Product RoadMap - AI Segment
by Sam Chu, CEO, Polywell Computers
Polywell continues to expand its product line to support the AI segment. Currently, it is very hard and expensive to obtain the NVIDIA A100 GPU module; many developers are hence using the RTX4090 as an alternative to develop their AI applications and modeling.
Polywell offers the following models with RTX4090:
5U Server with 8 RTX4090 GPU
4U / Tower Xeon with 2 RTX4090 for Mainstream AI Computing
On the high-end side, NVIDIA is now leading the GPU for AI Computing with its A100.
Polywell's 2U4GPU-A100 and 4U8GP-A100 support this platform.
https://www.polywell.com/4u8gpu-a100-e7003
https://www.polywell.com/4u8gpu-a100-4189a
https://www.polywell.com/2u4gpu-a100-e7003
See All servers HERE
Mainstream GPU for AI computing:
2U Systems integrating GPU Adapters in various configurations; can install up to 4 GPU cards
https://www.polywell.com/2u5e32tp-x4677a
https://www.polywell.com/2u-e7300c
https://www.polywell.com/2u6gpu-4189a
https://www.polywell.com/4ut4gpu-4189a
Edge AI
NVIDIA Cuda Platform:
NEW!
The i2317 (https://www.polywell.com/i2317) utilizes the NVIDIA® RTX4060, and is an evolution from the i2314 (https://www.polywell.com/i2314) that uses the RTX3050.
Other modesl with NVIDIA Graphics
i2315: https://www.polywell.com/i2315 (new)
i2314: https://www.polywell.com/i2314
i2313 https://www.polywell.com/i2313
i2312: https://www.polywell.com/i2312b
i2311: https://www.polywell.com/i2311
5000PS: This model has open slots, meaning you can custom build your system by choosing different NVIDIA Graphics Cards: https://www.polywell.com/5000ps
Other Edge AI Systems are HERE
Other small form factor NVIDIA-based GPU PCs:
https://www.polywell.com/h510mxm https://www.polywell.com/nano-u8mx250
Intel OpenVino platform:
https://www.polywell.com/edge-ai-pc
Intel is developing its Gaudi to compete with the NVIDIA A100. Their next generation Gaudi 3, based on the PCIe 6.0, will bypass the NVIDIA GPU. This is because Intel has its PCIe bus controller inside the CPU, which can block performance I/O from NVIDIA.
We will see if Intel can overtake NVIDIA in 3 to 5 years when the Gaudi becomes mature, and more software developers port their AI apps from Cuda into the Intel Gaudi platform. We are also seeing Intel developing the ARC Graphics card to compete with NVIDIA's GeForce Graphics. It is hard to predict how this will go, but we feel that Intel will do better than NVIDIA now that PCIe 6.0 Ultra-fast bus lanes are here.
Intel is making the next generation PC as an AI PC by utilizing its PCIe 5.0 bus I/O to talk to SSD under NVME 5.0 and memory I/O with DDR5; the GPU business could also be dominated after PCIe 5.0 and later by 6.0.
Intel has lost the market for a decade to NVIDIA, but I see the new Intel CPU is bringing Intel back on track to compete. We are counting on Intel to develop products on that path, and we are betting our future of AI with Intel because their software is overall much stronger than NVIDIA.
Besides AI, Polywell continues to focus on the embedded segment of the business with our Mini PCs and ARM Android platforms. We are also adding more resources to the Server line while aiming for AI Computing with both GPU and Network I/O SDN, with 200G and 400G Network I/O modules through PCIe 5.0 and PCIe 6.0.
Looking Ahead - the Intel OpenVino Platform
We will be adding the new generation of Ultra Core family in Q4 2024, which includes the H810/B860/Q870/W880 -based Socket-S family, u/H/V-series of Mobile processor family like Meteor Lake, Lunar Lake and Arrow Lake.
