Hardware & Chipsets

Pretraining · full fine-tuning · LoRA / QLoRA · inference

Disclaimer: This site is for informational purposes only. Model rankings, usage data, pricing, capabilities and hardware results change frequently and may vary by benchmark, provider and workload. Verify current source data, licensing, security, privacy, compliance and deployment requirements before making production or purchasing decisions.

Accelerators

10

Server configurations

6

Measured inference results

160

Hardware vendors

4

CompareAcceleratorMemory / bandwidthPower envelopeInterconnectEvidence

NVIDIA · Hopper

80 GB

HBM3 · 3.35 TB/s

Up to 700 WNVLink 4 · 900 GB/s
Manufacturer specification

Verified 2026-10-11

NVIDIA · Hopper

141 GB

HBM3e · 4.8 TB/s

Up to 700 WNVLink 4 · 900 GB/s
Manufacturer specification

Verified 2026-10-11

NVIDIA · Blackwell

180 GB

HBM3e · 8 TB/s

Up to 1000 WNVLink 5 · 1.8 TB/s
Manufacturer specification

Verified 2026-10-11

AMD · CDNA 3

192 GB

HBM3 · 5.3 TB/s

Up to 750 WInfinity Fabric · PCIe 5.0 x16
Manufacturer specification

Verified 2026-10-11

AMD · CDNA 3

256 GB

HBM3e · 6 TB/s

Up to 1000 WInfinity Fabric · PCIe 5.0 x16
Manufacturer specification

Verified 2026-10-11

AMD · CDNA 4

288 GB

HBM3e · 8 TB/s

Up to 1000 WInfinity Fabric · PCIe 5.0
Manufacturer specification

Verified 2026-10-11

Google · TPU v5p

102 GB

HBM · 95 GiB · 2.765 TB/s

Not disclosedICI 1,200 GB/s bidirectional · 3D torus (cube+ slices)
Manufacturer specification

Verified 2026-10-11

AWS · Neuron v3

103.1 GB

HBM · 96 GiB · 2.9 TB/s

Not disclosedNeuronLink v3 · 1.28 TB/s
Manufacturer specification

Verified 2026-10-11

AWS · Neuron v2

34.4 GB

HBM · 32 GiB · 0.88 TB/s

Not disclosedNeuronLink v2
Manufacturer specification

Verified 2026-10-11

NVIDIA · Blackwell

186 GB

HBM3e · 372 GB per Superchip · 8 TB/s

Not disclosedNVLink 5 · 1.8 TB/s
Manufacturer specification

Verified 2026-10-11

GB and TB/s are decimal; source GiB values are converted and retained in the memory-type label. Prices are not verified. Capacity, precision support and software compatibility are independent constraints.

Workload sizing

Planning estimate · decimal GB

Estimated memory envelope

26.3 GB

Model weights
16.0 GB
Adapter weights
0.0 GB
Gradients + Adam states
0.0 GB
KV cache (BF16)
1.1 GB
Activation / runtime budget
4.0 GB
Reserved headroom
5.3 GB

1–1 × H100 SXM

Memory-only lower bound to a 25% contingency envelope; not a validated deployment configuration.

Weights are not the whole workload. Sharding, communication, quantization metadata and framework support can change fit; no throughput or time-to-train is inferred.

Training assumes BF16 trainable weights and gradients, FP32 master weights and Adam moments (16 bytes per trainable parameter including weights). Activation memory is your explicit budget, not an architecture prediction. QLoRA base weights use 4-bit storage plus the separately budgeted overhead.

Measured inference performance

MLPerf · closed division · datacenter · Offline · checked latest source check · 160 results

Compare the same workload, accuracy target, precision and scenario. Offline tokens/s measures batched throughput, not interactive latency or model quality. System sizes differ.

Workload / roundSystem / submitterAcceleratorsWeight precisionTokens/sEvidence
deepseek-r1

v6.0 · Offline · closed

NVIDIA GB300 NVL72 (72x GB300-288GB_aarch64, TensorRT)

NVIDIA

NVIDIA GB300fp42,494,310
deepseek-r1

v6.0 · Offline · closed

Nebius GB300 NVL72 (72x GB300-288GB_aarch64, TensorRT)

Nebius

NVIDIA GB300fp4673,936
deepseek-r1

v6.0 · Offline · closed

NVIDIA GB300 NVL72 (72x GB300-288GB_aarch64, TensorRT)

NVIDIA

NVIDIA GB300fp4647,076
deepseek-r1

v6.0 · Offline · closed

NVIDIA GB200 NVL72 (72x GB200-186GB_aarch64, TensorRT)

