Z.ai GLM-4.7

GLM-4.7-Flash hardware requirements

Fits on one Instinct MI355X OAM at BF16. For 1,000 tokens/s at 32k context you need 2 x H100 SXM5 80GB at $3.66/hr or 4 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at $2.00/hr.

Parameters

31.2B

Active per token

3.0B

mixture of experts

KV cache per token

53 KB

16-bit cache

Native precision

BF16

Rent Model card

Top options for GLM-4.7-Flash

The strongest part each vendor offers for GLM-4.7-Flash 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.

1 x GB300 NVL72 GPU 288GB 1 x Instinct MI355X OAM 1 x Ascend 950DT 1 x Gaudi 2 96GB 1 x TPU v7 192GB 4 x Inferentia2

Memory fills at: 1 x GB300 NVL72 GPU 288GB 121 users, 1 x Instinct MI355X OAM 121 users, 1 x Ascend 950DT 42 users, 1 x Gaudi 2 96GB 16 users, 1 x TPU v7 192GB 68 users, 4 x Inferentia2 32 users. Roofline estimates at BF16.

Bare minimum

Fewest GPUs

1 x Instinct MI355X OAM

BF16, 4,096 tokens, one stream

Tokens/s: 952
Per hour: $5.99
Per million tokens: $1.75
Memory used: 23%

Rent on Daytona

Cheapest per hour

1 x RTX PRO 6000 Blackwell Max-Q Workstation Edition

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

Tokens/s: 212
Per hour: $0.50
Per million tokens: $0.65
Memory used: 68%

Rent on RunPod

Cheapest per million tokens

2 x RTX PRO 6000 Blackwell Max-Q Workstation Edition

BF16, 32,768 tokens, 32 streams

Tokens/s: 574
Per hour: $1.00
Per million tokens: $0.48
Memory used: 65%

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; 67 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	1	29 %	3.8k tok/s	157 tok/s	—	—	—	Specs →
2	Rubin SXM	1	43 %	3.4k tok/s	148 tok/s	—	—	—	Specs →
3	TPU 8t announced	1	57 %	~1.2k tok/s	43.9 tok/s	—	—	—	Specs →
4	TPU 8i announced	1	43 %	~1.5k tok/s	57.8 tok/s	—	—	—	Specs →
5	Instinct MI355X OAM	1	43 %	1.4k tok/s	53.8 tok/s	$5.99	$1.18	1.4 kW	Rent →
6	GB300 NVL72 GPU 288GB	1	43 %	1.4k tok/s	53.8 tok/s	$12.42	$2.44	—	Rent →
7	GB200 NVL72 GPU 186GB	1	67 %	1.4k tok/s	53.8 tok/s	$10.50	$2.06	—	Rent →
8	Instinct MI350X OAM	1	43 %	1.4k tok/s	53.8 tok/s	$6.16	$1.23	1.0 kW	Rent →
9	TPU v7 192GB	1	65 %	1.3k tok/s	49.6 tok/s	—	—	—	Specs →
10	Instinct MI430X OAM announced	1	29 %	~3.0k tok/s	157 tok/s	—	—	—	Specs →
11	B300 SXM 262GB	1	47 %	1.4k tok/s	53.8 tok/s	$4.99	$1.00	1.4 kW	Rent →
12	B200 SXM 180GB	1	69 %	1.3k tok/s	51.8 tok/s	$4.09	$0.85	1.0 kW	Rent →
13	Ascend 970 announced	1	43 %	2.1k tok/s	96.8 tok/s	—	—	—	Specs →
14	B100 SXM 192GB	1	65 %	1.3k tok/s	53.8 tok/s	—	—	0.7 kW	Specs →
15	Jalapeño announced	1	57 %	~2.1k tok/s	104 tok/s	—	—	0.7 kW	Specs →
16	Instinct MI325X OAM	1	48 %	984 tok/s	40.3 tok/s	$3.80	$1.07	1.0 kW	Rent →
17	Instinct MI300X 192GB	1	65 %	894 tok/s	35.6 tok/s	$2.39	$0.74	0.8 kW	Rent →
18	BR100	2	98 %	591 tok/s	20.4 tok/s	—	—	1.1 kW	Specs →
19	Ascend 960 announced	1	43 %	1.3k tok/s	64.5 tok/s	—	—	—	Specs →
20	H200 SXM 141GB	1	88 %	777 tok/s	32.3 tok/s	$2.48	$0.89	0.7 kW	Rent →
21	H100 SXM5 80GB	2	78 %	1.1k tok/s	42.8 tok/s	$3.66	$0.91	1.4 kW	Rent →
