Z.ai GLM-5

GLM-5.2 hardware requirements

Needs at least 8 GPUs at BF16 (Instinct MI355X OAM). For 1,000 tokens/s at 32k context you need 16 x Instinct MI300X 192GB at $38.24/hr.

Parameters

753B

Active per token

41.7B

mixture of experts

KV cache per token

88 KB

16-bit cache

Native precision

BF16

Rent Model card

Top options for GLM-5.2

The strongest part each vendor offers for GLM-5.2 at BF16 with 32,768 tokens per user, rentable parts first. Each card is one replica, the fewest accelerators that hold the model, and the line it draws on the graph below. Select a card to plot it.

Tokens per second as users grow

Each line is one replica, the fewest accelerators that hold the model, serving more users at once. It climbs while memory bandwidth is the limit, flattens once compute is, and stops where memory runs out at 32,768 tokens per user. Hover a line to bring it forward.

16 x GB200 NVL72 GPU 186GB 8 x Instinct MI355X OAM 16 x Ascend 910C 16 x Gaudi 3 128GB 16 x TPU v7 192GB

Memory fills at: 16 x GB200 NVL72 GPU 186GB 459 users, 8 x Instinct MI355X OAM 241 users, 16 x Ascend 910C 154 users, 16 x Gaudi 3 128GB 154 users, 16 x TPU v7 192GB 490 users. Roofline estimates at BF16.

Bare minimum

Fewest GPUs

8 x Instinct MI355X OAM

BF16, 4,096 tokens, one stream

Tokens/s: 481
Per hour: $47.92
Per million tokens: $27.69
Memory used: 68%

Rent on Daytona

Cheapest per hour

8 x Instinct MI325X OAM

BF16, 4,096 tokens, one stream, live price

Tokens/s: 358
Per hour: $30.40
Per million tokens: $23.56
Memory used: 76%

Rent on DigitalOcean

Cheapest per million tokens

16 x Instinct MI300X 192GB

BF16, 32,768 tokens, 32 streams

Tokens/s: 1.0k
Per hour: $38.24
Per million tokens: $10.40
Memory used: 54%

Rent on RunPod

Configurations

Every datacenter GPU in the catalog, sized for these settings. Change precision, context and concurrency and the whole page follows. Each point is one replica: the smallest tensor-parallel group of that GPU that holds the model.

Precision Context Streams

Vendor

Priced only

Priced configurations only; 20 more fit without a live price and appear in the table. Marker size grows with the tensor-parallel width. Roofline estimates.

#									Rent
1	Instinct MI455X	4	96 %	1.3k tok/s	58.9 tok/s	—	—	—	Specs →
2	Rubin SXM	8	72 %	2.2k tok/s	105 tok/s	—	—	—	Specs →
3	TPU 8t announced	8	96 %	~828 tok/s	31.1 tok/s	—	—	—	Specs →
4	TPU 8i announced	8	72 %	~984 tok/s	41.0 tok/s	—	—	—	Specs →
5	Instinct MI355X OAM	8	72 %	931 tok/s	38.2 tok/s	$47.92	$14.30	11.2 kW	Rent →
6	GB300 NVL72 GPU 288GB	8	72 %	929 tok/s	38.2 tok/s	$99.36	$29.70	—	Rent →
7	GB200 NVL72 GPU 186GB 2 nodes	16	56 %	1.6k tok/s	67.4 tok/s	$168	$28.46	—	Rent →
8	Instinct MI350X OAM	8	72 %	911 tok/s	38.2 tok/s	$49.28	$15.02	8.0 kW	Rent →
9	TPU v7 192GB 2 nodes	16	54 %	1.5k tok/s	62.1 tok/s	—	—	—	Specs →
10	Instinct MI430X OAM announced	4	96 %	~944 tok/s	58.9 tok/s	—	—	—	Specs →
11	B300 SXM 262GB	8	79 %	905 tok/s	38.2 tok/s	$39.92	$12.25	11.2 kW	Rent →
12	B200 SXM 180GB 2 nodes	16	58 %	1.6k tok/s	64.8 tok/s	$65.44	$11.71	16.0 kW	Rent →
13	Ascend 970 announced	8	72 %	1.3k tok/s	68.7 tok/s	—	—	—	Specs →
14	B100 SXM 192GB 2 nodes	16	54 %	1.5k tok/s	67.4 tok/s	—	—	11.2 kW	Specs →
15	Jalapeño announced	8	96 %	~1.2k tok/s	73.5 tok/s	—	—	5.6 kW	Specs →
16	Instinct MI325X OAM	8	81 %	631 tok/s	28.6 tok/s	$30.40	$13.37	8.0 kW	Rent →
17	Instinct MI300X 192GB 2 nodes	16	54 %	1.0k tok/s	44.6 tok/s	$38.24	$10.40	12.0 kW	Rent →
18	Ascend 960 announced	8	72 %	754 tok/s	45.8 tok/s	—	—	—	Specs →
19	H200 SXM 141GB 2 nodes	16	74 %	876 tok/s	40.4 tok/s	$39.68	$12.58	11.2 kW	Rent →
20	Data Center GPU Max 1550 128GB 2 nodes	16	82 %	638 tok/s	27.6 tok/s	—	—	9.6 kW	Specs →
21	H200 NVL 141GB 2 nodes	16	74 %	~827 tok/s	40.4 tok/s	$57.82	$19.43	9.6 kW	Rent →
22	Ascend 910C 2 nodes	16	82 %	619 tok/s	26.9 tok/s	—	—	9.6 kW	Specs →
23	Ascend 950DT 2 nodes	16	72 %	627 tok/s	33.7 tok/s	—	—	—	Specs →
24	Ascend 950PR 2 nodes	16	82 %	328 tok/s	13.5 tok/s	—	—	14.4 kW	Specs →
25	Gaudi 3 128GB 2 nodes	16	82 %	819 tok/s	31.2 tok/s	—	—	14.4 kW	Specs →
26	Cloud AI 100 Ultra 2 nodes	16	82 %	~131 tok/s	4.6 tok/s	—	—	2.4 kW	Specs →
27	Atlas 350 2 nodes	16	93 %	285 tok/s	11.8 tok/s	—	—	9.6 kW	Specs →
28	MI250X 128GB 2 nodes	16	82 %	480 tok/s	27.6 tok/s	—	—	9.0 kW	Specs →
29	Instinct MI250 128GB 2 nodes	16	82 %	467 tok/s	27.6 tok/s	—	—	8.0 kW	Specs →
30	H20 141GB HBM3e 2 nodes	16	74 %	295 tok/s	33.7 tok/s	—	—	6.4 kW	Specs →

