Z.ai GLM-5

GLM-5.3 hardware requirements

Needs at least 4 GPUs at FP8 (Instinct MI355X OAM). For 1,000 tokens/s at 32k context you need 16 x H100 SXM5 80GB at $29.28/hr or 32 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at $16.00/hr.

Parameters

753B

Active per token

41.7B

mixture of experts

KV cache per token

88 KB

16-bit cache

Native precision

FP8

Rent Model card

Top options for GLM-5.3

The strongest part each vendor offers for GLM-5.3 at FP8 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.

8 x GB200 NVL72 GPU 186GB 4 x Instinct MI355X OAM 8 x Ascend 950DT 16 x Gaudi 2 96GB 8 x TPU v7 192GB

Memory fills at: 8 x GB200 NVL72 GPU 186GB 229 users, 4 x Instinct MI355X OAM 120 users, 8 x Ascend 950DT 119 users, 16 x Gaudi 2 96GB 242 users, 8 x TPU v7 192GB 245 users. Roofline estimates at FP8.

Bare minimum

Fewest GPUs

4 x Instinct MI355X OAM

FP8, 4,096 tokens, one stream

Tokens/s: 507
Per hour: $23.96
Per million tokens: $13.13
Memory used: 68%

Rent on Daytona

Cheapest per hour

16 x RTX PRO 6000 Blackwell Max-Q Workstation Edition

FP8, 4,096 tokens, one stream, live price

Tokens/s: 377
Per hour: $8.00
Per million tokens: $5.89
Memory used: 51%

Rent on RunPod

Cheapest per million tokens

16 x RTX PRO 6000 Blackwell Max-Q Workstation Edition

FP8, 32,768 tokens, 32 streams

Tokens/s: 648
Per hour: $8.00
Per million tokens: $3.43
Memory used: 58%

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; 32 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	51 %	2.9k tok/s	108 tok/s	—	—	—	Specs →
2	Rubin SXM	4	76 %	2.7k tok/s	102 tok/s	—	—	—	Specs →
3	TPU 8t announced	8	51 %	~1.3k tok/s	57.2 tok/s	—	—	—	Specs →
4	TPU 8i announced	4	76 %	~791 tok/s	39.9 tok/s	—	—	—	Specs →
5	Instinct MI355X OAM	4	76 %	923 tok/s	37.1 tok/s	$23.96	$7.21	5.6 kW	Rent →
6	GB300 NVL72 GPU 288GB	4	76 %	922 tok/s	37.1 tok/s	$49.68	$14.97	—	Rent →
7	GB200 NVL72 GPU 186GB	8	59 %	1.7k tok/s	70.2 tok/s	$84.00	$13.40	—	Rent →
8	Instinct MI350X OAM	4	76 %	905 tok/s	37.1 tok/s	$24.64	$7.56	4.0 kW	Rent →
9	TPU v7 192GB	8	57 %	1.6k tok/s	64.7 tok/s	—	—	—	Specs →
10	Instinct MI430X OAM announced	4	51 %	~1.2k tok/s	108 tok/s	—	—	—	Specs →
11	B300 SXM 262GB	4	84 %	900 tok/s	37.1 tok/s	$19.96	$6.16	5.6 kW	Rent →
12	B200 SXM 180GB	8	61 %	1.7k tok/s	67.5 tok/s	$32.72	$5.51	8.0 kW	Rent →
13	Ascend 970 announced	4	76 %	1.3k tok/s	66.9 tok/s	—	—	—	Specs →
14	B100 SXM 192GB	8	57 %	1.6k tok/s	70.2 tok/s	—	—	5.6 kW	Specs →
15	Jalapeño announced	8	51 %	~2.4k tok/s	135 tok/s	—	—	5.6 kW	Specs →
16	Instinct MI325X OAM	4	86 %	630 tok/s	27.9 tok/s	$15.20	$6.70	4.0 kW	Rent →
