a2-ultragpu-2g by Google Cloud Platform

a2-ultragpu-2g is a Accelerator Optimized: 2 NVIDIA A100 80GB GPUs, 24 vCPUs, 340GB RAM, 2 local SSD server offered by Google Cloud Platform with 24 vCPUs, 340 GiB of memory and 804 GB of storage. The pricing starts at 3.5865 USD per hour.
24 vCPU
340 GiB Memory
804 GB Storage
2 GPU
Spare SCore
18,972
(All-cores)
1,201
(Single-core)

Specifications

Server Metadata

Vendor ID
gcp
Server ID
1020024
Name
a2-ultragpu-2g
Description
Accelerator Optimized: 2 NVIDIA A100 80GB GPUs, 24 vCPUs, 340GB RAM, 2 local SSD
Family
a2
Hw Virt
Status
active
Observed At
2026-10-26T11:40:28.909461

Availability

REGION / IDSPOTONDEMAND
Council Bluffs (US) / us-central1
6.0163 USD/h 10.1376 USD/h
Columbus (US) / us-east5
3.5865 USD/h 11.1298 USD/h
Eemshaven (NL) / europe-west4
6.1365 USD/h 11.1618 USD/h
Ashburn (US) / us-east4
5.0581 USD/h 11.4155 USD/h
Jurong West (SG) / asia-southeast1
7.2326 USD/h 12.4954 USD/h

Processor

vCPUs
24
CPU Allocation
Dedicated
CPU Cores
12
CPU Speed
2.2 GHz
CPU Architecture
x86_64
CPU Manufacturer
Intel
CPU Family
Xeon
CPU L1D Cache
32 KiB
CPU L1D Cache Total
384 KiB
CPU L1I Cache
32 KiB
CPU L1I Cache Total
384 KiB
CPU L2 Cache
1 MiB
CPU L2 Cache Total
12 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
15.8
Scalability
65.83

System Resources and Accelerators

MEMORY
Memory Amount
340 GiB
Memory Amount Actual
340 GiB
ACCELERATORS
Accelerator Count
2
Accelerator Memory Min
80 GiB
Accelerator Memory Total
160 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)
STORAGE
Storage Size
804 GB
Storage Type
nvme ssd
Storages
  • 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 dual NVIDIA A100 GPUs and 340 GB RAM, delivering top-tier performance for large language model inference.

GPU AcceleratedMemory OptimizedStorage & Database

Google Cloud Platform a2-ultragpu-2g is an accelerator-optimized server designed for GPU-intensive workloads. It features two NVIDIA Ampere A100 GPUs with 160 GB of total VRAM, paired with 24 dedicated Intel Xeon vCPUs running at 2.2 GHz. The system includes 340.0 GB of RAM and 804 GB of local NVMe SSD storage. Benchmark results show weak single-core CPU performance, with a Geekbench single-core score of 1039.0, and average multi-core capabilities. However, it excels in LLM inference, achieving top-tier results in prompt processing and text generation across small, medium, and large models. This hardware profile represents a high-density accelerator resource, making it highly efficient for deep learning and parallelized computation, though less cost-effective for CPU-bound tasks. It is best suited for large language model inference, data science, and GPU-accelerated simulations.

Economics

Average Price per Region

Prices per Zone

Lowest Prices

Performance

Workload Profiles

0.93Score
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) 1,885,620 ops/sec 1,919,289 ops/sec 1 50.00% 50.00% 0.00%
Redis RPS (pipeline=16, SET) 11,486,525 ops/sec 11,728,058 ops/sec 0.92 20.00% 20.00% -1.70%
PassMark Memory Mark (composite) 2,160 2,416 0.92 10.00% 10.00% -0.83%
Memory bandwidth (read, 16 MB ~ L3) 100,784 MB/sec 108,466 MB/sec 0.91 10.00% 10.00% -0.94%
PassMark single-thread CPU 1,698 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
22.3 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
225.6 GB/s
GPU-to-GPU P2P gather bandwidth
276.9 GB/s
GPU-to-GPU P2P latency
1,250 ns
GPU-to-GPU P2P bandwidth
274.6 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
1,245 ns
Benchmark Environment
Driver Version
 535.183.01
CUDA Runtime Version
 12.6
nvbandwidth Version
 v0.5
GPU Count
 2
P2P Supported
Yes
Efficiency
SM/CE copy efficiency
1.00

Memory Bandwidth

Compression

OpenSSL

Geekbench Single-Core

Score: 1,039

Geekbench Multi-Core

Score: 7,923

Passmark CPU Scores

BENCHMARKSCORE
Mark
 18,766
Compression
 278,916
Encryption
 8,683
Extended Instructions
 18,034
Floating Point Maths
 33,907
Integer Maths
 63,382
Physics
 1,738
Prime Numbers
 85
Single Threaded
 1,698
String Sorting
 32,780