By vendor

At BF16 the NVIDIA pick is 16 x B200 SXM 180GB at about 1.6k tokens/s for $65.44/hr; the AMD pick is 16 x Instinct MI300X 192GB at about 1.0k tokens/s for $38.24/hr. Per rental dollar AMD delivers 1.13x the tokens of NVIDIA here.

NVIDIA 9 parts fit

Best value: 16 x B200 SXM 180GB, 1.6k tok/s for $65.44/hr

Fewest GPUs: 8 x Rubin SXM, 2.2k tok/s

AMD 7 parts fit

Best value: 16 x Instinct MI300X 192GB, 1.0k tok/s for $38.24/hr

Fewest GPUs: 4 x Instinct MI455X, 1.3k tok/s

Huawei 4 parts fit

Fewest GPUs: 16 x Ascend 950DT, 627 tok/s

Intel 2 parts fit

Fewest GPUs: 16 x Gaudi 3 128GB, 819 tok/s

Google 1 parts fit

Fewest GPUs: 16 x TPU v7 192GB, 1.5k tok/s

Other 1 parts fit

Fewest GPUs: 16 x Cloud AI 100 Ultra, 131 tok/s

Fleet what-if

Compare whole fleets at these settings: a thousand of one part against a hundred of another. Each fleet splits into replicas of its tensor-parallel width; GPUs left over sit idle.

A GPUs Part

B GPUs Part

Fleet letter	Fleet	Replicas	Idle	Tokens/s	Per hour	Power	Per M tokens
A	1,008 x H200 SXM 141GB	63 x TP16	0	55k tok/s	$2,500	706 kW	$12.58
B	112 x GB200 NVL72 GPU 186GB	7 x TP16	0	11k tok/s	$1,176	—	$28.46

Size GLM-5.2 for a tokens per second target

The calculator starts from your traffic instead of a fleet: set a target rate and read the replica count for every GPU.

Open in the LLM GPU Calculator

Frequently asked questions

How much GPU memory does GLM-5.2 need?

The weights take 1,507 GB at the native BF16 precision (1,507 GB at BF16, 753 GB at FP8, 424 GB at INT4). Each concurrent stream adds 88 KB of KV cache per token: 0.4 GB at 4,096 tokens and 2.9 GB at 32,768 tokens.

What is the bare minimum to run GLM-5.2?

8 x Instinct MI355X OAM at BF16 with 4,096 tokens of context and one stream. The cheapest live rental for that footprint is 8 x Instinct MI325X OAM at $30.40 per hour.

How many H100s do you need to run GLM-5.2?

more than sixteen H100s at BF16 for 4,096 tokens of context and one stream. For 32,768 tokens and 32 concurrent streams the tensor-parallel group is more than sixteen H100s.

How much does GLM-5.2 cost per million tokens on an H100?

No live H100 rental price is on file right now, so the cost per token cannot be quoted. The configuration table lists every part that has one.

What precision does GLM-5.2 ship in?

The published checkpoint is BF16. FP8 and INT4 figures describe post-training quantisations that halve and quarter the footprint at a small accuracy cost.

All figures are roofline estimates from the model's config.json and each GPU's published memory, bandwidth and tensor-core peak, with fixed efficiency factors. Prices are the cheapest live on-demand listing per GPU when the site has one. They are a planning floor; a tuned serving stack can do better. Active parameters computed from the expert geometry, net of the multi-token-prediction layer. A sparse-attention indexer sits over the MLA cache; the cache figure ignores the indexer.