17	Instinct MI300X 192GB	8	57 %	1.1k tok/s	46.5 tok/s	$19.12	$4.88	6.0 kW	Rent →
18	BR100 2 nodes	16	87 %	~618 tok/s	24.8 tok/s	—	—	8.8 kW	Specs →
19	Ascend 960 announced	4	76 %	764 tok/s	44.6 tok/s	—	—	—	Specs →
20	H200 SXM 141GB	8	78 %	937 tok/s	42.1 tok/s	$19.84	$5.88	5.6 kW	Rent →
21	H100 SXM5 80GB 2 nodes	16	69 %	1.3k tok/s	51.8 tok/s	$29.28	$6.40	11.2 kW	Rent →
22	Data Center GPU Max 1550 128GB	8	86 %	~539 tok/s	28.7 tok/s	—	—	4.8 kW	Specs →
23	H200 NVL 141GB	8	78 %	~887 tok/s	42.1 tok/s	$28.91	$9.06	4.8 kW	Rent →
24	H100 NVL 94GB 2 nodes	16	59 %	~1.4k tok/s	60.4 tok/s	$51.04	$10.44	6.4 kW	Rent →
25	Ascend 910C	8	86 %	~521 tok/s	28.1 tok/s	—	—	4.8 kW	Specs →
26	H100 PCIe 80GB 2 nodes	16	69 %	800 tok/s	31.0 tok/s	$40.00	$13.90	5.6 kW	Rent →
27	H800 PCIe 80GB 2 nodes	16	69 %	800 tok/s	31.0 tok/s	—	—	5.6 kW	Specs →
28	Ascend 950DT	8	77 %	661 tok/s	35.1 tok/s	—	—	—	Specs →
29	RTX PRO 6000 Blackwell Workstation Edition 2 nodes	16	58 %	672 tok/s	27.7 tok/s	$28.80	$11.91	9.6 kW	Rent →
30	Ascend 950PR	8	86 %	348 tok/s	14.0 tok/s	—	—	7.2 kW	Specs →
31	MTT S5000 2 nodes	16	69 %	~615 tok/s	24.8 tok/s	—	—	—	Specs →
32	RTX PRO 6000 Blackwell Server Edition 2 nodes	16	58 %	614 tok/s	24.7 tok/s	$11.04	$5.00	9.6 kW	Rent →
33	Gaudi 3 128GB	8	86 %	743 tok/s	32.4 tok/s	—	—	7.2 kW	Specs →
34	TPU v5p 95GB 2 nodes	16	58 %	656 tok/s	42.8 tok/s	—	—	7.2 kW	Specs →
35	RTX PRO 6000 Blackwell Max-Q Workstation Edition 2 nodes	16	58 %	648 tok/s	27.7 tok/s	$8.00	$3.43	4.8 kW	Rent →
36	Cloud AI 100 Ultra	8	86 %	~125 tok/s	4.8 tok/s	—	—	1.2 kW	Specs →
37	Gaudi 2 96GB 2 nodes	16	58 %	803 tok/s	37.9 tok/s	$17.60	$6.09	9.6 kW	Rent →
38	Atlas 350	8	99 %	299 tok/s	12.3 tok/s	—	—	4.8 kW	Specs →
39	Ascend 910B 2 nodes	16	87 %	~460 tok/s	24.8 tok/s	—	—	6.4 kW	Specs →
40	MI250X 128GB	8	86 %	~361 tok/s	28.7 tok/s	—	—	4.5 kW	Specs →
41	Instinct MI250 128GB	8	86 %	~349 tok/s	28.7 tok/s	—	—	4.0 kW	Specs →
42	A800 PCIe 80GB 2 nodes	16	69 %	~452 tok/s	29.9 tok/s	—	—	4.8 kW	Specs →
43	A800 SXM4 2 nodes	16	69 %	~463 tok/s	31.6 tok/s	—	—	6.4 kW	Specs →
44	A100 PCIe 80GB 2 nodes	16	69 %	~452 tok/s	29.9 tok/s	$21.60	$13.27	4.8 kW	Rent →
45	A100 SXM4 80GB 2 nodes	16	69 %	~463 tok/s	31.6 tok/s	$22.40	$13.43	6.4 kW	Rent →
46	RTX PRO 5000 Blackwell 72GB 2 nodes	16	77 %	~343 tok/s	20.8 tok/s	—	—	4.8 kW	Specs →
47	Instinct MI210 PCIe 2 nodes	16	87 %	~308 tok/s	25.4 tok/s	—	—	4.8 kW	Specs →
48	H20 141GB HBM3e	8	78 %	326 tok/s	35.1 tok/s	—	—	3.2 kW	Specs →
49	H20 96GB 2 nodes	16	58 %	576 tok/s	61.9 tok/s	—	—	6.4 kW	Specs →
50	Atlas 300I Duo 96GB 2 nodes	16	58 %	~132 tok/s	6.3 tok/s	—	—	2.4 kW	Specs →
51	ATOM-Max 2 nodes	16	87 %	~207 tok/s	15.8 tok/s	—	—	5.6 kW	Specs →

