AMD Radeon VII 16 GB review (updated) - Architecture and Specifications

by Hilbert Hagedoorn on: 03/05/2019 07:43 PM [

] 403 comment(s)

Product overview

Vega 20 is being fabbed on the latest 7nm process technology, which is the densest foundry process available in an effort to reduce power consumption, die size and lower operating voltages. For the last six years, graphics processors have relied on 28nm high-k/metal nodes.

Two years ago we moved towards the last gen 14nm FinFET, and now we can half the size with 7nm node fabrication.Where Vega had a 56 and 64 edition, the Radeon VII will have just one card, fitted with 16 GB of HBM2 graphics memory, four stacks thus with a 4096-bit memory bus. The card as such will offer 1 TB/s of memory bandwidth. So when we break down the products in a more simplistic way we see the following:


AMD Radeon	Radeon VII	RX Vega 64 LCS	RX Vega 64 LE	RX Vega 56	RX 580
Graphics Core	Vega 20	Vega 10	Vega 10	Vega 10	Polaris 20 XTX
Fabrication Process	7nm	14 nm	14 nm	14 nm	14 nm
Single Precision fp32	14.2 TFLOPs	13.7 TFLOPs	12.66 TFLOPs	10.5 TFLOPs	6.1 TFLOPs
Memory	16GB HBM2	8GB HBM2	8GB HBM2	8 GB HBM2	4/8 GB GDDR5
Clock Frequency Base / Boost	1400 MHz 1750 MHz	1,406 MHz 1,677 MHz	1,247 MHz 1,546 MHz	1,156 MHz 1,471 MHz	1,257 MHz 1,340 MHz
Compute Units	60	64	64	56	36
Shader Processors	3840	4096	4096	3584	2304
Memory Interface	4096-bit	2048-bit	2048-bit	2048-bit	256-bit
ROPs	64	64	64	64	32
TMUs	240	256	256	224	144
Memory Speed	1000 MHz	945 MHz	945 MHz	800 MHz	8 Gbps
Memory Bandwidth	1 TB/s	484 GB/s	484 GB/s	410 GB/s	256 GB/s
Board power	300 W	435 W	295 W	210 W	185 W
MSRP	699 USD	600 USD	499 USD	399 USD	$199 (4 GB) $229 (8 GB)

AMD has not unlocked the entire GPU, normally the Vega GPU is based on 64 shader clusters with 64 shader processors each. So that would make 4096 shader processors in total. AMD enabled 60 clusters (shader partitions), tied to 64 ROPs. TMUs wise for a fully enabled unit AMD has always used a 4:1 ratio meaning 60 CUs x 4 = 240 texture units.

	Vega 64 Air cooled (64 CU Peak 1630MHz)	Radeon VII (60 CU Peak 1800 MHz)	MI60 (64 CU Peak 1800 MHz)
Peak FP 64	0.84 TFLOPS	3.46 TFLOPS	7.4 TFLOPS
Peak FP 32	13.4 TFLOPS	13.8 TFLOPS	14.7 TFLOPS
Peak FP 16	26.7 TFLOPS	27.7 TFLOPS	29.5 TFLOPS
Peak Int 8	53.4 TOPS	55.3 TOPS	59 TOPS
Peak Int 4	106.8 TOPS	110.7 TOPS	118.0 TOPS

Double precision (fp64), the cards are setup in a 1:4 ratio configuration for fp64. The Radeon VII graphics card was created for gamers and creators, enthusiasts and early adopters. AMD offers double precision compute performance to 3.46 TFLOPS (DP=1/4SP).

31 pages « < 4 5 6 7 next »

AMD Radeon Super Resolution (RSR) - preview
AMD has released its new Radeon Software 22.3.1 drivers, supporting Radeon Super Resolution technology as a broader answer to fight off DLSS from NVIDIA. Will the new feature make enough of a differen...

Unboxing: AMD Radeon RX 6600 XT
AMD has announced its NAVI23 based graphics cards, announced not launched are the Radeon RX 6600 and 6600 XT. In this item we'll tell you a thing or two on what you may expect when released....

AMD Radeon RX 6700 XT (reference) review
Priced at $479 USD, AMD released their 'mainstream to high-end Radeon RX 6700 XT. A product that is to battle with the RTX 3060 Ti and 3070 from team green. Armed with 12GB of graphics memory, ...

AMD Radeon RX 6900 XT review
It sounds like a movie trailer; but the trilogy ends today, the 3rd iteration of AMD Big Navi gets reviewed, oh yeah the shader unlocked megalodon is going to battle the GeForce RTX 3090, whilst being...