a2-ultragpu-8g by Google Cloud Platform

a2-ultragpu-8g is a Accelerator Optimized: 8 NVIDIA A100 80GB GPUs, 96 vCPUs, 1360GB RAM, 8 local SSD server offered by Google Cloud Platform with 96 vCPUs, 1360 GiB of memory and 3.22 TB of storage. The pricing starts at 14.3459 USD per hour.
96 vCPU
1360 GiB Memory
3.22 TB Storage
8 GPU
Spare SCore
75,780
(All-cores)
1,201
(Single-core)

Specifications

Server Metadata

Vendor ID
gcp
Server ID
1020096
Name
a2-ultragpu-8g
Description
Accelerator Optimized: 8 NVIDIA A100 80GB GPUs, 96 vCPUs, 1360GB RAM, 8 local SSD
Family
a2
Hw Virt
Status
active
Observed At
2026-10-30T08:56:45.622587

Availability

REGION / IDSPOTONDEMAND
Council Bluffs (US) / us-central1
24.0652 USD/h 40.5504 USD/h
Columbus (US) / us-east5
14.3459 USD/h 44.5193 USD/h
Eemshaven (NL) / europe-west4
24.5462 USD/h 44.6473 USD/h
Ashburn (US) / us-east4
20.2323 USD/h 45.6619 USD/h
Jurong West (SG) / asia-southeast1
28.9302 USD/h 49.9815 USD/h

Processor

vCPUs
96
CPU Allocation
Dedicated
CPU Cores
48
CPU Speed
2.2 GHz
CPU Architecture
x86_64
CPU Manufacturer
Intel
CPU Family
Xeon
CPU L1D Cache
32 KiB
CPU L1D Cache Total
2 MiB
CPU L1I Cache
32 KiB
CPU L1I Cache Total
2 MiB
CPU L2 Cache
1 MiB
CPU L2 Cache Total
48 MiB
CPU L3 Cache
39 MiB
CPU L3 Cache Total
77 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
63.1
Scalability
65.73

System Resources and Accelerators

MEMORY
Memory Amount
1 TB
Memory Amount Actual
1 TB
ACCELERATORS
Accelerator Count
8
Accelerator Memory Min
80 GiB
Accelerator Memory Total
640 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: 590.48.01, BIOS version: 92.00.94.00.04, Clock rate: 1410)
  • NVIDIA Ampere NVIDIA A100-SXM4-80GB (Memory amount: 81920, Firmware version: 590.48.01, BIOS version: 92.00.94.00.04, Clock rate: 1410)
  • NVIDIA Ampere NVIDIA A100-SXM4-80GB (Memory amount: 81920, Firmware version: 590.48.01, BIOS version: 92.00.94.00.04, Clock rate: 1410)
  • NVIDIA Ampere NVIDIA A100-SXM4-80GB (Memory amount: 81920, Firmware version: 590.48.01, BIOS version: 92.00.94.00.04, Clock rate: 1410)
  • NVIDIA Ampere NVIDIA A100-SXM4-80GB (Memory amount: 81920, Firmware version: 590.48.01, BIOS version: 92.00.94.00.04, Clock rate: 1410)
  • NVIDIA Ampere NVIDIA A100-SXM4-80GB (Memory amount: 81920, Firmware version: 590.48.01, BIOS version: 92.00.94.00.04, Clock rate: 1410)
  • NVIDIA Ampere NVIDIA A100-SXM4-80GB (Memory amount: 81920, Firmware version: 590.48.01, BIOS version: 92.00.94.00.04, Clock rate: 1410)
  • NVIDIA Ampere NVIDIA A100-SXM4-80GB (Memory amount: 81920, Firmware version: 590.48.01, BIOS version: 92.00.94.00.04, Clock rate: 1410)
STORAGE
Storage Size
3216 GB
Storage Type
nvme ssd
Storages
  • 402 GB nvme ssd
  • 402 GB nvme ssd
  • 402 GB nvme ssd
  • 402 GB nvme ssd
  • 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

CPU and System Topology

Server Description

An accelerator-optimized platform featuring eight NVIDIA A100 GPUs and massive memory for high-throughput deep learning and LLM inference.

