RTX PRO 5500 Price, 84GB VRAM & Local LLMs

The RTX PRO 5500 Blackwell, announced by NVIDIA in September 2026 with 84GB of VRAM, can run a 176B MoE model at 2-bit on one card. It isn't on sale yet, and its specs are preliminary.

eBay

RTX PRO 5500 price history

No quotes yet

eBay

Price updated —

Not enough data to show a trend

The median price trend will appear once enough valid quotes are available.

As an eBay Partner, VRAMGlass may be compensated if you buy through the eBay links on this page. The price you pay stays the same.

RTX PRO 5500 VRAM & full specs

Memory84 GB · GDDR7 ECC
Bandwidth1,398 GB/s

Bus width: 448-bit
Architecture: Blackwell (GB202)
Cores: 21,760
Tensor cores: 680
Ecosystem: CUDA
Interface: PCIe 5.0 x16
NVLink: Not supported
Power: 600 W
Slots: 2
Length: 282 mm
Type: Workstation
Launch date: Sep 14, 2026

What LLMs can a single RTX PRO 5500 run?

A single PRO 5500's 84GB just fits Qwen3.8 Flash Next at 2-bit, about 132 tokens/s. Up to 35B, go straight to 8-bit with full context.

Model	Size	Runs?	Speed t/s	Max context
Gemma 4 26B A4B25.2B (3.8B active)	10.5 GB	Runs	157.5	256K
Qwen3.6 27B27.8B	11.8 GB	Runs	79	256K
Qwen3.8 27B27.8B	9.8 GB	Runs	92.7	256K
GLM 4.7 Flash30B (3B active)	11.9 GB	Runs	166.3	198K
Gemma 4 31B30.7B	11.8 GB	Runs	76.3	256K
Qwen3.6 35B A3B36B (3B active)	12.3 GB	Runs	174.4	256K

Model	Size	Runs?	Speed t/s	Max context
Gemma 4 26B A4B25.2B (3.8B active)	16.9 GB	Runs	137.9	256K
Qwen3.6 27B27.8B	16.8 GB	Runs	57.7	256K
Qwen3.8 27B27.8B	16.5 GB	Runs	58.7	256K
GLM 4.7 Flash30B (3B active)	18.3 GB	Runs	151.3	198K
Gemma 4 31B30.7B	18.3 GB	Runs	52.2	256K
Qwen3.6 35B A3B36B (3B active)	22.1 GB	Runs	153.9	256K

Model	Size	Runs?	Speed t/s	Max context
Gemma 4 26B A4B25.2B (3.8B active)	26.9 GB	Runs	115.5	256K
Qwen3.6 27B27.8B	28.6 GB	Runs	35.3	256K
Qwen3.8 27B27.8B	29 GB	Runs	34.8	256K
GLM 4.7 Flash30B (3B active)	31.8 GB	Runs	127.1	198K
Gemma 4 31B30.7B	32.6 GB	Runs	30.8	256K
Qwen3.6 35B A3B36B (3B active)	36.9 GB	Runs	130.7	256K

RunsFits fully in VRAM; speed is estimated from VRAM bandwidth.

Needs RAM offloadMoE only: expert weights sit in system RAM, so speed is estimated from 70 GB/s RAM bandwidth.

Too bigWon’t fit on one card at this quantization, even with RAM offload.

Full calculation rules are on the Methodology page.

What LLMs can multiple RTX PRO 5500 cards run?

Two PRO 5500s (168GB) run DeepSeek V4 Flash at 4-bit at about 88 tokens/s; eight run the trillion-parameter Kimi K2.6 at 4-bit. No NVLink.

Model	Size	Cards	Total price	Total VRAM	Speed t/s	Total power
Qwen3.8 Flash Next176B (6B active)	78.9 GB	1 card	No price	84 GB	132	600 W
DeepSeek V4 Flash284B (13B active)	96.8 GB	2 cards	No price	168 GB	112.1	1,200 W
Hy3295B (21B active)	107.2 GB	2 cards	No price	168 GB	59.8	1,200 W
GLM 5.3 Flash320B (18B active)	108.7 GB	2 cards	No price	168 GB	94.6	1,200 W
GLM 5.3744B (40B active)	253.9 GB	4 cards	No price	336 GB	53.2	2,400 W
Hy4 preview770B (49B active)	307.2 GBest.	4 cards	No price	336 GB	41.2	2,400 W
Muse Spark Open816B (42B active)	311.2 GB	4 cards	No price	336 GB	48.1	2,400 W
Kimi K2.7 Code1,000B (32B active)	339.5 GB	8 cards	No price	672 GB	62.9	4,800 W
Kimi K2.61,000B (32B active)	340.5 GB	8 cards	No price	672 GB	62.8	4,800 W
Mistral Large 41,000B (52B active)	378.3 GB	8 cards	No price	672 GB	41.4	4,800 W
DeepSeek V4 Pro1,600B (49B active)	638.4 GBest.	8 cards	No price	672 GB	43.4	4,800 W

