Nvidia opens up speedy NVLink interconnect to custom CPUs, ASICs
Nvidia opens up speedy NVLink interconnect to custom CPUs, ASICs
One of the two just needs to be made by Nv
Computex Nvidia has opened the NVLink interconnect tech used to stitch its rack-scale compute platforms together to the broader ecosystem with the introduction of NVLink Fusion at Computex this week.…
dlvr.it
May 19, 2025 at 6:03 AM
Nvidia opens up speedy NVLink interconnect to custom CPUs, ASICs
You all wanted to see and SLI setup on the X99 rig, right?
Am I doing this right?
Am I doing this right?
August 15, 2026 at 12:21 AM
You all wanted to see and SLI setup on the X99 rig, right?
Am I doing this right?
Am I doing this right?
Nvidia has opened the NVLink interconnect to the broader ecosystem with the introduction of NVLink Fusion at Computex
nvidianews.nvidia.com/news/nvidia-...
#HPC #AI via @tobiasmann.bsky.social
nvidianews.nvidia.com/news/nvidia-...
#HPC #AI via @tobiasmann.bsky.social
NVIDIA Unveils NVLink Fusion for Industry to Build Semi-Custom AI Infrastructure With NVIDIA Partner Ecosystem
NVIDIA today unveiled NVIDIA NVLink Fusion™ — new silicon that lets industries build semi-custom AI infrastructure with the vast ecosystem of partners building with NVIDIA NVLink™, the world’s most ad...
nvidianews.nvidia.com
May 19, 2025 at 10:45 AM
Nvidia has opened the NVLink interconnect to the broader ecosystem with the introduction of NVLink Fusion at Computex
nvidianews.nvidia.com/news/nvidia-...
#HPC #AI via @tobiasmann.bsky.social
nvidianews.nvidia.com/news/nvidia-...
#HPC #AI via @tobiasmann.bsky.social
of course NVLink is basically the same thing, but at least you can pretend that it’s a Special Graphics Card Thing, which *feels* less perverse than Literally Just Ethernet Lol
November 19, 2024 at 12:42 AM
of course NVLink is basically the same thing, but at least you can pretend that it’s a Special Graphics Card Thing, which *feels* less perverse than Literally Just Ethernet Lol
What they didn't tell you is that single GPU is an NVLink fabric the size of a rugby field.
April 9, 2026 at 5:15 PM
What they didn't tell you is that single GPU is an NVLink fabric the size of a rugby field.
i‘m going to need you to step away from the nvlink and handover the spine leaf
June 27, 2026 at 7:57 PM
i‘m going to need you to step away from the nvlink and handover the spine leaf
AI networking startups race to replace Nvidia's NVLink
AI networking startups race to replace Nvidia's NVLink
Intel spin-off Cornelis and newcomer Delos Data pitch open alternatives for scaling AI beyond the rack
www.theregister.com
September 15, 2026 at 8:58 PM
AI networking startups race to replace Nvidia's NVLink
NVIDIA’s Kyber rack combines NVLink switch ASICs into what used to be the cable backplane. It’s physically huge—height of the whole rack. I wonder what the FRU is on this monstrosity.
#GTC26
#GTC26
March 18, 2026 at 5:46 AM
NVIDIA’s Kyber rack combines NVLink switch ASICs into what used to be the cable backplane. It’s physically huge—height of the whole rack. I wonder what the FRU is on this monstrosity.
#GTC26
#GTC26
Apple joins UALink, an industry group working on a next generation AI accelerator architecture standard meant to challenge Nvidia's dominance with NVLink (Zac Hall/9to5Mac)
Main Link | Techmeme Permalink
Main Link | Techmeme Permalink
January 14, 2025 at 8:18 PM
Apple joins UALink, an industry group working on a next generation AI accelerator architecture standard meant to challenge Nvidia's dominance with NVLink (Zac Hall/9to5Mac)
Main Link | Techmeme Permalink
Main Link | Techmeme Permalink
my actual best gpu option right now is probably a second 3090 and a nvlink bridge
October 17, 2025 at 8:20 PM
my actual best gpu option right now is probably a second 3090 and a nvlink bridge
This is completely correct other than the infrastructure not existing. I think people need to realize WHY this is happening is because, for example, NVIDIA sells GB200 NVL72s (racks of 72 GPUs) with *17TB of RAM* and likely over 500TB of storage.