NVIDIA

NVIDIA GB200fp4486,141
deepseek-r1

v6.0 · Offline · closed

CoreWeave GB200 NVL72 (64x GB200-186GB_aarch64, TensorRT)

CoreWeave

NVIDIA GB200fp4468,665
deepseek-r1

v6.0 · Offline · closed

NVIDIA GB200 NVL72 (72x GB200-186GB_aarch64, TensorRT)

Google

NVIDIA GB200fp4307,874
deepseek-r1

v6.0 · Offline · closed

CoreWeave GB300 NVL72 (8x GB300-288GB_aarch64, TensorRT)

CoreWeave

NVIDIA GB300fp478,564
deepseek-r1

v6.0 · Offline · closed

Nebius GB300 NVL72 (4x GB300-288GB_aarch64, TensorRT)

Nebius

NVIDIA GB300fp476,346.6
deepseek-r1

v6.0 · Offline · closed

NVIDIA DGX B300 (8x B300-SXM-270GB, TensorRT)

NVIDIA

NVIDIA B300-SXM-270GBfp470,325.7
deepseek-r1

v6.0 · Offline · closed

G894-SD3-AAX7

GigaComputing

NVIDIA B300-SXM-270GBfp469,617.4
deepseek-r1

v6.0 · Offline · closed

Nebius B300 n1 (8x B300-SXM-270GB, TensorRT)

Nebius

NVIDIA B300-SXM-270GBfp469,318.9
deepseek-r1

v6.0 · Offline · closed

QuantaGrid D75H-10U (8x B300-SXM-270GB, TensorRT)

Quanta_Cloud_Technology

NVIDIA B300-SXM-270GBfp469,263.2
deepseek-r1

v6.0 · Offline · closed

Cisco UCS C880A M8 (8x NVIDIA B300-SXM-270GB, TensorRT)

Cisco

NVIDIA B300-SXM-270GBfp469,251.4
deepseek-r1

v6.0 · Offline · closed

PowerEdge XE9780 (8x B300-SXM-270GB, TensorRT)

Dell

NVIDIA B300-SXM-270GBfp469,021.2
deepseek-r1

v6.0 · Offline · closed

PowerEdge XE9780L (8x B300-SXM-270GB, TensorRT)

Dell

NVIDIA B300-SXM-270GBfp466,287.9
deepseek-r1

v6.0 · Offline · closed

Nebius B200 n1 (8x B200-SXM-180GB, TensorRT)

Nebius

NVIDIA B200-SXM-180GBfp458,581.6
deepseek-r1

v6.0 · Offline · closed

PowerEdge XE9780 (8x B200-SXM-180GB, TensorRT)

Dell

NVIDIA B200-SXM-180GBfp458,235.4
deepseek-r1

v6.0 · Offline · closed

NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)

Google

NVIDIA B200-SXM-180GBfp457,116.9
gpt-oss-120b

v6.0 · Offline · closed

Nebius GB300 NVL72 (72x GB300-288GB_aarch64, TensorRT)

Nebius

NVIDIA GB300fp41,046,150
gpt-oss-120b

v6.0 · Offline · closed

NVIDIA GB300 NVL72 (72x GB300-288GB_aarch64, TensorRT)

NVIDIA

NVIDIA GB300fp41,042,980
gpt-oss-120b

v6.0 · Offline · closed

AMD MI355X Custom Cluster

AMD

AMD Instinct MI355X 288GB HBM3e (x94)fp41,031,070
gpt-oss-120b

v6.0 · Offline · closed

CoreWeave GB300 NVL72 (64x GB300-288GB_aarch64, TensorRT)

CoreWeave

NVIDIA GB300fp4910,468
gpt-oss-120b

v6.0 · Offline · closed

NVIDIA GB200 NVL72 (72x GB200-186GB_aarch64, TensorRT)

NVIDIA

NVIDIA GB200fp4879,542
gpt-oss-120b

v6.0 · Offline · closed

CoreWeave GB200 NVL72 (64x GB200-186GB_aarch64, TensorRT)

CoreWeave

NVIDIA GB200fp4774,539
gpt-oss-120b

v6.0 · Offline · closed

Cisco UCS C880A M8 (8x NVIDIA B300-SXM-270GB, TensorRT)

Cisco

NVIDIA B300-SXM-270GBfp4111,496
gpt-oss-120b

v6.0 · Offline · closed

XA NB3I-E12

ASUSTeK

NVIDIA B300-SXM-270GBfp4110,077
gpt-oss-120b

v6.0 · Offline · closed

ThinkSystem SR680a V4 (8x B300-SXM-270GB, TensorRT)

