From WikiChip
Difference between revisions of "qualcomm/cloud ai"
< qualcomm

(cloud ai)
 
Line 32: Line 32:
 
}}
 
}}
 
'''Cloud AI''' is a family of [[neural processors]] for the edge and data center market designed by [[Qualcomm]] and introduced in early 2021.
 
'''Cloud AI''' is a family of [[neural processors]] for the edge and data center market designed by [[Qualcomm]] and introduced in early 2021.
 +
 +
== Cloud AI 100 ==
 +
{{main|qualcomm/microarchitectures/cloud ai|l1=Cloud AI}}
 +
[[File:qualcomm-cloud-ai-100-2.png|right]]
 +
Launched in early 2021, the Cloud AI 100 series are the first series of AI inference processors. Those processors range from 70 [[TOPS]] / 35 [[FLOPS]] to 400 [[TOPS]] / 200 [[FLOPS]] and are offered in both PCIe cards and M.2 modules (with and without heatsinks).
 +
 +
{| class="wikitable" style="text-align: center;"
 +
|-
 +
! FF
 +
| PCIe (HHHL) || Dual M.2 || Dual M.2e (No Heatsink)
 +
|-
 +
! Size
 +
| 68.9 mm x 169.5 mm || 46 mm x 110 mm || 46 mm x 110 mm
 +
|-
 +
! TDP
 +
| 75 W || 15-25 W || 15-25 W
 +
|-
 +
! colspan="4" | Peak Compute
 +
|-
 +
! Int8
 +
| 400 TOPS || 200 TOPS || 70 TOPS
 +
|-
 +
! FP16
 +
| 200 teraFLOPS || 100 teraFLOPS || 35 teraFLOPS
 +
|-
 +
! colspan="4" | Configuration
 +
|-
 +
! SRAM
 +
| 144 MiB || 144 MiB || 72 MiB
 +
|-
 +
! DRAM (w/ECC)
 +
| 16 GiB LPDDR4x-4266
 +
| 32 GiB LPDDR4x-4266
 +
| 8 GiB LPDDR4x-4266
 +
|-
 +
! DRAM B/W
 +
| 136.5 GB/s
 +
| 136.5 GB/s
 +
| 68.25 GB/s
 +
|-
 +
! Host Interface
 +
| PCIe Gen 4 (x8)
 +
| PCIe Gen 4 (x8)
 +
| PCIe Gen 3 (x4)
 +
|}

Revision as of 04:48, 15 September 2021

Qualcomm Cloud AI
Developer Qualcomm
Type System on chips
Introduction Apr 9, 2019 (announced)
March, 2021 (launch)
Architecture Multi-core Neural Processor
ISA Custom
µarch Cloud AI
Process 7 nm
0.007 μm
7.0e-6 mm
Technology CMOS

Cloud AI is a family of neural processors for the edge and data center market designed by Qualcomm and introduced in early 2021.

Cloud AI 100

Main article: Cloud AI
qualcomm-cloud-ai-100-2.png

Launched in early 2021, the Cloud AI 100 series are the first series of AI inference processors. Those processors range from 70 TOPS / 35 FLOPS to 400 TOPS / 200 FLOPS and are offered in both PCIe cards and M.2 modules (with and without heatsinks).

FF PCIe (HHHL) Dual M.2 Dual M.2e (No Heatsink)
Size 68.9 mm x 169.5 mm 46 mm x 110 mm 46 mm x 110 mm
TDP 75 W 15-25 W 15-25 W
Peak Compute
Int8 400 TOPS 200 TOPS 70 TOPS
FP16 200 teraFLOPS 100 teraFLOPS 35 teraFLOPS
Configuration
SRAM 144 MiB 144 MiB 72 MiB
DRAM (w/ECC) 16 GiB LPDDR4x-4266 32 GiB LPDDR4x-4266 8 GiB LPDDR4x-4266
DRAM B/W 136.5 GB/s 136.5 GB/s 68.25 GB/s
Host Interface PCIe Gen 4 (x8) PCIe Gen 4 (x8) PCIe Gen 3 (x4)
Facts about "Cloud AI - Qualcomm"
designerQualcomm +
first announcedApril 9, 2019 +
first launchedMarch 2021 +
full page namequalcomm/cloud ai +
instance ofsystem on a chip family +
instruction set architectureCustom +
main designerQualcomm +
microarchitectureCloud AI +
nameQualcomm Cloud AI +
process7 nm (0.007 μm, 7.0e-6 mm) +
technologyCMOS +