By vendor

At FP8 the NVIDIA pick is 16 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at about 648 tokens/s for $8.00/hr; the AMD pick is 8 x Instinct MI300X 192GB at about 1.1k tokens/s for $19.12/hr. Per rental dollar NVIDIA delivers 1.42x the tokens of AMD here.

NVIDIA 22 parts fit

Best value: 16 x RTX PRO 6000 Blackwell Max-Q Workstation Edition, 648 tok/s for $8.00/hr

Fewest GPUs: 4 x Rubin SXM, 2.7k tok/s

AMD 8 parts fit

Best value: 8 x Instinct MI300X 192GB, 1.1k tok/s for $19.12/hr

Fewest GPUs: 4 x Instinct MI455X, 2.9k tok/s

Huawei 6 parts fit

Fewest GPUs: 8 x Ascend 950DT, 661 tok/s

Intel 3 parts fit

Best value: 16 x Gaudi 2 96GB, 803 tok/s for $17.60/hr

Fewest GPUs: 8 x Gaudi 3 128GB, 743 tok/s

Google 2 parts fit

Fewest GPUs: 8 x TPU v7 192GB, 1.6k tok/s

Other 4 parts fit

Fewest GPUs: 8 x Cloud AI 100 Ultra, 125 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	1,000 x H200 SXM 141GB	125 x TP8	0	117k tok/s	$2,480	700 kW	$5.88
B	104 x GB200 NVL72 GPU 186GB	13 x TP8	0	23k tok/s	$1,092	—	$13.40

Size GLM-5.3 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-5.3 need?

The weights take 753 GB at the native FP8 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-5.3?

4 x Instinct MI355X OAM at FP8 with 4,096 tokens of context and one stream. The cheapest live rental for that footprint is 16 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at $8.00 per hour.

How many H100s do you need to run GLM-5.3?

16 H100s at FP8 for 4,096 tokens of context and one stream. For 32,768 tokens and 32 concurrent streams the tensor-parallel group is 16 H100s, and 1,000 tokens per second takes 16 H100s across 1 replicas.

How much does GLM-5.3 cost per million tokens on an H100?

About $6.40 per million output tokens at FP8, 32,768 tokens of context and 32 streams, using the cheapest live on-demand price of $1.83 per GPU-hour. The cheapest part per token is 16 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at $3.43.

What precision does GLM-5.3 ship in?

The published checkpoint is FP8. FP8 halves the BF16 footprint with the accuracy the lab validated; INT4 is a further community quantisation.

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. Same geometry as GLM-5.2, shipped in FP8. Active parameters computed from the expert geometry, net of the multi-token-prediction layer.