B200
Pick the B200 for Blackwell with 180 GB per GPU. An 8-GPU node holds 1.44 TB of HBM3e. Per-GPU TDP is configurable up to 1,000 W.
NVIDIA B200B200 vs B300
Have a B200 or a B300, or need one? Tell us which. We verify the hardware, match a Have with a Need, and handle the deal. If you're selling, you pick which offer to take.

NVIDIA
B200

NVIDIA
B300
B200
Pick the B200 for Blackwell with 180 GB per GPU. An 8-GPU node holds 1.44 TB of HBM3e. Per-GPU TDP is configurable up to 1,000 W.
NVIDIA B200B300
Pick the B300 for Blackwell Ultra with 288 GB of HBM3e per GPU. An 8-GPU node holds 2.3 TB. Official DGX B300 is 10U at 14.5 kW.
NVIDIA B300Ampere
Hopper
Blackwell
B200
Blackwell Ultra
B300
180 GB HBM3e
288 GB HBM3e
Up to 8 TB/s per GPU
Up to 8 TB/s per GPU
1.44 TB HBM3e in the node
2.3 TB HBM3e in the node
Official DGX B200 is 10U, 14.3 kW max
Official DGX B300 is 10U, 14.5 kW
Architecture
B200. Blackwell
B300. Blackwell Ultra
Form factor
B200. SXM (HGX)
B300. SXM (HGX)
Per-GPU power
B200. Configurable up to 1,000 W per GPU
B300. Official DGX B300 is 14.5 kW for the 8-GPU node. A per-GPU wattage is not in the published specs.
Common systems
B200. HGX B200 8-GPU servers, DGX B200, GB200 NVL72 racks
B300. HGX B300 8-GPU servers, DGX B300, GB300 NVL72 racks
Both are Blackwell SXM GPUs for 8-GPU HGX servers, not standard PCIe cards. The B200 is 180 GB per GPU. The B300 is the Ultra version, 288 GB per GPU, for dense training and inference. A full liquid-cooled rack is a different system: GB200 NVL72 or GB300 NVL72.
The usual unit is a complete 8-GPU server, air-cooled or liquid-cooled. It can also be a GPU tray or part of an NVL72 rack. Tell us which, and whether the node is air or liquid, before we match it.
Say whether you have a complete 8-GPU node, a bare GPU tray, or a full rack. Note air or liquid cooling. These nodes are dense, and the cooling type decides where they can land. Ask for test, burn-in, and data-wipe records, and for missing PSUs, rails, optics, or the networking that ships with the node.
On a B200, note hours and whether GPUs were power-capped below the 1,000 W maximum. A tested, complete air-cooled 8-GPU node is the straightforward case. A rack that still needs CDUs, busbars, or switches takes longer.
On every Replenysh deal, both sides pass KYC, every buyer and destination is screened against export-control and sanctions lists, and we confirm serials and condition. The same NDA and deal terms apply. Payment, shipping, data wiping, and colocation are arranged based on what each deal needs. How a used GPU deal works.
Pick the B200 for Blackwell with 180 GB per GPU, 1.44 TB in an 8-GPU node. Pick the B300 for Blackwell Ultra with 288 GB per GPU, 2.3 TB in an 8-GPU node. Tell us which you have or need.
Both publish memory bandwidth up to 8 TB/s per GPU. The B200 is 180 GB, configurable up to 1,000 W per GPU, and an official DGX B200 is 14.3 kW max in 10U. The B300 is 288 GB, and an official DGX B300 is 14.5 kW in 10U.
Often, yes. Both usually move as a complete 8-GPU HGX or DGX-class server. Either can also be a tray or part of an NVL72 rack. We collect which one it is before we match it.
Complete node versus tray versus rack, air or liquid cooling, test records, and missing power supplies, optics, or networking. On a B200, note if GPUs were power-capped below 1,000 W. We confirm serials and condition before we bring the hardware to buyers.
No. We don't publish prices. Tell us what you have or what you need, and vetted buyers make offers. If you're selling, you pick which offer to take.
Every buyer and every seller passes a KYC identity and company check, and a Replenysh person reviews every buyer before they see a deal. Every buyer and destination is screened against export-control and sanctions lists before the hardware moves.
Tell us what you have, or what you need. We verify, match, and handle the deal.