Model	Size	Cards	Total price	Total VRAM	Speed t/s	Total power
Qwen3.8 Flash Next176B (6B active)	111.3 GB	2 cards	No price	168 GB	116.2	1,200 W
DeepSeek V4 Flash284B (13B active)	155.1 GB	2 cards	No price	168 GB	87.7	1,200 W
Hy3295B (21B active)	182.2 GB	4 cards	No price	336 GB	46.1	2,400 W
GLM 5.3 Flash320B (18B active)	199.7 GB	4 cards	No price	336 GB	65.1	2,400 W
GLM 5.3744B (40B active)	467.3 GB	8 cards	No price	672 GB	33.9	4,800 W
Hy4 preview770B (49B active)	467.3 GB	8 cards	No price	672 GB	29.6	4,800 W
Muse Spark Open816B (42B active)	505.5 GB	8 cards	No price	672 GB	33.2	4,800 W
DeepSeek V4.1 Flash763B (8 / 16B active)	510.3 GB	8 cards	No price	672 GB	57.2	4,800 W
Kimi K2.61,000B (32B active)	583.7 GB	8 cards	No price	672 GB	43.1	4,800 W
Kimi K2.7 Code1,000B (32B active)	583.7 GB	8 cards	No price	672 GB	43.1	4,800 W
Mistral Large 41,000B (52B active)	604.8 GB	8 cards	No price	672 GB	28.5	4,800 W

Model	Size	Cards	Total price	Total VRAM	Speed t/s	Total power
DeepSeek V4 Flash284B (13B active)	161.9 GB	4 cards	No price	336 GB	85.5	2,400 W
Qwen3.8 Flash Next176B (6B active)	188.2 GB	4 cards	No price	336 GB	90.5	2,400 W
Hy3295B (21B active)	317.7 GB	8 cards	No price	672 GB	32.6	4,800 W
GLM 5.3 Flash320B (18B active)	341 GB	8 cards	No price	672 GB	43.9	4,800 W
Kimi K2.61,000B (32B active)	594.5 GB	8 cards	No price	672 GB	42.5	4,800 W
Kimi K2.7 Code1,000B (32B active)	594.5 GB	8 cards	No price	672 GB	42.5	4,800 W

Rows marked “Engram in RAM” need about 203 GB of extra system RAM. Full calculation rules are on the Methodology page.

Who it suits

Good fit

Running Qwen3.8 Flash Next at 2-bit on one card
About 132 tokens/s
Running dense models of 70B and up on one card
Llama 3.3 70B at 4-bit fits with plenty to spare
Building rack-mounted workstations
NVIDIA says it is designed for rack deployment, and there is also a liquid-cooled version

Not a good fit

You need it now
The website still says coming soon, with no release date
Comparing prompt processing speed
NVIDIA hasn't published compute figures yet
Tight system power
Each card draws 600W, as much as the RTX PRO 6000 Workstation Edition

Before you buy

The specs are still preliminary
NVIDIA's website marks the specs as preliminary and subject to change, with a status of coming soon and no release date. The core count and bus width come from PNY's spec page; clock speeds and AI compute haven't been published, so we have left our compute data blank until the official datasheet appears.
Size and power of the air-cooled version
The air-cooled version is a dual-slot, full-height, full-length card, about 282 mm long, noticeably longer than typical workstation cards, and it takes one 16-pin power connector. Measure the GPU clearance in your case before you build.
MIG splits it only in two
MIG can split it into at most two 42GB instances (the RTX PRO 6000 can split into four 24GB ones). Each is noticeably larger than the 36GB instances of the PRO 5000 72GB, big enough to hold even Gemma 4 31B at 8-bit (32.6 GB).

Compared with similar GPUs

RTX PRO 5500 vs RTX PRO 6000D, RTX PRO 6000, and RTX PRO 5000 72GB: price, VRAM, bandwidth, compute and local LLM speed compared.

Spec	RTX PRO 5500	RTX PRO 6000D	RTX PRO 6000	RTX PRO 5000 72GB
Memory	84 GB	84 GBSame	96 GB14% more	72 GB14% less
Bandwidth	1,398 GB/s	1,398 GB/sSame	1,792 GB/s28% higher	1,344 GB/s4% lower
BF16 / FP16 compute	—	385 TFLOPS	500 TFLOPS	258 TFLOPS
INT8 compute	—	771 TOPS	1,000 TOPS	516 TOPS
Qwen3.8 27B speed (2-bit)	92.7 t/s	92.7 t/sSame	114.6 t/s24% faster	89.5 t/s3% slower
Largest dense model (4-bit)	143B	143BSame	165B15% more	122B15% less
Power	600 W	600 WSame	600 WSame	300 W50% lower
Interface	PCIe 5.0 x16	PCIe 5.0 x16	PCIe 5.0 x16	PCIe 5.0 x16
Released	2026-09	2025-09	2025-03	2025-12
eBay median	View on eBay · RTX PRO 5500	View on eBay · RTX PRO 6000D	$18,550Buy RTX PRO 6000	$9,497Buy RTX PRO 5000 72GB

As an eBay Partner, VRAMGlass may be compensated if you buy through the eBay links on this page. The price you pay stays the same.

Other markets

Amazon.jp · Japan

N/A

No listings

View

Xianyu · Mainland China

N/A