Nano bandwidth: Cortex-A78AE, 2 threads (raw CSV file).

Jetson Orin Nano bandwidth: Cortex-A78AE, 4 threads (raw CSV file).

Jetson Orin Nano bandwidth: Cortex-A78AE, 6 threads (raw CSV file).

Orange Pi 5 Plus

Caches

• Cortex-A55
  • L1D private to core: 32 KB
  • L1I private to core: 32 KB
  • L2 private to core: 128 KB
• Cortex-A76
  • L1D private to core: 64 KB
  • L1I private to core: 64 KB
  • L2 private to core: 512 KB
• L3 shared by all the cores: 3 MB

Topology

hwloc: Topology of the Orange Pi 5 Plus.

Memory Bandwidth

Cortex-A55, 1 threadCortex-A55, 4 threadsCortex-A76, 1 threadCortex-A76, 4 threadsMixed, 8 threads

Orange Pi 5 Plus bandwidth: Cortex-A55, 1 thread (raw CSV file).

Orange Pi 5 Plus bandwidth: Cortex-A55, 4 threads (raw CSV file).

Orange Pi 5 Plus bandwidth: Cortex-A76, 1 thread (raw CSV file).

Orange Pi 5 Plus bandwidth: Cortex-A76, 4 threads (raw CSV file).

Orange Pi 5 Plus bandwidth: Mixed, 8 threads (raw CSV file).

Raspberry Pi 5 Model B

Caches

• Cortex-A76
  • L1D private to core: 64 KB (guessed size)
  • L1I private to core: 64 KB (guessed size)
  • L2 private to core: 512 KB
  • L3 shared by all the cores: 2 MB

Topology

hwloc: Topology of the Raspberry Pi 5 (be careful, the caches organization is partially wrong on this diagram).

Memory Bandwidth

Cortex-A76, 1 threadCortex-A76, 2 threadsCortex-A76, 3 threadsCortex-A76, 4 threads

Raspberry Pi 5 bandwidth: Cortex-A76, 1 thread (raw CSV file).

Raspberry Pi 5 bandwidth: Cortex-A76, 2 threads (raw CSV file).

Raspberry Pi 5 bandwidth: Cortex-A76, 3 threads (raw CSV file).

Raspberry Pi 5 bandwidth: Cortex-A76, 4 threads (raw CSV file).

Mercury EM780

Caches

• Ryzen 7 7840U
  • L1D private to core: 32 KB
  • L1I private to core: 32 KB
  • L2 private to core: 1 MB
  • L3 shared by all the cores: 16 MB

Topology

hwloc: Topology of the Mercury EM780.

Memory Bandwidth

Ryzen 7 7840U, 1 threadRyzen 7 7840U, 2 threadsRyzen 7 7840U, 3 threadsRyzen 7 7840U, 4 threadsRyzen 7 7840U, 5 threadsRyzen 7 7840U, 6 threadsRyzen 7 7840U, 7 threadsRyzen 7 7840U, 8 threadsRyzen 7 7840U, 16 threads

Mercury EM780 bandwidth: 7840U, 1 thread (raw CSV file).

Mercury EM780 bandwidth: 7840U, 2 threads (raw CSV file).

Mercury EM780 bandwidth: 7840U, 3 threads (raw CSV file).

Mercury EM780 bandwidth: 7840U, 4 threads (raw CSV file).

Mercury EM780 bandwidth: 7840U, 5 threads (raw CSV file).

Mercury EM780 bandwidth: 7840U, 6 threads (raw CSV file).

Mercury EM780 bandwidth: 7840U, 7 threads (raw CSV file).

Mercury EM780 bandwidth: 7840U, 8 threads (raw CSV file).

Mercury EM780 bandwidth: 7840U, 16 threads (raw CSV file).

Miscellaneous

Hardware Counters

On all the SBCs, the system perf counters have been enabled for non-root users. Thereafter is the applied configuration:

sudo sysctl -w kernel.perf_event_paranoid=2
# the following command makes the modification permanent after reboot
echo "kernel.perf_event_paranoid=2" | sudo tee --append /etc/sysctl.d/99-sysctl.conf