Lenovo

NVIDIA B300-SXM-270GBfp4109,141
gpt-oss-120b

v6.0 · Offline · closed

G894-SD3-AAX7

GigaComputing

NVIDIA B300-SXM-270GBfp4107,696
gpt-oss-120b

v6.0 · Offline · closed

Nebius B300 n1 (8x B300-SXM-270GB, TensorRT)

Nebius

NVIDIA B300-SXM-270GBfp4106,885
gpt-oss-120b

v6.0 · Offline · closed

NVIDIA DGX B300 (8x B300-SXM-270GB, TensorRT)

NVIDIA

NVIDIA B300-SXM-270GBfp4103,961
gpt-oss-120b

v6.0 · Offline · closed

smci355-ccs-aus-m09-17 (AS -4126GS-NMR-LCC)

AMD

AMD Instinct MI355X 288GB HBM3efp495,004
gpt-oss-120b

v6.0 · Offline · closed

LLM-D v0.5.0,Openshift 4.20.12,NVIDIA 8xB200-SXM-180GB

RedHat

NVIDIA B200-SXM-180GBmxfp493,070.7
gpt-oss-120b

v6.0 · Offline · closed

G893-ZX1-AAX4

GigaComputing

AMD Instinct MI355X 288GB HBM3efp492,332.2
gpt-oss-120b

v6.0 · Offline · closed

HPE ProLiant Compute XD685 (8x AMD Instinct MI355X 288GB, vLLM)

HPE

AMD Instinct MI355X 288GB HBM3efp491,846.6
gpt-oss-120b

v6.0 · Offline · closed

BM.GPU.MI355X.8

ORACLE

AMD Instinct MI355X 288GB HBM3efp491,706.3
gpt-oss-120b

v6.0 · Offline · closed

PowerEdge XE9785L (8x MI355X)

Dell

AMD Instinct MI355X 288GB HBM3efp491,395.5
gpt-oss-120b

v6.0 · Offline · closed

Nebius B200 n1 (8x B200-SXM-180GB, TensorRT)

Nebius

NVIDIA B200-SXM-180GBfp485,921.2
gpt-oss-120b

v6.0 · Offline · closed

NVIDIA GB300 NVL72 (4x GB300-288GB_aarch64, TensorRT)

Lambda_SIT

NVIDIA GB300fp460,220.4
gpt-oss-120b

v6.0 · Offline · closed

LLM-D v0.5.0,OpenShift 4.21,IBM cloud gx3d-160x1792x8h200,NVIDIA 8xH200-SXM-141GB

RedHat

NVIDIA H200-SXM-141GBmxfp428,680
gpt-oss-120b

v6.0 · Offline · closed

HPE ProLiant Compute DL380a Gen12 (10x NVIDIA RTX PRO 6000 Blackwell Server Edition, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp419,040.8
gpt-oss-120b

v6.0 · Offline · closed

Nebius GB300 NVL72 (4x GB300-288GB_aarch64, TensorRT)

Nebius

NVIDIA GB300fp415,371.7
gpt-oss-120b

v6.0 · Offline · closed

HPE ProLiant Compute DL380a Gen12 (8x NVIDIA RTX PRO 6000 Blackwell Server Edition, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp415,189.9
gpt-oss-120b

v6.0 · Offline · closed

HPE ProLiant DL385 Gen11 (4x NVIDIA RTX PRO 6000 Blackwell Server Edition-450W, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp47,297.05
gpt-oss-120b

v6.0 · Offline · closed

1-node-4x-BMG-Pro-B60-Dual

Intel

MS-Intel Arc Pro B60 Dual 48G TurboMXFP41,601.91
gpt-oss-120b

v6.0 · Offline · closed

1-node-4x-BMG-B70

Intel

Intel(R) Arc Pro(R) B70MXFP41,536.9
gpt-oss-120b

v6.0 · Offline · closed

1-node-4x-BMG-B60

Intel

Intel(R) Arc Pro(R) B60MXFP4841.038
llama2-70b-99

v6.0 · Offline · closed

NVIDIA GB300 NVL72 (72x GB300-288GB_aarch64, TensorRT)

NVIDIA

NVIDIA GB300fp41,126,850
llama2-70b-99

v6.0 · Offline · closed

AMD MI355X Custom Cluster

AMD

AMD Instinct MI355X 288GB HBM3e (x87)fp41,042,110
llama2-70b-99

v6.0 · Offline · closed

NVIDIA GB200 NVL72 (72x GB200-186GB_aarch64, TensorRT)

NVIDIA

NVIDIA GB200fp4888,054
llama2-70b-99

v6.0 · Offline · closed

Cisco UCS 16xMI350X

Cisco

AMD Instinct MI350X 288GB HBM3efp4159,932
llama2-70b-99

v6.0 · Offline · closed

Dell_MangoBoost Heterogeneous Cluster (8x MI300X + 8x MI325X + 8x MI355X, LLMBoost)

