Back to search

GPU Workload Scorer

Not all GPUs are equal for every task. We score every GPU in our catalog across five real AI workloads using weighted scoring on core specs — VRAM, FP16 TFLOPS, memory bandwidth, and TDP.

Training weights VRAM and compute highest. Inference prioritizes memory bandwidth and throughput. Edge deployment optimizes for low power. Select a workload to see the rankings.

Choose a workload profile

LLM Training ranking

RankGPUScoreVRAMFP16Mem BWTDPTags
#1
NVIDIA A100
Ampere
Excellent
97%
80 GB3122039 GB/s400 W
datacentertraininginference
#2
AMD Instinct MI250X
CDNA 2
Excellent
95%
128 GB3833200 GB/s560 W
datacentertraininginference
#3
NVIDIA H100
Hopper
Excellent
95%
80 GB4953350 GB/s700 W
datacentertraininginference
#4
NVIDIA H200
Hopper
Excellent
95%
141 GB4954800 GB/s700 W
datacentertraininginference
#5
NVIDIA H200 NVL
Hopper
Excellent
95%
141 GB4184800 GB/s600 W
datacentertraininginference
#6
NVIDIA B200
Blackwell
Excellent
95%
192 GB11258000 GB/s1000 W
datacentertraininginference
#7
NVIDIA B300
Blackwell Ultra
Excellent
95%
288 GB11008000 GB/s1300 W
datacentertraininginference
#8
NVIDIA GB200
Blackwell
Excellent
95%
192 GB11258000 GB/s1200 W
datacentertraininginference
#9
NVIDIA GH200
Hopper
Excellent
95%
96 GB4954000 GB/s1000 W
datacentertraininginference
#10
AMD Instinct MI300X
CDNA 3
Excellent
95%
192 GB13075300 GB/s750 W
datacentertraininginference
#11
NVIDIA RTX 6000 Ada
Ada Lovelace
Good
72%
48 GB365960 GB/s300 W
workstationinferencetraining
#12
NVIDIA L40S
Ada Lovelace
Good
71%
48 GB366864 GB/s350 W
datacenterinferencegraphics
#13
NVIDIA RTX A6000
Ampere
Good
63%
48 GB155768 GB/s300 W
workstationinferencetraining
#14
NVIDIA A40
Ampere
Good
61%
48 GB150696 GB/s300 W
datacenterinferencegraphics
#15
NVIDIA A16
Ampere
Fair
53%
64 GB36928 GB/s250 W
datacentervdiinference