readygpu

Lenovo ThinkStation PGX GB10 128GB vs NVIDIA DGX Spark 128GB

For running models locally, priced in Germany. At 128k context the first runs 80 models fully in memory, the second 80.

In short

Both have 128 GB of memory for models; the Lenovo ThinkStation PGX GB10 128GB has 273 GB/s of bandwidth and the NVIDIA DGX Spark 128GB 273 GB/s. At 128k context both run the same 80 models we track fully in memory. On the models both run fully in memory, the Lenovo ThinkStation PGX GB10 128GB and the NVIDIA DGX Spark 128GB are about as fast as each other (estimated).

Updated . Speeds are estimates; how they are worked out.

Lenovo ThinkStation PGX GB10 128GB

Not priced here

Add to build
NVIDIA DGX Spark 128GB

9.877 €

eBay

Add to build

Specifications

SpecLenovo ThinkStation PGX GB10 128GBNVIDIA DGX Spark 128GB
ChipGB10 Grace BlackwellGB10 Grace Blackwell
Memory bus256-bit256-bit
Memory bandwidth273 GB/s273 GB/s
FP16 compute62 TFLOPS62 TFLOPS
SoftwareCUDACUDA
Unified memory128 GB128 GB
GPU can use by default119 GB119 GB
GPU can use, tuned122 GB122 GB
Operating systemdgx-osdgx-os
Processor20-core Arm (GB10)20-core Arm (GB10)
Storage1000 GB4000 GB

What each runs

Estimated decode speed with the most precise quant that fits each one est.