Dell_MangoBoost

AMD Instinct MI300X 192GB HBM3 (x8) and AMD Instinct MI325X 256GB HBM3e (x8) and AMD Instinct MI355X 288GB HBM3e (x8)fp8151,843
llama2-70b-99

v6.0 · Offline · closed

Cisco UCS C880A M8 (8x NVIDIA B300-SXM-270GB, TensorRT)

Cisco

NVIDIA B300-SXM-270GBfp4115,730
llama2-70b-99

v6.0 · Offline · closed

NVIDIA DGX B300 (8x B300-SXM-270GB, TensorRT)

NVIDIA

NVIDIA B300-SXM-270GBfp4112,954
llama2-70b-99

v6.0 · Offline · closed

XA NB3I-E12

ASUSTeK

NVIDIA B300-SXM-270GBfp4109,625
llama2-70b-99

v6.0 · Offline · closed

G894-SD3-AAX7

GigaComputing

NVIDIA B300-SXM-270GBfp4109,343
llama2-70b-99

v6.0 · Offline · closed

PowerEdge XE9780L (8x B300-SXM-270GB, TensorRT)

Dell

NVIDIA B300-SXM-270GBfp4108,669
llama2-70b-99

v6.0 · Offline · closed

HPE ProLiant Compute XD685 (8x NVIDIA B200 180GB, TensorRT)

HPE

NVIDIA B200-SXM-180GBfp4103,954
llama2-70b-99

v6.0 · Offline · closed

smci355-ccs-aus-m09-17 (AS -4126GS-NMR-LCC)

AMD

AMD Instinct MI355X 288GB HBM3efp4103,480
llama2-70b-99

v6.0 · Offline · closed

SYS-822GS-NBRT (8x B200-SXM-180GB, TensorRT)

Supermicro

NVIDIA B200-SXM-180GBfp4101,595
llama2-70b-99

v6.0 · Offline · closed

QuantaGrid D74A-7U

Supermicro

AMD Instinct MI355X 288GB HBM3efp4100,754
llama2-70b-99

v6.0 · Offline · closed

G893-ZX1-AAX4

GigaComputing

AMD Instinct MI355X 288GB HBM3efp4100,310
llama2-70b-99

v6.0 · Offline · closed

HPE ProLiant Compute XD685 (8x AMD Instinct MI355X 288GB, vLLM)

HPE

AMD Instinct MI355X 288GB HBM3efp4100,019
llama2-70b-99

v6.0 · Offline · closed

PowerEdge XE9785L (8x MI355X)

Dell

AMD Instinct MI355X 288GB HBM3efp4100,008
llama2-70b-99

v6.0 · Offline · closed

BM.GPU.MI355X.8

ORACLE

AMD Instinct MI355X 288GB HBM3efp499,874.3
llama2-70b-99

v6.0 · Offline · closed

PowerEdge XE9785

Dell_MangoBoost

AMD Instinct MI355X 288GB HBM3efp897,535.1
llama2-70b-99

v6.0 · Offline · closed

PowerEdge XE9785 (MI355X Power Cap 1000W)

Dell_MangoBoost

AMD Instinct MI355X 288GB HBM3e (Power Cap 1000 W)fp881,870.5
llama2-70b-99

v6.0 · Offline · closed

AS-8126GS-TNMR,vLLM

Supermicro_RedHat

AMD Instinct MI350X 288GB HBM3efp480,478.1
llama2-70b-99

v6.0 · Offline · closed

G8825Z5

MiTAC

AMD Instinct MI350X 288GB HBM3efp478,988.6
llama2-70b-99

v6.0 · Offline · closed

Cisco UCS 8xMI350X

Cisco

AMD Instinct MI350X 288GB HBM3efp478,936.7
llama2-70b-99

v6.0 · Offline · closed

HPE ProLiant Compute DL380a Gen12 (10x NVIDIA RTX PRO 6000 Blackwell Server Edition, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp436,710.8
llama2-70b-99

v6.0 · Offline · closed

MangoBoost Mi325X (8x MI325X, LLMBoost)

MangoBoost

AMD Instinct MI325X 256GB HBM3efp833,880.4
llama2-70b-99

v6.0 · Offline · closed

PowerEdge XE7740 (8x H200-NVL-141GB, TensorRT)

