Instances›AWS EC2›g4dn›g4dn.metal
Compute
vCPUs96
Memory384 GiB
Mem / vCPU4.00 GiB
ProcessorXeon Family
Clock speed2.5 GHz
Architecturex86_64
Storage
Type2 × 900 GB NVMe SSD
Bandwidth19,000 Mbps
Throughput2375 MB/s
IOPS80,000
Networking
Performance100 Gigabit
Max ENIs15
IPv6Yes
GPUNVIDIA T4 Tensor Core
Family sizes
g4dn.xlarge4 vCPU · 16 GiB
g4dn.2xlarge8 vCPU · 32 GiB
g4dn.4xlarge16 vCPU · 64 GiB
g4dn.8xlarge32 vCPU · 128 GiB
g4dn.12xlarge48 vCPU · 192 GiB
g4dn.16xlarge64 vCPU · 256 GiB
g4dn.metal96 vCPU · 384 GiB
g4dn.metal

The g4dn.metal is a GPU AWS EC2 instance: 96 vCPU and 384 GiB of memory, 4.00 GiB per vCPU. In the name, "g" is graphics, "4" is the generation, "dn" marks local NVMe storage and enhanced networking and "metal" is the size within the g4dn family. It runs on Intel Xeon Family at 2.5 GHz, and AWS introduced it in June 2020. AWS has since shipped g5, g6 and g7; g5g.metal is the direct replacement at this size. Rackspace Spot has no equivalent to offer: 8 × NVIDIA T4 Tensor Core wired together behind a high-speed interconnect is one tightly coupled machine, not 8 independent nodes — and that interconnect is usually what the instance was bought for. Spot's largest VM tops out at 16 vCPU and 128 GiB, and even that is not the missing piece here. The table below prices what AWS charges; the honest answer on the Spot side is that this workload stays where it is.

Match qualityNo equivalent

g4dn.metal is 8 Turing, Turing, Turing, Turing, Turing, Turing, Turing, Turing GPUs (128 GiB VRAM) wired together inside one chassis. The interconnect is what you bought — that is what makes it one accelerator with pooled memory rather than several machines — and Rackspace Spot has nothing that reproduces it. The Spot GPU class is built on the NVIDIA A30 (Ampere, 24 GB VRAM), and separate nodes of it on ordinary datacentre networking are not a substitute for a single interconnect-joined node.

Split the job across independent single-GPU nodes only if it is data-parallel with no per-step synchronisation and the model fits one card. Model-parallel work, or anything synchronising gradients every step, does not survive the move. Rackspace Spot sells x86_64 Kubernetes nodes up to 16 vCPU, plus one NVIDIA A30 GPU class.

Pricing by region
1 hr
RegionAWS on-demandAWS Spot1-yr Reserved3-yr ReservedRackspace SpotYou save
US East, N. Virginia$7.824$4.765$4.929$3.380
US Central, Chicagono AWS region
US Central, Dallasno AWS region
Hong Kong$12.048$7.590$5.205
Sydney, Australia$10.174$4.470$6.410$4.945
London, UK$9.154$1.143$6.247$4.449
US West, San Jose$9.389$1.273$6.605$4.563
US Central, Dallas 2no AWS region
US East, N. Virginia 2$7.824$4.765$4.929$3.380
When to use it
Good for
  • ●CUDA-accelerated batch jobs
  • ●Model inference and embedding generation
  • ●Video transcoding and image processing
  • ●Workloads that fit inside 384 GiB of RAM
  • ●Jobs that want 2 × 900 GB of local NVMe scratch space
  • ●Network-heavy services — AWS rates this size at 100 Gigabit
Avoid if
  • ●Starting a new deployment. g5g.metal is the newer generation at this size.
  • ●The working set needs more than 384 GiB of RAM
  • ●You need dedicated hardware. Rackspace Spot sells virtualised compute only.
  • ●Anything the CPU can do fast enough. GPU hours are the most expensive on the price list.
  • ●More than 4.00 GiB of RAM per vCPU is needed — that ratio is fixed for the whole family
  • ●The work is single-threaded. Most of these 96 vCPU would sit idle.
Rackspace Spot sells spare capacity at market prices. You pay what the market decides, not a list price.
  • Spot sells spare capacity at market prices. It is the cheapest compute on the internet.
  • Prices are set by supply and demand. You pay what the market decides.
  • Compute is delivered as fully managed Kubernetes. All infrastructure is taken care of.
  • Egress costs $0.01/GB after the free allowance. AWS charges $0.09/GB.
Build your Cloudspace
Frequently asked questions