February 17, 2026 at 8:42 PM
This is completely correct other than the infrastructure not existing. I think people need to realize WHY this is happening is because, for example, NVIDIA sells GB200 NVL72s (racks of 72 GPUs) with *17TB of RAM* and likely over 500TB of storage.
Bislang nutzte Nvidia die schnelle NVLink-Schnittstelle exklusiv für eigene Rechenbeschleuniger. In Zukunft dürfen sie auch andere Firmen integrieren. #Nvidia
NVLink Fusion: Nvidia öffnet die NVLink-Schnittstelle für andere Chiphersteller
Bislang nutzte Nvidia die schnelle NVLink-Schnittstelle exklusiv für eigene Rechenbeschleuniger. In Zukunft dürfen sie auch andere Firmen integrieren.
www.heise.de
May 19, 2025 at 6:17 AM
Bislang nutzte Nvidia die schnelle NVLink-Schnittstelle exklusiv für eigene Rechenbeschleuniger. In Zukunft dürfen sie auch andere Firmen integrieren. #Nvidia
For those who are not aware of these details with NVLink Fusion:
For custom XPU configurations, the interface to NVLink utilizes integration of Universal Chiplet Interconnect Express (UCIe) IP and interface. NVIDIA provides the bridge / NVLink chiplet for UCIe to NVLink.
For custom XPU configurations, the interface to NVLink utilizes integration of Universal Chiplet Interconnect Express (UCIe) IP and interface. NVIDIA provides the bridge / NVLink chiplet for UCIe to NVLink.
August 22, 2025 at 4:04 PM
For those who are not aware of these details with NVLink Fusion:
For custom XPU configurations, the interface to NVLink utilizes integration of Universal Chiplet Interconnect Express (UCIe) IP and interface. NVIDIA provides the bridge / NVLink chiplet for UCIe to NVLink.
For custom XPU configurations, the interface to NVLink utilizes integration of Universal Chiplet Interconnect Express (UCIe) IP and interface. NVIDIA provides the bridge / NVLink chiplet for UCIe to NVLink.
Are we finally read to scale the "copper wall"?
Nvidia’s NVLink Fusion Quietly Pushes Optics Inside The Rack
NVLink Fusion hints that optics may be moving inside the rack.
spectrum.ieee.org
July 9, 2026 at 1:01 PM
Are we finally read to scale the "copper wall"?
Apple is reportedly considering a return to the server market it left when it abandoned Xserve, but this time incorporating the future M8 Ultra chip, fully integrated with NVLink Fusion technology.
Apple's Xserve may return with Nvidia tech
Apple is reportedly considering a return to the server market it left when it abandoned Xserve, but this time incorporating the future M8 Ultra chip, fully integrated with NVLink Fusion technology.
appleinsider.com
September 16, 2026 at 1:14 PM
Apple is reportedly considering a return to the server market it left when it abandoned Xserve, but this time incorporating the future M8 Ultra chip, fully integrated with NVLink Fusion technology.
"Don't be Evil" seems to have been replaced with "Don't be Functional."
July 23, 2026 at 11:45 PM
"Don't be Evil" seems to have been replaced with "Don't be Functional."
NVLinkなしのTesla V100 32GB 2枚でQwen3.8-Flash-Nextを動かした | toyoshi #zenn
https://zenn.dev/phpmyadmin/articles/v100-2x-qwen38-flash-next
https://zenn.dev/phpmyadmin/articles/v100-2x-qwen38-flash-next
NVLinkなしのTesla V100 32GB 2枚でQwen3.8-Flash-Nextを動かした
zenn.dev
September 27, 2026 at 8:35 AM
NVLinkなしのTesla V100 32GB 2枚でQwen3.8-Flash-Nextを動かした | toyoshi #zenn
https://zenn.dev/phpmyadmin/articles/v100-2x-qwen38-flash-next
https://zenn.dev/phpmyadmin/articles/v100-2x-qwen38-flash-next
One brain cell call that NVLink
January 5, 2026 at 10:07 PM
One brain cell call that NVLink
They fumbled that when they stopped EVGA from making dual-gpu cards, and killed off SLI in favor of NVlink that nobody uses.