Dell

NVIDIA H200-NVL-141GBfp832,004
llama2-70b-99

v6.0 · Offline · closed

HPE ProLiant Compute DL380a Gen12 (8x NVIDIA RTX PRO 6000 Blackwell Server Edition, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp429,908.4
llama2-70b-99

v6.0 · Offline · closed

ESC8000A_E13P_RTX_PRO_6000_PCIE_96GBx8

ASUSTeK

NVIDIA RTX PRO 6000 Blackwell Server Editionfp427,812.9
llama2-70b-99

v6.0 · Offline · closed

ESC8000_E12P_RTX_PRO_6000_PCIE_96GBx8

ASUSTeK

NVIDIA RTX PRO 6000 Blackwell Server Editionfp427,730.1
llama2-70b-99

v6.0 · Offline · closed

MangoBoost Mi300X (8x MI300X, LLMBoost)

MangoBoost

AMD Instinct MI300X 192GB HBM3fp827,598.7
llama2-70b-99

v6.0 · Offline · closed

PowerEdge XE9680 (8x MI300X)

Dell

AMD Instinct MI300X 192GB HBM3fp827,072.5
llama2-70b-99

v6.0 · Offline · closed

PowerEdge XE7740 (8x RTXPro6000-PCIe-96GB, TensorRT)

Dell

NVIDIA RTX PRO 6000 Blackwell Server Editionfp427,034.8
llama2-70b-99

v6.0 · Offline · closed

ThinkSystem SR675i V3 (8x NVIDIA RTX PRO 6000 Blackwell Server Edition, TensorRT)

Lenovo

NVIDIA RTX PRO 6000 Blackwell Server Editionfp426,889.6
llama2-70b-99

v6.0 · Offline · closed

PowerEdge XE7745 (8x RTXPro6000-PCIe-96GB, TensorRT)

Dell

NVIDIA RTX PRO 6000 Blackwell Server Editionfp426,829.8
llama2-70b-99

v6.0 · Offline · closed

Cisco UCS X215c M8 - X580P PCIe Node (4x RTXPro6000-PCIe-96GB, TensorRT)

Cisco

NVIDIA RTX PRO 6000 Blackwell Server Editionfp413,529.8
llama2-70b-99

v6.0 · Offline · closed

HPE ProLiant DL385 Gen11 (4x NVIDIA RTX PRO 6000 Blackwell Server Edition-450W, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp413,361.9
llama2-70b-99

v6.0 · Offline · closed

HPE ProLiant DL385 Gen11 (2x NVIDIA RTX PRO 6000 Blackwell Server Edition-600W, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp46,901.9
llama2-70b-99

v6.0 · Offline · closed

HPE ProLiant DL385 Gen11 (2x NVIDIA RTX PRO 6000 Blackwell Server Edition-600W, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp46,862.3
llama2-70b-99

v6.0 · Offline · closed

1-node-4x-BMG-Pro-B60-Dual

Intel

MS-Intel Arc Pro B60 Dual 48G TurboINT83,270.66
llama2-70b-99

v6.0 · Offline · closed

1-node-4x-BMG-B70

Intel

Intel(R) Arc Pro(R) B70INT82,459.18
llama2-70b-99

v6.0 · Offline · closed

1-node-4x-BMG-B60

Intel

Intel(R) Arc Pro(R) B60INT81,697.66
llama2-70b-99.9

v6.0 · Offline · closed

NVIDIA GB300 NVL72 (72x GB300-288GB_aarch64, TensorRT)

NVIDIA

NVIDIA GB300fp41,126,850
llama2-70b-99.9

v6.0 · Offline · closed

AMD MI355X Custom Cluster

AMD

AMD Instinct MI355X 288GB HBM3e (x87)fp41,042,110
llama2-70b-99.9

v6.0 · Offline · closed

NVIDIA GB200 NVL72 (72x GB200-186GB_aarch64, TensorRT)

NVIDIA

NVIDIA GB200fp4888,054
llama2-70b-99.9

v6.0 · Offline · closed

Dell_MangoBoost Heterogeneous Cluster (8x MI300X + 8x MI325X + 8x MI355X, LLMBoost)

Dell_MangoBoost

AMD Instinct MI300X 192GB HBM3 (x8) and AMD Instinct MI325X 256GB HBM3e (x8) and AMD Instinct MI355X 288GB HBM3e (x8)fp8151,843
llama2-70b-99.9

v6.0 · Offline · closed

NVIDIA DGX B300 (8x B300-SXM-270GB, TensorRT)

NVIDIA

NVIDIA B300-SXM-270GBfp4112,954
llama2-70b-99.9

v6.0 · Offline · closed

G894-SD3-AAX7

GigaComputing

NVIDIA B300-SXM-270GBfp4109,343
llama2-70b-99.9

v6.0 · Offline · closed

PowerEdge XE9780L (8x B300-SXM-270GB, TensorRT)