GPU AcceleratedMemory OptimizedStorage & Database

Google Cloud Platform a2-ultragpu-8g is an accelerator-optimized server designed for demanding parallel computing workloads. It features 96 vCPUs powered by Intel Xeon processors, 1360.0 GB of system memory, and 3216 GB of local NVMe SSD storage. The server is equipped with eight NVIDIA A100 GPUs, providing a total of 640 GB of VRAM. While single-core CPU benchmarks and single-threaded compression tasks perform in the lower tier, the instance delivers strong multi-core performance and exceptional memory read bandwidth. Benchmark data highlights top-tier capabilities in large language model inference, especially for prompt processing and text generation across medium and large models. This server is highly suited for deep learning, large-scale LLM deployment, ray tracing, and high-throughput database operations like Redis, where massive GPU acceleration and high memory density are required.

Economics

Average Price per Region

Prices per Zone

Lowest Prices

Performance

Workload Profiles

2.4Score
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) 6,657,746 ops/sec 1,919,289 ops/sec 3.5 50.00% 50.00% +87.00%
Redis RPS (pipeline=16, SET) 38,935,432 ops/sec 11,728,058 ops/sec 3.2 20.00% 20.00% +26.00%
PassMark Memory Mark (composite) 2,374 2,416 1 10.00% 10.00% 0.00%
Memory bandwidth (read, 16 MB ~ L3) 178,306 MB/sec 108,466 MB/sec 1.6 10.00% 10.00% +4.80%
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
41.3 GB/s
Node GPU-to-host bandwidth
52.8 GB/s
Node host-to-GPU bandwidth
49.5 GB/s
Peer-to-peer
GPU-to-GPU P2P duplex bandwidth
213.9 GB/s
GPU-to-GPU P2P gather bandwidth
274.2 GB/s
GPU-to-GPU P2P latency
1,272 ns
GPU-to-GPU P2P bandwidth
272.1 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
1,234 ns
Benchmark Environment
Driver Version
 535.183.01
CUDA Version (driver)
 12.6
nvbandwidth Version
 v0.5
GPU Count
 8
P2P Supported
Yes
Efficiency
SM/CE copy efficiency
1.00

Memory Bandwidth

Compression

OpenSSL

Geekbench Single-Core

Score: 1,060

Geekbench Multi-Core

Score: 11,609

Passmark CPU Scores

BENCHMARKSCORE
Mark
 48,614
Compression
 1,011,247
Encryption
 30,571
Extended Instructions
 64,383
Floating Point Maths
 135,191
Integer Maths
 253,165
Physics
 4,759
Prime Numbers
 299
Single Threaded
 1,696
String Sorting
 113,705

Passmark Memory Scores

BENCHMARKSCORE
Memory Mark
 2,374
Database Operations
 19,517
Memory Latency
 56
Memory Read Cached
 20,863
Memory Read Uncached
 8,442
Memory Write
 8,537