June 19, 2026 at 3:28 AM
They fumbled that when they stopped EVGA from making dual-gpu cards, and killed off SLI in favor of NVlink that nobody uses.
You'd have to quantize the model down to int4. There's no fp8 hardware on Ampere. Only 2-way NVLink, no PCIe-card-to-card, it would definitely be a different experience from running on a single NVLink domain. It all depends on how you define "usable."
June 20, 2026 at 9:42 PM
You'd have to quantize the model down to int4. There's no fp8 hardware on Ampere. Only 2-way NVLink, no PCIe-card-to-card, it would definitely be a different experience from running on a single NVLink domain. It all depends on how you define "usable."
Sources: Apple is developing enterprise AI servers powered by two or four M8 Ultra chips, set for 2029, and is considering integrating Nvidia's NVLink Fusion (The Information)
Main Link | Techmeme Permalink
Main Link | Techmeme Permalink
September 16, 2026 at 1:16 PM
Sources: Apple is developing enterprise AI servers powered by two or four M8 Ultra chips, set for 2029, and is considering integrating Nvidia's NVLink Fusion (The Information)
Main Link | Techmeme Permalink
Main Link | Techmeme Permalink
AWS stellt mit Trainium3 einen viermal schnelleren und energieeffizienteren Chip vor. Ab der nächsten Chip-Generation kommt Nvidias NVLink Fusion zum Einsatz. #Amazon
Amazon mit neuem KI-Beschleuniger und künftig mit Nvidia-Technologie
AWS stellt mit Trainium3 einen viermal schnelleren und energieeffizienteren Chip vor. Ab der nächsten Chip-Generation kommt Nvidias NVLink Fusion zum Einsatz.
www.heise.de
December 2, 2025 at 8:46 PM
AWS stellt mit Trainium3 einen viermal schnelleren und energieeffizienteren Chip vor. Ab der nächsten Chip-Generation kommt Nvidias NVLink Fusion zum Einsatz. #Amazon
For custom CPU configurations, integration of NVIDIA NVLink-C2C IP for connectivity to NVIDIA GPUs is recommended for optimal performance. There is no NVLink chipset necessary.
August 22, 2025 at 4:04 PM
For custom CPU configurations, integration of NVIDIA NVLink-C2C IP for connectivity to NVIDIA GPUs is recommended for optimal performance. There is no NVLink chipset necessary.
⚡ NVIDIA Rubin R100 GPU Architecture: HBM4 288GB, TSMC 3nm GAA & NVLink 6 Break Through Datacenter Memory Walls
👉 Full Story: https://imtechboss.com/post.html?id=nvidia-rubin-r100-gpu-architecture-hbm4-288gb-tsmc-3nm-nvlink-6
#Tech #Hardware #NVIDIA
👉 Full Story: https://imtechboss.com/post.html?id=nvidia-rubin-r100-gpu-architecture-hbm4-288gb-tsmc-3nm-nvlink-6
#Tech #Hardware #NVIDIA
NVIDIA Rubin R100 GPU Architecture: HBM4 288GB, TSMC 3nm GAA & NVLink 6 Break Through Datacenter Memory Walls
NVIDIA has revealed architecture specifications for the Rubin R100 GPU family succeeding Blackwell. Featuring 288GB of ultra-fast HBM4 memory, 3nm GAA nanosheets, and NVLink 6 delivering 3.6 TB/s bi-directional fabric bandwidth, Rubin is engineered to eliminate KV-cache bottlenecks in trillion-parameter mixture-of-experts models.
imtechboss.com
September 26, 2026 at 7:10 AM
⚡ NVIDIA Rubin R100 GPU Architecture: HBM4 288GB, TSMC 3nm GAA & NVLink 6 Break Through Datacenter Memory Walls
👉 Full Story: https://imtechboss.com/post.html?id=nvidia-rubin-r100-gpu-architecture-hbm4-288gb-tsmc-3nm-nvlink-6
#Tech #Hardware #NVIDIA
👉 Full Story: https://imtechboss.com/post.html?id=nvidia-rubin-r100-gpu-architecture-hbm4-288gb-tsmc-3nm-nvlink-6
#Tech #Hardware #NVIDIA