Dell

NVIDIA B300-SXM-270GBfp4108,669
llama2-70b-99.9

v6.0 · Offline · closed

HPE ProLiant Compute XD685 (8x NVIDIA B200 180GB, TensorRT)

HPE

NVIDIA B200-SXM-180GBfp4104,572
llama2-70b-99.9

v6.0 · Offline · closed

smci355-ccs-aus-m09-17 (AS -4126GS-NMR-LCC)

AMD

AMD Instinct MI355X 288GB HBM3efp4103,480
llama2-70b-99.9

v6.0 · Offline · closed

SYS-822GS-NBRT (8x B200-SXM-180GB, TensorRT)

Supermicro

NVIDIA B200-SXM-180GBfp4101,526
llama2-70b-99.9

v6.0 · Offline · closed

QuantaGrid D74A-7U

Supermicro

AMD Instinct MI355X 288GB HBM3efp4100,754
llama2-70b-99.9

v6.0 · Offline · closed

G893-ZX1-AAX4

GigaComputing

AMD Instinct MI355X 288GB HBM3efp4100,310
llama2-70b-99.9

v6.0 · Offline · closed

HPE ProLiant Compute XD685 (8x AMD Instinct MI355X 288GB, vLLM)

HPE

AMD Instinct MI355X 288GB HBM3efp4100,019
llama2-70b-99.9

v6.0 · Offline · closed

PowerEdge XE9785L (8x MI355X)

Dell

AMD Instinct MI355X 288GB HBM3efp4100,008
llama2-70b-99.9

v6.0 · Offline · closed

BM.GPU.MI355X.8

ORACLE

AMD Instinct MI355X 288GB HBM3efp499,874.3
llama2-70b-99.9

v6.0 · Offline · closed

PowerEdge XE9785

Dell_MangoBoost

AMD Instinct MI355X 288GB HBM3efp897,535.1
llama2-70b-99.9

v6.0 · Offline · closed

PowerEdge XE9785 (MI355X Power Cap 1000W)

Dell_MangoBoost

AMD Instinct MI355X 288GB HBM3e (Power Cap 1000 W)fp881,870.5
llama2-70b-99.9

v6.0 · Offline · closed

G8825Z5

MiTAC

AMD Instinct MI350X 288GB HBM3efp478,988.6
llama2-70b-99.9

v6.0 · Offline · closed

HPE ProLiant Compute DL380a Gen12 (10x NVIDIA RTX PRO 6000 Blackwell Server Edition, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp436,704.2
llama2-70b-99.9

v6.0 · Offline · closed

MangoBoost Mi325X (8x MI325X, LLMBoost)

MangoBoost

AMD Instinct MI325X 256GB HBM3efp833,880.4
llama2-70b-99.9

v6.0 · Offline · closed

PowerEdge XE7740 (8x H200-NVL-141GB, TensorRT)

Dell

NVIDIA H200-NVL-141GBfp832,004
llama2-70b-99.9

v6.0 · Offline · closed

HPE ProLiant Compute DL380a Gen12 (8x NVIDIA RTX PRO 6000 Blackwell Server Edition, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp429,900.1
llama2-70b-99.9

v6.0 · Offline · closed

MangoBoost Mi300X (8x MI300X, LLMBoost)

MangoBoost

AMD Instinct MI300X 192GB HBM3fp827,598.7
llama2-70b-99.9

v6.0 · Offline · closed

PowerEdge XE9680 (8x MI300X)

Dell

AMD Instinct MI300X 192GB HBM3fp827,072.5
llama2-70b-99.9

v6.0 · Offline · closed

PowerEdge XE7740 (8x RTXPro6000-PCIe-96GB, TensorRT)

Dell

NVIDIA RTX PRO 6000 Blackwell Server Editionfp427,034.8
llama2-70b-99.9

v6.0 · Offline · closed

ThinkSystem SR675i V3 (8x NVIDIA RTX PRO 6000 Blackwell Server Edition, TensorRT)

Lenovo

NVIDIA RTX PRO 6000 Blackwell Server Editionfp426,889.6
llama2-70b-99.9

v6.0 · Offline · closed

PowerEdge XE7745 (8x RTXPro6000-PCIe-96GB, TensorRT)

