a3-highgpu-2g by Google Cloud Platform

a3-highgpu-2g is a Accelerator Optimized: 2 NVIDIA H100 GPU, 52 vCPUs, 468GB RAM server offered by Google Cloud Platform with 52 vCPUs, 468 GiB of memory and 1.61 TB of storage. The pricing starts at 2.2844 USD per hour.
52 vCPU
468 GiB Memory
1.61 TB Storage
2 GPU
Spare SCore
42,151
(All-cores)
1,854
(Single-core)

Specifications #

Server Metadata #

Vendor ID
gcp
Server ID
1720362
Name
a3-highgpu-2g
Description
Accelerator Optimized: 2 NVIDIA H100 GPU, 52 vCPUs, 468GB RAM
Family
GPU Machines
Series
a3
Hw Virt
Average Time To Start
35
Status
active
Observed At
2026-11-18T22:52:17.003960

Availability #

Region / IDSpotOndemand
Council Bluffs (US) / us-central1
13.1751 USD/h
22.1225 USD/h
The Dalles (US) / us-west1
13.1751 USD/h
22.1225 USD/h
Ashburn (US) / us-east4
12.6246 USD/h
22.1389 USD/h
Mumbai (IN) / asia-south1
11.7134 USD/h
23.0337 USD/h
St. Ghislain (BE) / europe-west1
11.8415 USD/h
24.3425 USD/h
Hamina (FI) / europe-north1
2.2844 USD/h
24.3589 USD/h
Columbus (US) / us-east5
9.5007 USD/h
24.8958 USD/h
Las Vegas (US) / us-west4
11.6756 USD/h
24.9122 USD/h
London (GB) / europe-west2
5.9545 USD/h
25.2295 USD/h
Changhua County (TW) / asia-east1
15.3535 USD/h
25.5897 USD/h
Paris (FR) / europe-west9
8.7808 USD/h
25.6621 USD/h
Frankfurt (DE) / europe-west3
14.8865 USD/h
26.1078 USD/h
Delhi (IN) / asia-south2
8.1021 USD/h
26.5712 USD/h
Sydney (AU) / australia-southeast1
6.9557 USD/h
27.6696 USD/h
Eemshaven (NL) / europe-west4
16.8741 USD/h
28.1351 USD/h
Seoul (KR) / asia-northeast3
14.3197 USD/h
28.4123 USD/h
Jurong West (SG) / asia-southeast1
11.9958 USD/h
28.5696 USD/h
Melbourne (AU) / australia-southeast2
2.6981 USD/h
28.7675 USD/h
Tokyo (JP) / asia-northeast1
5.9853 USD/h
31.6608 USD/h

Processor #

vCPUs
52
CPU Allocation
Dedicated
CPU Cores
26
CPU Speed
2.7 GHz
CPU Architecture
x86_64
CPU Manufacturer
Intel
CPU Family
Xeon
CPU Model
8481C
CPU L1D Cache
48 KiB
CPU L1D Cache Total
1 MiB
CPU L1I Cache
32 KiB
CPU L1I Cache Total
832 KiB
CPU L2 Cache
2 MiB
CPU L2 Cache Total
52 MiB
CPU L3 Cache
105 MiB
CPU L3 Cache Total
105 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, vmx, 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, tpr_shadow, flexpriority, ept, vpid, ept_ad, fsgsbase, tsc_adjust, bmi1, avx2, smep, bmi2, erms, invpcid, rtm, avx512f, avx512dq, rdseed, adx, smap, avx512ifma, clflushopt, clwb, avx512cd, sha_ni, avx512bw, avx512vl, xsaveopt, xsavec, xgetbv1, xsaves, avx_vnni, avx512_bf16, arat, vnmi, avx512vbmi, umip, avx512_vbmi2, gfni, vaes, vpclmulqdq, avx512_vnni, avx512_bitalg, avx512_vpopcntdq, la57, rdpid, cldemote, movdiri, movdir64b, fsrm, md_clear, serialize, tsxldtrk, amx_bf16, avx512_fp16, amx_tile, amx_int8, arch_capabilities
Ecpus
22.7
Scalability
43.65