22	TPU v6e 32GB	4	100 %	1.1k tok/s	39.6 tok/s	—	—	—	Specs →
23	Data Center GPU Max 1550 128GB	1	97 %	557 tok/s	22.0 tok/s	—	—	0.6 kW	Specs →
24	H200 NVL 141GB	1	88 %	~743 tok/s	32.3 tok/s	$3.61	$1.35	0.6 kW	Rent →
25	H100 NVL 94GB	2	66 %	~1.2k tok/s	49.8 tok/s	$6.38	$1.46	0.8 kW	Rent →
26	Ascend 910C	1	97 %	542 tok/s	21.5 tok/s	—	—	0.6 kW	Specs →
27	H100 PCIe 80GB	2	78 %	693 tok/s	25.5 tok/s	$5.00	$2.00	0.7 kW	Rent →
28	H800 PCIe 80GB	2	78 %	693 tok/s	25.5 tok/s	—	—	0.7 kW	Specs →
29	Ascend 950DT	1	86 %	575 tok/s	26.9 tok/s	—	—	—	Specs →
30	RTX PRO 6000 Blackwell Workstation Edition	2	65 %	590 tok/s	22.9 tok/s	$3.60	$1.69	1.2 kW	Rent →
31	Ascend 950PR	1	97 %	283 tok/s	10.8 tok/s	—	—	0.9 kW	Specs →
32	MTT S5000	2	78 %	~537 tok/s	20.4 tok/s	—	—	—	Specs →
33	RTX PRO 6000 Blackwell Server Edition	2	65 %	537 tok/s	20.4 tok/s	$1.38	$0.71	1.2 kW	Rent →
34	Gaudi 3 128GB	1	97 %	692 tok/s	24.9 tok/s	—	—	0.9 kW	Specs →
35	TPU v5p 95GB	2	66 %	802 tok/s	35.3 tok/s	—	—	0.9 kW	Specs →
36	RTX PRO 6000 Blackwell Max-Q Workstation Edition	2	65 %	574 tok/s	22.9 tok/s	$1.00	$0.48	0.6 kW	Rent →
37	Cloud AI 100 Ultra	1	97 %	~109 tok/s	3.7 tok/s	—	—	0.1 kW	Specs →
38	Gaudi 2 96GB	2	65 %	723 tok/s	31.3 tok/s	$2.20	$0.84	1.2 kW	Rent →
39	Atlas 350	2	56 %	468 tok/s	17.9 tok/s	—	—	1.2 kW	Specs →
40	Data Center GPU Max 1100 48GB	4	66 %	794 tok/s	29.7 tok/s	—	—	1.2 kW	Specs →
41	GeForce RTX 5090 32GB	4	100 %	878 tok/s	43.4 tok/s	$4.76	$1.51	2.3 kW	Rent →
42	Ascend 910B	2	98 %	514 tok/s	20.4 tok/s	—	—	0.8 kW	Specs →
43	MI250X 128GB	1	97 %	446 tok/s	22.0 tok/s	—	—	0.6 kW	Specs →
44	RTX 6000 Ada 48GB	4	66 %	~631 tok/s	23.2 tok/s	$3.88	$1.71	1.2 kW	Rent →
45	Instinct MI250 128GB	1	97 %	437 tok/s	22.0 tok/s	—	—	0.5 kW	Specs →
46	L40S 48GB	4	66 %	576 tok/s	20.9 tok/s	$3.04	$1.47	1.4 kW	Rent →
47	A800 PCIe 80GB	2	78 %	557 tok/s	24.7 tok/s	—	—	0.6 kW	Specs →
48	A800 SXM4	2	78 %	578 tok/s	26.0 tok/s	—	—	0.8 kW	Specs →
49	A100 PCIe 80GB	2	78 %	557 tok/s	24.7 tok/s	$2.70	$1.35	0.6 kW	Rent →
50	A100 SXM4 80GB	2	78 %	578 tok/s	26.0 tok/s	$2.80	$1.35	0.8 kW	Rent →
51	A100 PCIe 40GB	4	79 %	900 tok/s	37.6 tok/s	—	—	1.0 kW	Specs →
52	A100 SXM4 40GB	4	79 %	900 tok/s	37.6 tok/s	$6.24	$1.93	1.6 kW	Rent →
53	GeForce RTX 5090 D V2 24GB	8	67 %	~1.5k tok/s	61.4 tok/s	—	—	4.6 kW	Specs →
54	RTX 5880 Ada 48GB	4	66 %	~602 tok/s	23.2 tok/s	—	—	1.1 kW	Specs →
55	TPU v4 32GB	4	100 %	717 tok/s	29.0 tok/s	—	—	—	Specs →
56	RTX 5000 Ada 32GB	4	100 %	~388 tok/s	13.9 tok/s	$3.32	$2.38	1.0 kW	Rent →
57	RTX PRO 5000 Blackwell 48GB	4	66 %	~769 tok/s	32.5 tok/s	$3.84	$1.39	1.2 kW	Rent →
58	RTX PRO 5000 Blackwell 72GB	2	87 %	~406 tok/s	17.2 tok/s	—	—	0.6 kW	Specs →
59	RTX PRO 4500 Blackwell Server Edition	4	100 %	489 tok/s	19.4 tok/s	—	—	0.7 kW	Specs →
60	RTX PRO 4500 Blackwell 32GB	4	100 %	534 tok/s	21.7 tok/s	$2.88	$1.50	0.8 kW	Rent →
61	Radeon AI PRO R9700 32GB	4	100 %	404 tok/s	15.5 tok/s	—	—	1.2 kW	Specs →
62	Inferentia2	4	100 %	491 tok/s	19.8 tok/s	—	—	—	Specs →