ModelLenovo ThinkStation PGX GB10 128GBNVIDIA DGX Spark 128GB
DeepSeek V4 Flash6tok/sUD-Q2_K_XL6tok/sUD-Q2_K_XL
GLM 5.3 Flash10tok/sUD-Q2_K_XL10tok/sUD-Q2_K_XL
Qwen 3.8 Flash Next2tok/sQ4_K_M2tok/sQ4_K_M
Qwen 3.8 27B5tok/sQ8_05tok/sQ8_0
MiniMax M2.73tok/sUD-Q2_K_XL3tok/sUD-Q2_K_XL
Qwen 3.5 122B-A10B9tok/sQ6_K9tok/sQ6_K
Qwen 3.6 27B5tok/sQ8_05tok/sQ8_0
Qwen3 Coder Next14tok/sQ8_014tok/sQ8_0
gpt-oss 120B13tok/sMXFP413tok/sMXFP4
Nemotron 3 Super 120B-A12B8tok/sUD-Q6_K8tok/sUD-Q6_K
Gemma 4 31B4tok/sQ8_04tok/sQ8_0
Mistral Small 4 119B11tok/sUD-Q6_K11tok/sUD-Q6_K
Qwen 3.6 35B-A3B17tok/sQ8_017tok/sQ8_0
Qwen 3.5 27B5tok/sQ8_05tok/sQ8_0
Gemma 4 26B-A4B14tok/sQ8_014tok/sQ8_0
GLM 4.5 Air3tok/sQ6_K3tok/sQ6_K
Qwen 3.5 35B-A3B17tok/sQ8_017tok/sQ8_0
Qwen3 Next 80B-A3B16tok/sQ8_016tok/sQ8_0
GLM 4.7 Flash8tok/sQ8_08tok/sQ8_0
Devstral Small 2 24B4tok/sQ8_04tok/sQ8_0
Nemotron 3.5 Lightning 30B-A3B24tok/sUD-Q6_K_XL24tok/sUD-Q6_K_XL
Kimi Linear 48B-A3B24tok/sQ8_024tok/sQ8_0
gpt-oss 20B18tok/sMXFP418tok/sMXFP4
Qwen3 Coder 30B-A3B6tok/sQ8_06tok/sQ8_0
Qwen3 32B2tok/sQ8_02tok/sQ8_0
Gemma 4 12B11tok/sQ8_011tok/sQ8_0
Qwen 3.5 9B12tok/sQ8_012tok/sQ8_0
Gemma 3 27B4tok/sQ8_04tok/sQ8_0
Mistral Small 3.2 24B4tok/sQ8_04tok/sQ8_0
DeepSeek R1 Distill Qwen 32B2tok/sQ8_02tok/sQ8_0
Llama 3.3 70B1tok/sQ8_01tok/sQ8_0
Ministral 3 14B5tok/sQ8_05tok/sQ8_0
Qwen3 14B5tok/sQ8_05tok/sQ8_0
DeepSeek R1 0528 Qwen3 8B6tok/sQ8_06tok/sQ8_0
Qwen3 8B6tok/sQ8_06tok/sQ8_0
Phi-4 14B4tok/sQ8_04tok/sQ8_0
Qwen 3.5 4B19tok/sQ8_019tok/sQ8_0
Ministral 3 8B6tok/sQ8_06tok/sQ8_0
Granite 4.2 8B6tok/sQ8_06tok/sQ8_0
Gemma 4 E4B14tok/sQ8_014tok/sQ8_0
LFM2.5 8B-A1B14tok/sQ8_014tok/sQ8_0
Llama 3.1 8B7tok/sQ6_K7tok/sQ6_K
Qwen2.5 0.5B79tok/sQ8_079tok/sQ8_0
Qwen3.5 0.8B69tok/sQ8_069tok/sQ8_0
Qwen3.5 2B55tok/sQ5_K_M55tok/sQ5_K_M
LFM3 1.7B49tok/sQ8_049tok/sQ8_0
DeepSeek R1 Distill Qwen 1.5B30tok/sQ8_030tok/sQ8_0
Mellum2.1 12B A2.5B Thinking22tok/sQ8_022tok/sQ8_0
Spark X2.5 4B21tok/smlx-4bit21tok/smlx-4bit
MiniCPM5 2B20tok/sQ8_020tok/sQ8_0
FrogNano 4B 260919tok/sQ8_019tok/sQ8_0
LFM2.5 2.6B16tok/sQ8_016tok/sQ8_0
LFM2.5 VL 3B16tok/sQ8_016tok/sQ8_0
Kimi VL A3B Thinking 250614tok/sQ8_014tok/sQ8_0
Kimi VL A3B14tok/smlx-8bit14tok/smlx-8bit
GLM 4 9B 041412tok/smlx-5bit12tok/smlx-5bit
granite vision 4.1 4b12tok/sQ8_012tok/sQ8_0
granite 4.2 3b12tok/sQ8_012tok/sQ8_0
GLM 4.6V Flash11tok/sQ8_011tok/sQ8_0
DeepSeek R1 Distill Qwen 7B11tok/sQ8_011tok/sQ8_0
GLM 4.1V 9B Thinking10tok/smlx-8bit10tok/smlx-8bit
Ministral 3 3B Instruct 251210tok/sQ8_010tok/sQ8_0
LFM2.5 24B A2B8tok/sQ8_08tok/sQ8_0
DeepSeek R1 Distill Llama 8B7tok/sQ8_07tok/sQ8_0
Ministral 3 8B Reasoning 25126tok/sQ8_06tok/sQ8_0
Qwen3.8 27B OBLITERATED6tok/smlx-4bit6tok/smlx-4bit
Muse Glimmer 30B5tok/sQ8_05tok/sQ8_0
Ministral 3 14B Reasoning 25125tok/sQ8_05tok/sQ8_0
DeepSeek R1 Distill Qwen 14B4tok/sQ8_04tok/sQ8_0
Devstral Small 25074tok/sQ8_04tok/sQ8_0
Magistral Small 25064tok/sQ8_04tok/sQ8_0
Magistral Small 25074tok/sQ8_04tok/sQ8_0
Magistral Small 25094tok/sQ8_04tok/sQ8_0
GLM 4.5V3tok/sQ4_K_M3tok/sQ4_K_M
GLM 4.6V3tok/sQ4_K_M3tok/sQ4_K_M
granite 4.2 30b3tok/sQ8_03tok/sQ8_0
DeepSeek R1 Distill Llama 70B1tok/sQ8_01tok/sQ8_0
Devstral 2 123B Instruct 25121tok/sQ4_K_M1tok/sQ4_K_M
Kimi Dev 72B1tok/sQ8_01tok/sQ8_0
Mistral Medium 3.5 128B1tok/sQ4_K_M1tok/sQ4_K_M
DeepSeek R1won't runwon't run
DeepSeek R1 0528won't runwon't run
DeepSeek R1 Zerowon't runwon't run
DeepSeek V3won't runwon't run
DeepSeek V3 0324won't runwon't run
DeepSeek V3.1won't runwon't run
DeepSeek V3.1 Terminuswon't runwon't run
DeepSeek V3.2won't runwon't run
DeepSeek V4 Flash 0731won't runwon't run
DeepSeek V4 Flash Vision Expwon't runwon't run
GLM 4.5won't runwon't run
GLM 4.6won't runwon't run
GLM 4.7won't runwon't run
GLM 5won't runwon't run
GLM 5.1won't runwon't run
GLM 5.2won't runwon't run
GLM 5.3won't runwon't run
Kimi K2won't runwon't run
Kimi K2 Instruct 0905won't runwon't run
Kimi K2 Thinkingwon't runwon't run
Kimi K2.5won't runwon't run
MiniMax M2won't runwon't run
MiniMax M2.1won't runwon't run
MiniMax M2.5won't runwon't run
MiniMax M3won't runwon't run

Questions

Which can run bigger models, the Lenovo ThinkStation PGX GB10 128GB or the NVIDIA DGX Spark 128GB?
At 128k context both run the same 80 models we track fully in memory.
Which is faster for local AI, the Lenovo ThinkStation PGX GB10 128GB or the NVIDIA DGX Spark 128GB?
On the models both run fully in memory, the Lenovo ThinkStation PGX GB10 128GB and the NVIDIA DGX Spark 128GB are about as fast as each other (estimated).

More comparisons

More hardware for local AI