Passmark Memory Scores

BENCHMARKSCORE
Memory Mark
 2,160
Database Operations
 7,282
Memory Latency
 62
Memory Read Cached
 20,741
Memory Read Uncached
 7,791
Memory Write
 7,838

Stress-ng Stressors

CPU integerSingleMultiScaling
atomic
 201.68 181.82 0.90×
bitops
 92,137 1,269,799 13.78×
branch
 131,260 2,385,184 18.17×
funccall
 462.22 5,699 12.33×
funcret
 9,529 164,625 17.28×
hash
 107,462 1,837,387 17.10×
intmath
 18,951 340,341 17.96×
nop
 163,061 1,689,611 10.36×
prime
 271,389 3,642,826 13.42×
rotate
 82,350 1,359,593 16.51×
str
 625,812 8,357,939 13.36×
wcs
 400,936 5,017,403 12.51×
CryptographySingleMultiScaling
crypt
 2,079 24,294 11.68×
CompressionSingleMultiScaling
jpeg
 509.56 7,124 13.98×
zlib
 35.45 844.07 23.81×
HypervisorSingleMultiScaling
clock
 25,622 256,249 10.00×
clone
 305.13 2,592 8.50×
context
 2,211 30,982 14.01×
cpu-sched
 60.89 1,571 25.79×
eventfd
 46,558 556,051 11.94×
fault
 9,466 132,582 14.01×
fork
 3,437 43,032 12.52×
getrandom
 38,144 557,557 14.62×
hrtimers
 284,723 2,326,456 8.17×
mprotect
 1,043,083 3,315,575 3.18×
mtx
 8,405,816 548,141,383 65.21×
munmap
 2,271 30,490 13.43×
null
 250,055 3,086,927 12.35×
pipe
 831,939 7,325,024 8.80×
priv-instr
 470,152 6,448,672 13.72×
resched
 3,797,159 885,896 0.23×
switch
 76,006 3,487,203 45.88×
timermix
 169,466 2,886,776 17.03×
tlb-shootdown
 10,212 37,662 3.69×
tsc
 850,034 14,188,652 16.69×
usersyscall
 324,218 4,412,776 13.61×
vdso
 70,683,487 1,187,609,367 16.80×
vfork
 24,005 201,317 8.39×
x86syscall
 2,883,073 35,646,333 12.36×
yield
 11,698,597 11,597,743 0.99×
zero
 1,729,033 22,282,505 12.89×
CPU float/SIMDSingleMultiScaling
eigen
 13,759 199,679 14.51×
expmath
 2,276 41,791 18.36×
fma
 510,499 9,966,238 19.52×
fp
 103.2 1,581 15.32×
fractal
 4.2 73.81 17.57×
hyperbolic
 2,381 44,627 18.74×
logmath
 3,861 68,231 17.67×
matrix
 2,598 50,166 19.31×
matrix-3d
 563.43 2,990 5.31×
monte-carlo
 118.31 1,562 13.20×
mpfr
 1,030 13,268 12.88×
powmath
 1,350 24,686 18.28×
trig
 1,457 23,092 15.85×
veccmp
 13,247 165,240 12.47×
vecfp
 1,151 17,027 14.79×
vecmath
 1,865 35,221 18.88×
vecshuf
 480.77 6,150 12.79×
vecwide
 22,094 286,562 12.97×
AlgorithmsSingleMultiScaling
bitonicsort
 6.2 86.2 13.90×
bsearch
 205.71 2,803 13.62×
bubblesort
 0.6 4.8 8.00×
fibsearch
 208.24 2,870 13.78×
heapsort
 7.4 105.6 14.27×
hsearch
 4,275 63,261 14.80×
insertionsort
 7 81.6 11.66×
malloc
 1,922,939 25,917,649 13.48×
mcontend
 113.66 4,588 40.36×
mergesort
 19.8 316.39 15.98×
qsort
 12.6 172.99 13.73×
skiplist
 4,160 70,548 16.96×
tree
 1.2 11.79 9.82×
MemorySingleMultiScaling
cache
 245,373 516,446 2.10×
cachehammer
 336,464 6,709,408 19.94×
cacheline
 33.56 771.43 22.99×
l1cache
 193.51 2,508 12.96×
memcpy
 296.01 3,809 12.87×
memrate
 0.97 8.97 9.25×
memthrash
 133.92 667.31 4.98×
ptr-chase
 24,603,654 203,180,092 8.26×
spinmem
 188.19 884.67 4.70×
stream
 24.72 5,249 212.35×
vm
 40,335 295,090 7.32×

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-highgpu-4g 48340 GiB4
a2-ultragpu-4g 48680 GiB4
a2-highgpu-8g 96680 GiB8
a2-ultragpu-8g 961360 GiB8

Similar Servers

INSTANCEVENDORvCPUsMEMORYAccelerators

a2-ultragpu-2g FAQs