a2-ultragpu-4g by Google Cloud Platform

a2-ultragpu-4g is a Accelerator Optimized: 4 NVIDIA A100 80GB GPUs, 48 vCPUs, 680GB RAM, 4 local SSD server offered by Google Cloud Platform with 48 vCPUs, 680 GiB of memory and 1.61 TB of storage. The pricing starts at 7.527 USD per hour.
48 vCPU
680 GiB Memory
1.61 TB Storage
4 GPU
Spare SCore
37,882
(All-cores)
1,195
(Single-core)

Specifications

Server Metadata

Vendor ID
gcp
Server ID
1020048
Name
a2-ultragpu-4g
Description
Accelerator Optimized: 4 NVIDIA A100 80GB GPUs, 48 vCPUs, 680GB RAM, 4 local SSD
Family
a2
Hw Virt
Status
active
Observed At
2026-11-04T00:40:08.354907

Availability

REGION / IDSPOTONDEMAND
Council Bluffs (US) / us-central1
11.9071 USD/h 20.2752 USD/h
Columbus (US) / us-east5
7.527 USD/h 22.2597 USD/h
Eemshaven (NL) / europe-west4
12.1548 USD/h 22.3237 USD/h
Ashburn (US) / us-east4
10.1162 USD/h 22.8309 USD/h
Jurong West (SG) / asia-southeast1
14.4698 USD/h 24.9907 USD/h

Processor

vCPUs
48
CPU Allocation
Dedicated
CPU Cores
24
CPU Speed
2.2 GHz
CPU Architecture
x86_64
CPU Manufacturer
Intel
CPU Family
Xeon
CPU L1D Cache
32 KiB
CPU L1D Cache Total
768 KiB
CPU L1I Cache
32 KiB
CPU L1I Cache Total
768 KiB
CPU L2 Cache
1 MiB
CPU L2 Cache Total
24 MiB
CPU L3 Cache
39 MiB
CPU L3 Cache Total
39 MiB
CPU Flags
fpu, vme, de, pse, tsc, msr, pae, mce, cx8, apic, sep, mtrr, pge, mca, cmov, pat, pse36, clflush, mmx, fxsr, sse, sse2, ss, ht, syscall, nx, pdpe1gb, rdtscp, lm, constant_tsc, rep_good, nopl, xtopology, nonstop_tsc, cpuid, tsc_known_freq, pni, pclmulqdq, ssse3, fma, cx16, pcid, sse4_1, sse4_2, x2apic, movbe, popcnt, aes, xsave, avx, f16c, rdrand, hypervisor, lahf_lm, abm, 3dnowprefetch, ssbd, ibrs, ibpb, stibp, ibrs_enhanced, fsgsbase, tsc_adjust, bmi1, hle, avx2, smep, bmi2, erms, invpcid, rtm, mpx, avx512f, avx512dq, rdseed, adx, smap, clflushopt, clwb, avx512cd, avx512bw, avx512vl, xsaveopt, xsavec, xgetbv1, xsaves, arat, avx512_vnni, md_clear, arch_capabilities
Ecpus
31.7
Scalability
66.04

System Resources and Accelerators

MEMORY
Memory Amount
680 GiB
Memory Amount Actual
680 GiB
ACCELERATORS
Accelerator Count
4
Accelerator Memory Min
80 GiB
Accelerator Memory Total
320 GiB
Accelerator Type
GPU
Accelerator Manufacturer
NVIDIA
Accelerator Family
Ampere
Accelerator Model
A100
Accelerators
  • NVIDIA Ampere NVIDIA A100-SXM4-80GB (Memory amount: 81920, Firmware version: 535.183.01, BIOS version: 92.00.94.00.04, Clock rate: 1410)
  • NVIDIA Ampere NVIDIA A100-SXM4-80GB (Memory amount: 81920, Firmware version: 535.183.01, BIOS version: 92.00.94.00.04, Clock rate: 1410)
  • NVIDIA Ampere NVIDIA A100-SXM4-80GB (Memory amount: 81920, Firmware version: 535.183.01, BIOS version: 92.00.94.00.04, Clock rate: 1410)
  • NVIDIA Ampere NVIDIA A100-SXM4-80GB (Memory amount: 81920, Firmware version: 535.183.01, BIOS version: 92.00.94.00.04, Clock rate: 1410)
STORAGE
Storage Size
1608 GB
Storage Type
nvme ssd
Storages
  • 402 GB nvme ssd
  • 402 GB nvme ssd
  • 402 GB nvme ssd
  • 402 GB nvme ssd
NETWORK
Inbound Traffic
0 GB/month
Outbound Traffic
0 GB/month
IPv4
0

Server Description

An accelerator-optimized platform featuring four dedicated GPUs and high-capacity memory for demanding parallel processing and large language model inference.