63	Trainium	4	100 %	491 tok/s	19.8 tok/s	—	—	—	Specs →
64	Instinct MI100 32GB	4	100 %	500 tok/s	29.7 tok/s	—	—	1.2 kW	Specs →
65	Arc Pro B70 32GB	4	100 %	~384 tok/s	14.7 tok/s	—	—	0.9 kW	Specs →
66	L40 48GB	4	66 %	506 tok/s	20.9 tok/s	$3.28	$1.80	1.2 kW	Rent →
67	Instinct MI210 PCIe	2	98 %	415 tok/s	20.9 tok/s	—	—	0.6 kW	Specs →
68	GeForce RTX 4090 24GB	8	67 %	1.0k tok/s	46.1 tok/s	$7.12	$1.90	3.6 kW	Rent →
69	A30 24GB	8	67 %	987 tok/s	42.6 tok/s	—	—	1.3 kW	Specs →
70	RTX 4500 Ada 24GB	8	67 %	~533 tok/s	19.7 tok/s	—	—	1.7 kW	Specs →
71	RTX A6000 48GB	4	66 %	445 tok/s	18.6 tok/s	$2.00	$1.25	1.2 kW	Rent →
72	A40 48GB	4	66 %	410 tok/s	16.8 tok/s	$2.36	$1.60	1.2 kW	Rent →
73	H20 141GB HBM3e	1	88 %	304 tok/s	26.9 tok/s	—	—	0.4 kW	Specs →
74	H20 96GB	2	65 %	577 tok/s	51.1 tok/s	—	—	0.8 kW	Specs →
75	RTX PRO 4000 Blackwell 24GB	8	67 %	~751 tok/s	30.7 tok/s	$4.56	$1.69	1.2 kW	Rent →
76	BI-V100 Tiangai 100	4	100 %	598 tok/s	29.0 tok/s	—	—	1.0 kW	Specs →
77	Atlas 300I Duo 48GB	4	66 %	264 tok/s	9.9 tok/s	—	—	0.6 kW	Specs →
78	Atlas 300I Duo 96GB	2	65 %	139 tok/s	5.2 tok/s	—	—	0.3 kW	Specs →
79	RTX A5500 24GB	8	67 %	~813 tok/s	35.1 tok/s	—	—	1.8 kW	Specs →
80	Wormhole n300s	8	67 %	649 tok/s	26.3 tok/s	—	—	2.4 kW	Specs →
81	V100S PCIe 32GB	4	100 %	552 tok/s	27.4 tok/s	$3.52	$1.77	1.0 kW	Rent →
82	ATOM-Max	2	98 %	271 tok/s	13.1 tok/s	—	—	0.7 kW	Specs →
83	A10 24GB	8	67 %	662 tok/s	27.4 tok/s	$8.81	$3.70	1.2 kW	Rent →
84	Radeon PRO W7900 Dual Slot 48GB	4	66 %	453 tok/s	20.9 tok/s	—	—	1.2 kW	Specs →
85	Radeon PRO W7900 48GB	4	66 %	453 tok/s	20.9 tok/s	—	—	1.2 kW	Specs →
86	Radeon RX 7900 XTX 24GB	8	67 %	918 tok/s	43.9 tok/s	—	—	2.8 kW	Specs →
87	L4 24GB	8	67 %	376 tok/s	13.7 tok/s	$4.72	$3.49	0.6 kW	Rent →
88	L20 48GB	4	66 %	449 tok/s	20.9 tok/s	—	—	1.1 kW	Specs →
89	Radeon AI PRO R9600D 32GB	4	100 %	345 tok/s	15.5 tok/s	—	—	0.6 kW	Specs →
90	Arc Pro B65 32GB	4	100 %	~332 tok/s	14.7 tok/s	—	—	0.8 kW	Specs →
91	Arc Pro B60 24GB	8	67 %	~508 tok/s	20.8 tok/s	—	—	1.6 kW	Specs →
92	L2 24GB	8	67 %	362 tok/s	13.7 tok/s	—	—	—	Specs →
93	RTX PRO 4000 Blackwell SFF 24GB	8	67 %	~485 tok/s	19.7 tok/s	—	—	0.6 kW	Specs →
94	MLU370-X8	4	66 %	332 tok/s	14.9 tok/s	—	—	1.0 kW	Specs →
95	Radeon PRO W7800 48GB	4	66 %	406 tok/s	20.9 tok/s	—	—	1.0 kW	Specs →
96	Radeon PRO W7800 32GB	4	100 %	312 tok/s	13.9 tok/s	—	—	1.0 kW	Specs →
97	GeForce RTX 3090 24GB	8	67 %	723 tok/s	42.8 tok/s	$4.00	$1.54	2.8 kW	Rent →
98	A10G 24GB	8	67 %	555 tok/s	27.4 tok/s	$8.05	$4.03	2.4 kW	Rent →
99	Atlas 300V Pro	4	66 %	132 tok/s	5.0 tok/s	—	—	—	Specs →
100	Atlas 300I Pro	8	67 %	250 tok/s	9.4 tok/s	—	—	—	Specs →
101	Atlas 300V	8	67 %	234 tok/s	9.4 tok/s	—	—	0.6 kW	Specs →
102	GeForce RTX 3090 Ti 24GB	8	67 %	545 tok/s	46.1 tok/s	$3.68	$1.88	3.6 kW	Rent →
103	Radeon PRO W6800 32GB	4	100 %	201 tok/s	12.4 tok/s	—	—	1.0 kW	Specs →
104	RTX A5000 24GB	8	67 %	391 tok/s	35.1 tok/s	$2.16	$1.53	1.8 kW	Rent →
105	Radeon PRO V710 28GB	8	57 %	312 tok/s	20.5 tok/s	—	—	1.3 kW	Specs →