System Resources and Accelerators #

Memory
Memory Amount
468 GiB
Memory Amount Actual
468 GiB
Accelerators
Accelerator Count
2
Accelerator Memory Min
80 GiB
Accelerator Memory Total
160 GiB
Accelerator Type
GPU
Accelerator Manufacturer
NVIDIA
Accelerator Family
Hopper
Accelerator Model
H100
Accelerators
  • NVIDIA Hopper NVIDIA H100 80GB HBM3 (Memory amount: 81559, Firmware version: 595.71.05, BIOS version: 96.00.CF.00.01, Clock rate: 1980)
  • NVIDIA Hopper NVIDIA H100 80GB HBM3 (Memory amount: 81559, Firmware version: 595.71.05, BIOS version: 96.00.CF.00.01, Clock rate: 1980)
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
Compatible Storage IDs
30014, 30017, 30013, 30016, 30007, 30002
Network
Inbound Traffic
0 GB/month
Outbound Traffic
0 GB/month
IPv4
0

CPU and System Topology #

Server Description #

An accelerator-optimized server featuring dual NVIDIA H100 GPUs and high-speed NVMe storage for demanding machine learning and inference workloads.

Google Cloud Platform a3-highgpu-2g is an accelerator-optimized server designed for high-compute and machine learning workloads. It features two NVIDIA H100 Hopper GPUs with 160 GB of total VRAM, paired with 52 vCPUs from an Intel Xeon 8481C processor running at 2.7 GHz. The system includes 468 GB of RAM and a 1608 GB local NVMe SSD. Benchmarks demonstrate top-tier performance in LLM inference, with prompt processing speeds of 5644.37 tokens per second on a 7B model, alongside strong memory bandwidth and multi-threaded decompression capabilities. This server is optimized for deep learning, large language model inference, and high-performance database operations.

Economics #

Average Price per Region #

Prices per Zone #

Lowest Prices #

Performance #

Workload Profiles #

2.6Score
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) 7,257,322 ops/sec 1,936,298 ops/sec 3.8 50.00% 50.00% +95.00%
Redis RPS (pipeline=16, SET) 37,434,610 ops/sec 11,817,332 ops/sec 3.1 20.00% 20.00% +25.00%
PassMark Memory Mark (composite) 2,230 2,408 0.92 10.00% 10.00% -0.83%
Memory bandwidth (read, 16 MB ~ L3) 232,792 MB/sec 108,674 MB/sec 2.1 10.00% 10.00% +7.70%
PassMark single-thread CPU 2,720 Mops/s 2,365 Mops/s 1.1 10.00% 10.00% +0.96%

GPU Bandwidth and Latency #

Node (all GPUs)
Node baseline directional bandwidth
24.5 GB/s
Node GPU-to-host bandwidth
28.6 GB/s
Node host-to-GPU bandwidth
27.7 GB/s
Peer-to-peer
GPU-to-GPU P2P duplex bandwidth
393.0 GB/s
GPU-to-GPU P2P gather bandwidth
397.6 GB/s
GPU-to-GPU P2P latency
827 ns
GPU-to-GPU P2P bandwidth
397.5 GB/s
Single GPU slot
Per-GPU GPU-to-host bandwidth
28.6 GB/s
Per-GPU host-to-GPU bandwidth
27.7 GB/s
Per-GPU host-device latency
1,198 ns
Per-GPU device memory copy bandwidth
1498.2 GB/s
Benchmark Environment
Driver Version
595.71.05
CUDA Version (driver)
13.2
nvbandwidth Version
v0.8
GPU Count
2
P2P Supported
Yes
Efficiency
SM/CE copy efficiency
0.91

Memory Bandwidth#

Compression #

OpenSSL #

Geekbench Single-Core #

Score: 2,079

Geekbench Multi-Core #

Score: 17,415

Passmark CPU Scores #

BENCHMARKSCORE
Mark
54,314
Compression
716,947
Encryption
43,985
Extended Instructions
46,057
Floating Point Maths
142,147
Integer Maths
199,432
Physics
6,503
Prime Numbers
349
Single Threaded
2,720
String Sorting
97,370

Passmark Memory Scores #

BENCHMARKSCORE
Memory Mark
2,230
Database Operations
23,460
Memory Latency
74
Memory Read Cached
25,230
Memory Read Uncached
9,444
Memory Write
8,322