Stress-ng Stressors

CPU integerSingleMultiScaling
atomic
 223.82 118.26 0.53×
bitops
 92,150 5,070,457 55.02×
branch
 131,462 9,532,625 72.51×
funccall
 462.35 22,352 48.34×
funcret
 9,682 645,997 66.72×
hash
 107,610 7,383,741 68.62×
intmath
 19,109 1,358,451 71.09×
nop
 163,189 6,756,025 41.40×
prime
 271,749 14,192,072 52.22×
rotate
 82,451 5,403,906 65.54×
str
 603,813 28,964,734 47.97×
wcs
 412,313 19,182,693 46.52×
CryptographySingleMultiScaling
crypt
 2,083 96,726 46.45×
CompressionSingleMultiScaling
jpeg
 499.24 26,734 53.55×
zlib
 40.66 3,233 79.52×
HypervisorSingleMultiScaling
clock
 21,694 166,564 7.68×
clone
 293.01 6,452 22.02×
context
 2,209 119,096 53.91×
cpu-sched
 58.67 6,190 105.50×
eventfd
 45,512 1,851,953 40.69×
fault
 9,541 130,068 13.63×
fork
 3,248 44,085 13.57×
getrandom
 38,199 2,204,328 57.71×
hrtimers
 450,277 7,829,543 17.39×
mprotect
 576,262 11,812,944 20.50×
mtx
 7,729,541 1,763,098,279 228.10×
munmap
 2,257 52,420 23.23×
null
 254,776 11,551,730 45.34×
pipe
 824,442 20,878,997 25.33×
priv-instr
 480,156 9,210,608 19.18×
resched
 3,930,026 5,321,193 1.35×
switch
 78,113 12,631,038 161.70×
timermix
 168,274 11,175,658 66.41×
tlb-shootdown
 7,424 86,952 11.71×
tsc
 850,416 56,709,784 66.68×
usersyscall
 331,602 8,784,204 26.49×
vdso
 70,579,968 4,541,531,901 64.35×
vfork
 21,258 98,526 4.63×
x86syscall
 2,898,030 131,537,562 45.39×
yield
 44,658,430 50,647,077 1.13×
zero
 1,738,045 88,247,016 50.77×
CPU float/SIMDSingleMultiScaling
eigen
 13,776 784,146 56.92×
expmath
 2,282 162,798 71.35×
fma
 509,093 39,645,582 77.87×
fp
 107.26 6,247 58.24×
fractal
 4.2 288.8 68.76×
hyperbolic
 2,373 176,352 74.32×
logmath
 3,862 266,767 69.07×
matrix
 2,592 194,289 74.96×
matrix-3d
 599.22 5,991 10.00×
monte-carlo
 118.28 6,178 52.23×
mpfr
 1,033 51,056 49.44×
powmath
 1,354 96,049 70.96×
trig
 1,459 86,506 59.29×
veccmp
 13,270 626,492 47.21×
vecfp
 1,151 69,274 60.17×
vecmath
 1,864 136,797 73.39×
vecshuf
 479.89 24,694 51.46×
vecwide
 21,924 1,109,872 50.62×
AlgorithmsSingleMultiScaling
bitonicsort
 6.2 324.99 52.42×
bsearch
 203.65 11,361 55.79×
bubblesort
 0.6 19.2 32.00×
fibsearch
 208.48 11,484 55.09×
heapsort
 7.4 403.59 54.54×
hsearch
 4,297 240,335 55.93×
insertionsort
 7 326 46.57×
malloc
 2,005,461 57,155,870 28.50×
mcontend
 81.12 16,780 206.85×
mergesort
 19.8 1,168 58.98×
qsort
 12.6 658.98 52.30×
skiplist
 4,156 279,641 67.29×
tree
 1.2 38.36 31.97×
MemorySingleMultiScaling
cache
 235,832 1,684,401 7.14×
cachehammer
 335,368 13,587,480 40.52×
cacheline
 14.97 1,366 91.23×
l1cache
 183.56 9,427 51.36×
memcpy
 297.21 14,526 48.87×
memrate
 1.17 14.94 12.77×
memthrash
 272.17 916.7 3.37×
ptr-chase
 31,174,214 765,978,734 24.57×
spinmem
 190.99 3,551 18.59×
stream
 12.22 21,227 1737.03×
vm
 53,541 366,626 6.85×

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-ultragpu-4g 48680 GiB4
a2-highgpu-8g 96680 GiB8

Similar Servers

INSTANCEVENDORvCPUsMEMORYAccelerators

a2-ultragpu-8g FAQs