By vendor

At BF16 the NVIDIA pick is 2 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at about 574 tokens/s for $1.00/hr; the AMD pick is 1 x Instinct MI300X 192GB at about 894 tokens/s for $2.39/hr. Per rental dollar NVIDIA delivers 1.53x the tokens of AMD here.

NVIDIA 51 parts fit

Best value: 2 x RTX PRO 6000 Blackwell Max-Q Workstation Edition, 574 tok/s for $1.00/hr

Fewest GPUs: 1 x Rubin SXM, 3.4k tok/s

AMD 18 parts fit

Best value: 1 x Instinct MI300X 192GB, 894 tok/s for $2.39/hr

Fewest GPUs: 1 x Instinct MI455X, 3.8k tok/s

Huawei 10 parts fit

Fewest GPUs: 1 x Ascend 950DT, 575 tok/s

Intel 7 parts fit

Best value: 2 x Gaudi 2 96GB, 723 tok/s for $2.20/hr

Fewest GPUs: 1 x Gaudi 3 128GB, 692 tok/s

Google 4 parts fit

Fewest GPUs: 1 x TPU v7 192GB, 1.3k tok/s

AWS 2 parts fit

Fewest GPUs: 4 x Inferentia2, 491 tok/s

Other 7 parts fit

Fewest GPUs: 1 x Cloud AI 100 Ultra, 109 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	1,000 x TP1	0	777k tok/s	$2,480	700 kW	$0.89
B	100 x GB200 NVL72 GPU 186GB	100 x TP1	0	141k tok/s	$1,050	—	$2.06

Size GLM-4.7-Flash 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-4.7-Flash need?

The weights take 62.4 GB at the native BF16 precision (62.4 GB at BF16, 31.2 GB at FP8, 17.6 GB at INT4). Each concurrent stream adds 53 KB of KV cache per token: 0.2 GB at 4,096 tokens and 1.8 GB at 32,768 tokens.

What is the bare minimum to run GLM-4.7-Flash?

1 x Instinct MI355X OAM at BF16 with 4,096 tokens of context and one stream. The cheapest live rental for that footprint is 1 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at $0.50 per hour.

How many H100s do you need to run GLM-4.7-Flash?

one H100 at BF16 for 4,096 tokens of context and one stream. For 32,768 tokens and 32 concurrent streams the tensor-parallel group is 2 H100s, and 1,000 tokens per second takes 2 H100s across 1 replicas.

How much does GLM-4.7-Flash cost per million tokens on an H100?

About $0.91 per million output tokens at BF16, 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 2 x RTX PRO 6000 Blackwell Max-Q Workstation Edition at $0.48.

What precision does GLM-4.7-Flash 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. Lightweight MLA model; the card describes it as 30B-A3B.