Stress-ng Stressors #

CPU integerSingleMultiScaling
atomic
213.07 127.92 0.60×
bitops
146,387 3,835,096 26.20×
branch
134,818 4,567,078 33.88×
funccall
733.71 20,479 27.91×
funcret
12,512 415,473 33.21×
hash
173,927 5,497,702 31.61×
intmath
63,200 1,635,639 25.88×
nop
276,104 4,912,657 17.79×
prime
495,294 12,873,891 25.99×
rotate
105,367 2,391,360 22.70×
str
962,051 30,203,757 31.40×
wcs
738,749 19,589,729 26.52×
CryptographySingleMultiScaling
crypt
3,299 69,088 20.94×
CompressionSingleMultiScaling
jpeg
779.57 23,593 30.26×
zlib
62.55 2,365 37.81×
HypervisorSingleMultiScaling
clock
38,359 398,441 10.39×
clone
406.71 6,378 15.68×
context
5,059 182,954 36.17×
cpu-sched
69.04 3,372 48.84×
eventfd
63,360 1,535,459 24.23×
fault
28,210 246,813 8.75×
fork
4,811 100,203 20.83×
getrandom
57,339 1,838,745 32.07×
hrtimers
299,066 5,991,144 20.03×
mprotect
1,083,351 13,493,323 12.46×
mtx
30,370,794 1,582,333,521 52.10×
munmap
4,045 96,539 23.87×
null
928,956 31,603,396 34.02×
pipe
658,781 42,066,008 63.85×
priv-instr
857,569 13,534,548 15.78×
resched
15,014,953 4,983,265 0.33×
switch
97,361 14,619,443 150.16×
timermix
234,164 8,947,336 38.21×
tlb-shootdown
9,352 78,451 8.39×
tsc
605,926 21,774,768 35.94×
usersyscall
732,000 13,935,670 19.04×
vdso
90,669,565 3,146,850,684 34.71×
vfork
33,688 240,866 7.15×
x86syscall
10,423,521 404,851,024 38.84×
yield
48,364,703 68,519,150 1.42×
zero
6,821,543 254,812,376 37.35×
CPU float/SIMDSingleMultiScaling
eigen
15,388 442,670 28.77×
expmath
3,475 109,208 31.43×
fma
1,241,960 41,174,203 33.15×
fp
282.09 10,242 36.31×
fractal
6.6 213.87 32.40×
hyperbolic
3,918 124,329 31.73×
logmath
5,505 187,662 34.09×
matrix
6,683 152,787 22.86×
matrix-3d
648.48 6,893 10.63×
monte-carlo
302.04 10,553 34.94×
mpfr
1,835 44,149 24.05×
powmath
2,046 68,198 33.34×
trig
2,769 83,252 30.06×
veccmp
27,252 685,696 25.16×
vecfp
2,402 63,355 26.37×
vecmath
4,146 135,824 32.76×
vecshuf
4,073 175,890 43.18×
vecwide
85,828 2,477,077 28.86×
AlgorithmsSingleMultiScaling
bitonicsort
8.6 228.59 26.58×
bsearch
270.05 8,344 30.90×
bubblesort
1 20.8 20.80×
fibsearch
301.76 7,684 25.46×
heapsort
9.6 284.39 29.62×
hsearch
8,372 221,492 26.46×
insertionsort
12.4 239.2 19.29×
malloc
2,813,251 57,847,066 20.56×
mcontend
78.52 8,502 108.28×
mergesort
32.2 980.18 30.44×
qsort
23 681.19 29.62×
skiplist
5,739 193,882 33.79×
tree
1.6 41.33 25.83×
MemorySingleMultiScaling
cache
297,520 1,533,198 5.15×
cachehammer
212,142 7,154,593 33.73×
cacheline
41.97 2,921 69.60×
l1cache
1,145 9,438 8.24×
memcpy
461.49 9,328 20.21×
memrate
0.17 7.58 44.59×
memthrash
255.5 1,390 5.44×
ptr-chase
26,187,472 439,057,785 16.77×
spinmem
251.59 2,354 9.36×
stream
14.23 9,910 696.41×
vm
133,604 299,530 2.24×

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
a3-highgpu-1g 26234 GiB1
a3-highgpu-4g 104936 GiB4
a3-highgpu-8g 2081872 GiB8
a3-megagpu-8g 2081872 GiB8
a3-edgegpu-8g 2081872 GiB8
a3-edgegpu-8g-nolssd 2081872 GiB8
a3-ultragpu-8g 2242952 GiB8

Similar Servers #

INSTANCEVENDORvCPUsMEMORYAccelerators

a3-highgpu-2g FAQs #