AMD announces Radeon Pro VII featuring 16GB HBM2 memory and Infinity Fabric Link

Published: 13th May 2020, 08:48 | Comments

newz

Today AMD launches Radeon Pro VII.

AMD Radeon PRO VII for professionals

The new Radeon Pro features 16 GB of HBM2 memory, which is the same capacity as NVIDIA Quadro RTX 5000, although the bandwidth is much higher (1 TB/s). Unlike the Radeon RX (Radeon VII) model, the Pro variant does not have an HDMI display output. Instead, AMD opted for six mini-DisplayPort connectors which are 8K capable.

AMD Radeon Pro VII is a PCIe Gen 4.0 graphics card, which is not surprising considering all Navi-based Pro models are compatible with the new interface. However, this is an advantage over the Turing-based Quadro RTX series, which are only PCIe Gen 3 capable. It is also worth mentioning that Radeon VII (from RX series) is only PCIe Gen 3.0.

The card has a typical board power of 250W, which is slightly less than RTX 5000 (260W). It requires a single 8-pin and a single 6-pin power connector. This configuration is another difference compared to the RX model, as the card was spec’ed at 295W TBP with dual 8-pin.

AMD is also bringing back a new multi-GPU bridge for the Radeon Pro VII. The bridge is called ‘Infinity Fabric Link’ and it has a bandwidth up to 168 GB/s. AMD is comparing its Infinity Fabric Link with NVIDIA NVLINK which can offer a maximum bidirectional bandwidth of 100 GB/s (for RTX 5000 it is 50 GB/s).

Update:

AMD confirmed that Radeon Pro VII features Vega 20 GPU with 3840 Stream Processors. The maximum double-precision compute performance is twice as high as Radeon VII (RX Series). The card will retail at 1,899 USD.

AMD Radeon Pro vs NVIDIA Quadro RTX
VideoCardz.comRadeon Pro VIIRadeon VIIRadeon Pro W5700Radeon Pro W5500Quarto RTX 6000Quadro RTX 5000
Picture
GPU7nm VEGA 207nm VEGA 207nm NAVI 10 GXL7nm NAVI 14 PRO XL12nm TURING TU10212nm TURING TU104
Cores
 
3840
 
3840
 
2304
 
1408
 
4608
 
3072
TMUs
 
240
 
240
 
144
 
88
 
288
 
192
ROPs
 
64
 
64
 
64
 
32
 
96
 
64
RT CoresN/AN/AN/AN/A
 
72
 
48
Tensor CoresN/AN/AN/AN/A
 
576
 
384
FP32 Perf.
 
13.1 TFLOPS
 
13.4 TFLOPS
 
8.7 TFLOPS
 
5.4 TFLOPS
 
16.3 TFLOPS
 
11.2 TFLOPS
FP64 Perf.
 
6.5 TFLOPS
 
3.5 TFLOPS
 
0.6 TFLOPS
 
0.3 TFLOPS
 
0.5 TFLOPS
 
0.3 TFLOPS
Peak GPU Clock
 
1705MHz
 
1750 MHz
 
1930 MHz
 
1900 MHz
 
1770 MHz
 
1815 MHz
Memory Config.
 
16 GB HBM2
 
16 GB HBM2
 
8 GB GDDR6
 
8 GB GDDR6
 
24 GB GDDR6
 
16 GB GDDR6
Memory Bandwidth
 
1 TB/s
 
1 TB/s
 
448 GB/s
 
448 GB/s
 
624 GB/s
 
448 GB/s
Board Power
 
250W
 
295W
 
205W
 
125W
 
295W
 
250W
PCIeGen 4Gen 3Gen 4Gen 4Gen 3Gen 3
LINK (bi-dir)
 
IFL 168 GB/s
N/AN/AN/A
 
NVL 100 GB/s
 
NVL 50 GB/s

.


by WhyCry

Previous Post
NVIDIA CEO Jensen Huang teases DGX A100 as 'world's largest graphics card'
Next Post
Intel announces Xeon W-1200 "Comet Lake" series





Back to Top ↑