Dell

NVIDIA RTX PRO 6000 Blackwell Server Editionfp426,829.8
llama2-70b-99.9

v6.0 · Offline · closed

HPE ProLiant DL385 Gen11 (4x NVIDIA RTX PRO 6000 Blackwell Server Edition-450W, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp413,200.3
llama2-70b-99.9

v6.0 · Offline · closed

HPE ProLiant DL385 Gen11 (2x NVIDIA RTX PRO 6000 Blackwell Server Edition-600W, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp46,893.36
llama2-70b-99.9

v6.0 · Offline · closed

1-node-4x-BMG-Pro-B60-Dual

Intel

MS-Intel Arc Pro B60 Dual 48G TurboINT83,270.66
llama2-70b-99.9

v6.0 · Offline · closed

1-node-4x-BMG-B70

Intel

Intel(R) Arc Pro(R) B70INT82,459.18
llama2-70b-99.9

v6.0 · Offline · closed

1-node-4x-BMG-B60

Intel

Intel(R) Arc Pro(R) B60INT81,697.66
llama3.1-405b

v6.0 · Offline · closed

NVIDIA GB300 NVL72 (72x GB300-288GB_aarch64, TensorRT)

NVIDIA

NVIDIA GB300fp419,511.7
llama3.1-405b

v6.0 · Offline · closed

NVIDIA GB200 NVL72 (72x GB200-186GB_aarch64, TensorRT)

NVIDIA

NVIDIA GB200fp415,461.5
llama3.1-405b

v6.0 · Offline · closed

Cisco UCS C880A M8 (8x NVIDIA B300-SXM-270GB, TensorRT)

Cisco

NVIDIA B300-SXM-270GBfp41,971.11
llama3.1-405b

v6.0 · Offline · closed

QuantaGrid D75H-10U (8x B300-SXM-270GB, TensorRT)

Quanta_Cloud_Technology

NVIDIA B300-SXM-270GBfp41,953.77
llama3.1-405b

v6.0 · Offline · closed

NVIDIA DGX B300 (8x B300-SXM-270GB, TensorRT)

NVIDIA

NVIDIA B300-SXM-270GBfp41,951.6
llama3.1-405b

v6.0 · Offline · closed

G894-SD3-AAX7

GigaComputing

NVIDIA B300-SXM-270GBfp41,947.91
llama3.1-405b

v6.0 · Offline · closed

NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)

Google

NVIDIA B200-SXM-180GBfp41,350.24
llama3.1-8b

v6.0 · Offline · closed

XA NB3I-E12

ASUSTeK

NVIDIA B300-SXM-270GBfp4166,745
llama3.1-8b

v6.0 · Offline · closed

NVIDIA DGX B300 (8x B300-SXM-270GB, TensorRT)

NVIDIA

NVIDIA B300-SXM-270GBfp4165,432
llama3.1-8b

v6.0 · Offline · closed

NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)

Lambda_SIT

NVIDIA B200-SXM-180GBfp4160,403
llama3.1-8b

v6.0 · Offline · closed

HPE ProLiant Compute XD685 (8x NVIDIA B200 180GB, TensorRT)

HPE

NVIDIA B200-SXM-180GBfp4152,625
llama3.1-8b

v6.0 · Offline · closed

G894-SD3-AAX7

GigaComputing

NVIDIA B300-SXM-270GBfp4150,707
llama3.1-8b

v6.0 · Offline · closed

HPE ProLiant Compute DL380a Gen12 (10x NVIDIA RTX PRO 6000 Blackwell Server Edition, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp460,679.9
llama3.1-8b

v6.0 · Offline · closed

PowerEdge XE7740 (8x H200-NVL-141GB, TensorRT)

Dell

NVIDIA H200-NVL-141GBfp853,193.6
llama3.1-8b

v6.0 · Offline · closed

HPE ProLiant Compute DL380a Gen12 (8x NVIDIA RTX PRO 6000 Blackwell Server Edition, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp449,580.2
llama3.1-8b

v6.0 · Offline · closed

PowerEdge XE7745 (8x RTXPro6000-PCIe-96GB, TensorRT)

Dell

NVIDIA RTX PRO 6000 Blackwell Server Editionfp449,036.1
llama3.1-8b

v6.0 · Offline · closed

ESC8000A_E13P_RTX_PRO_6000_PCIE_96GBx8

ASUSTeK

NVIDIA RTX PRO 6000 Blackwell Server Editionfp448,791
llama3.1-8b

v6.0 · Offline · closed

ESC8000_E12P_RTX_PRO_6000_PCIE_96GBx8

ASUSTeK

NVIDIA RTX PRO 6000 Blackwell Server Editionfp448,613.8
llama3.1-8b

v6.0 · Offline · closed

Cisco UCS X210 M8 - X580P PCIe Node (4x H200-NVL-141GB, TensorRT)

Cisco

NVIDIA H200-NVL-141GBfp824,216.6
llama3.1-8b

v6.0 · Offline · closed

Cisco UCS X215c M8 - X580P PCIe Node (4x RTXPro6000-PCIe-96GB, TensorRT)