GPU AcceleratedMemory OptimizedStorage & Database

Google Cloud Platform a2-ultragpu-4g is an accelerator-optimized server featuring four NVIDIA Ampere A100 GPUs with 320 GB of total VRAM, 48 vCPUs on an Intel Xeon x86_64 architecture, 680.0 GB of RAM, and 1608 GB of NVMe SSD storage. While single-core CPU benchmarks place the instance in the bottom 25%, its multi-core and memory bandwidth performance sit in the middle 50%. The server achieves top-tier status in LLM inference prompt processing and text generation for medium and large models. This hardware profile is optimized for GPU-heavy workloads, trading off single-threaded CPU speed for massive parallel processing capabilities. It is designed for deep learning, large language model inference, and complex data science workloads.

Economics

Average Price per Region

Prices per Zone

Lowest Prices

Performance

Workload Profiles

1.4Score
Precomputed compound score for Cache Intensive workloads. A weighted average (geometric mean) of benchmark scores compared to their medians: score = ∏ (x_i / m_i)^(w_i / Σw). The score of 1.0 represents a synthetic baseline server with the median performance of each component benchmark; 0.5 means roughly half the performance; and 2.0 means twice the performance of that reference profile. Component weights: 50% Redis RPS (pipeline=1, SET), 20% Redis RPS (pipeline=16, SET), 10% PassMark Memory Mark (composite), 10% Memory bandwidth (read, 16 MB ~ L3), 10% PassMark single-thread CPU. Rationale for component selection: In-memory key-value store workload, mixing direct Redis performance metrics with memory speed and latency benchmarks, and single-core CPU performance profiles.
Component Score Weight Impact
Raw Reference Normalized Target Actual
Redis RPS (pipeline=1, SET) 3,548,377 ops/sec 1,930,206 ops/sec 1.8 50.00% 50.00% +34.00%
Redis RPS (pipeline=16, SET) 20,539,926 ops/sec 11,792,140 ops/sec 1.8 20.00% 20.00% +12.00%
PassMark Memory Mark (composite) 2,265 2,412 0.96 10.00% 10.00% -0.41%
Memory bandwidth (read, 16 MB ~ L3) 89,817 MB/sec 108,490 MB/sec 0.82 10.00% 10.00% -2.00%
PassMark single-thread CPU 1,696 Mops/s 2,365 Mops/s 0.71 10.00% 10.00% -3.40%

GPU Bandwidth and Latency

Node (all GPUs)
Node baseline directional bandwidth
20.6 GB/s
Node GPU-to-host bandwidth
26.4 GB/s
Node host-to-GPU bandwidth
24.7 GB/s
Peer-to-peer
GPU-to-GPU P2P duplex bandwidth
214.3 GB/s
GPU-to-GPU P2P gather bandwidth
280.0 GB/s
GPU-to-GPU P2P latency
1,248 ns
GPU-to-GPU P2P bandwidth
273.3 GB/s
Single GPU slot
Per-GPU GPU-to-host bandwidth
13.2 GB/s
Per-GPU host-to-GPU bandwidth
12.3 GB/s
Per-GPU host-device latency
1,240 ns
Benchmark Environment
Driver Version
 535.183.01
CUDA Version (driver)
 12.6
nvbandwidth Version
 v0.5
GPU Count
 4
P2P Supported
Yes
Efficiency
SM/CE copy efficiency
1.00

Memory Bandwidth

Compression

OpenSSL

Geekbench Single-Core

Score: 1,051

Geekbench Multi-Core

Score: 10,350

Passmark CPU Scores

BENCHMARKSCORE
Mark
 30,679
Compression
 518,172
Encryption
 15,662
Extended Instructions
 32,899
Floating Point Maths
 68,172
Integer Maths
 126,890
Physics
 2,398
Prime Numbers
 153
Single Threaded
 1,696
String Sorting
 58,640

Passmark Memory Scores

BENCHMARKSCORE
Memory Mark
 2,265
Database Operations
 14,059
Memory Latency
 56
Memory Read Cached
 20,825
Memory Read Uncached
 7,606
Memory Write
 7,974

Stress-ng Raw Scores

Stress-ng Relative Multicore Performance

LLM Inference Speed for Prompt Processing

LLM Inference Speed for Text Generation

Static Web Server

Redis

Alternatives

Servers of the Same Family

INSTANCEvCPUsMEMORYAccelerators
a2-highgpu-1g 1285 GiB1
a2-ultragpu-1g 12170 GiB1
a2-highgpu-2g 24170 GiB2
a2-ultragpu-2g 24340 GiB2
a2-highgpu-4g 48340 GiB4
a2-highgpu-8g 96680 GiB8
a2-ultragpu-8g 961360 GiB8

Similar Servers

INSTANCEVENDORvCPUsMEMORYAccelerators

a2-ultragpu-4g FAQs