Cisco

NVIDIA RTX PRO 6000 Blackwell Server Editionfp424,013.5
llama3.1-8b

v6.0 · Offline · closed

HPE ProLiant DL385 Gen11 (4x NVIDIA RTX PRO 6000 Blackwell Server Edition-450W, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp423,556.7
llama3.1-8b

v6.0 · Offline · closed

HPE ProLiant DL385 Gen11 (4x NVIDIA RTX PRO 6000 Blackwell Server Edition-450W, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp423,265.7
llama3.1-8b

v6.0 · Offline · closed

HPE ProLiant DL385 Gen11 (2x NVIDIA RTX PRO 6000 Blackwell Server Edition-600W, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp412,495.2
llama3.1-8b

v6.0 · Offline · closed

HPE ProLiant DL385 Gen11 (2x NVIDIA RTX PRO 6000 Blackwell Server Edition-600W, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp412,433.3
llama3.1-8b

v6.0 · Offline · closed

HPE ProLiant Compute DL380 Gen12 (5x NVIDIA RTX PRO 4500 Blackwell, TensorRT)

HPE

NVIDIA RTX PRO 4500 Blackwellfp411,665
llama3.1-8b

v6.0 · Offline · closed

1-node-4x-BMG-Pro-B60-Dual

Intel

MS-Intel Arc Pro B60 Dual 48G TurboINT86,762.3
llama3.1-8b

v6.0 · Offline · closed

1-node-4x-BMG-B70

Intel

Intel(R) Arc Pro(R) B70INT84,617.4
llama3.1-8b

v6.0 · Offline · closed

1-node-4x-BMG-B60

Intel

Intel(R) Arc Pro(R) B60INT83,347.43
llama3.1-8b

v6.0 · Offline · closed

HPE ProLiant DL145 Gen11 (1x NVIDIA RTX PRO 4500 Blackwell, TensorRT)

HPE

NVIDIA RTX PRO 4500 Blackwellfp42,340.75
llama3.1-8b

v6.0 · Offline · closed

1-node-4x-BMG-B50

Intel

Intel(R) Arc Pro(R) B50INT81,721.84
llama3.1-8b

v6.0 · Offline · closed

1-node-2S-GNR_128C

Intel

INT81,229.56
llama3.1-8b

v6.0 · Offline · closed

1-node-2S-GNR_128C

Supermicro

INT81,185.98
llama3.1-8b

v6.0 · Offline · closed

1-node-2S-GNR_86C

Dell

INT8807.675
llama3.1-8b

v6.0 · Offline · closed

C240M8-1-node-2S-GNR_86C

Cisco

INT8806.105
llama3.1-8b

v6.0 · Offline · closed

1-node-2S-GNR_86C

Quanta_Cloud_Technology

INT8777.671
llama3.1-8b

v6.0 · Offline · closed

1-node-2S-GNR_86C

Lenovo

INT8775.982
mixtral-8x7b

v6.0 · Offline · closed

HPE ProLiant Compute XD685 (8x NVIDIA B200 180GB, TensorRT)

HPE

NVIDIA B200-SXM-180GBfp4126,468
mixtral-8x7b

v6.0 · Offline · closed

HPE ProLiant Compute DL380a Gen12 (10x NVIDIA RTX PRO 6000 Blackwell Server Edition, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp841,242.4
mixtral-8x7b

v6.0 · Offline · closed

HPE ProLiant Compute DL380a Gen12 (8x NVIDIA RTX PRO 6000 Blackwell Server Edition, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp833,637.3
mixtral-8x7b

v6.0 · Offline · closed

HPE ProLiant DL385 Gen11 (4x NVIDIA RTX PRO 6000 Blackwell Server Edition-450W, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp814,828.3
mixtral-8x7b

v6.0 · Offline · closed

HPE ProLiant DL385 Gen11 (2x NVIDIA RTX PRO 6000 Blackwell Server Edition-600W, TensorRT)

HPE

NVIDIA RTX PRO 6000 Blackwell Server Editionfp88,198.25

MLPerf Training is currently reference-only: time-to-target is tied to prescribed tasks and cannot predict arbitrary LLM pretraining or fine-tuning duration.

Reported model–hardware relationships

Llama 3.1 405B → H100 SXM

Meta reports training on up to 16,384 H100 GPUs. Applies to this model version, not all Llama models.

The Llama 3 Herd of Models · reported training · verified 2026-10-11

Undisclosed training, fine-tuning and production-serving hardware remain unknown. A measured MLPerf workload is not evidence of a model’s original training hardware.

Sources & verification

Refresh checks machine-readable MLPerf results; it does not automatically reverify curated product sheets or overwrite their verification dates.