Reference image
NVIDIA RTX PRO 6000 Blackwell
3 versions96 GB workstation and server GPU. Three editions with different power, cooling and host platforms.
Available to quote
Models
3 models · 8 versions to quote (Professional GPUs · Data center GPUs). Availability is held per version, not as one site-wide stock label.
Professional GPUs
1 model · 3 versions to quote
Reference image
Data center GPUs
2 models · 5 versions to quote
Reference image
NVIDIA H100
3 versionsTraining, fine-tuning and high-throughput inference, in three form factors.
Available to quote
SXM5 module SXM5 module · requires an HGX baseboard
- Form factor
- SXM
- GPU memory
- 80 GB HBM3
- GPU memory bandwidth
- 3.35 TB/s
- Max TDP
- Up to 700 W (configurable)
- Interconnect
- NVLink 900 GB/s; PCIe Gen5 128 GB/s
- MIG
- Up to 7 instances @ 10 GB
- Host platform
- HGX H100 partner/Certified systems with 4 or 8 GPUs
Specifications verified against NVIDIA H100.
Fits Mounts on an HGX H100 baseboard in a 4- or 8-GPU server. It is a module, not a PCIe card — it cannot be installed in a standard slot.
Best for Large-model training and fine-tuning where NVLink bandwidth between GPUs matters.
Get a quote for SXM5 moduleNVL (PCIe, NVLink pair) PCIe card · air-cooled, used in NVLink-bridged pairs
- Form factor
- PCIe, dual-slot air-cooled
- GPU memory
- 94 GB HBM3
- GPU memory bandwidth
- 3.9 TB/s
- Max TDP
- 350-400 W per GPU (configurable)
- Interconnect
- NVLink 600 GB/s; PCIe Gen5 128 GB/s
- MIG
- Up to 7 instances @ 12 GB
- Host platform
- Partner/Certified systems with 1-8 GPUs
Specifications verified against NVIDIA H100.
Fits PCIe card slot. The 350-400 W figure is per GPU — power-plan and quote for the pair, not for a single card.
Best for High-throughput inference, especially large language models, where the extra memory per GPU helps fit the model.
Get a quote for NVL (PCIe, NVLink pair)PCIe Add-in card · fits a standard server slot
- Form factor
- PCIe Gen5 x16, dual-slot air-cooled (unverified)
- GPU memory
- 80 GB HBM2e (unverified)
- GPU memory bandwidth
- Confirm with NVIDIA documentation (unverified)
- Max TDP
- Confirm with NVIDIA documentation (unverified)
Not verified yet: Memory, bandwidth, TDP and interface for the PCIe form; The NVIDIA H100 product page publishes SXM and NVL columns only; the PCIe row was not available to verify.. Ask us to confirm before you rely on it.
Fits Standard PCIe Gen5 x16 slot, with chassis airflow. Not interchangeable with an SXM module.
Best for Existing servers with free PCIe slots where an SXM baseboard is not an option.
Get a quote for PCIe
Reference image
NVIDIA A100 80 GB
2 versionsEstablished training and inference platform, in SXM4 and PCIe forms.
Available to quote
80 GB PCIe Add-in card · PCIe 4.0 x16
- Form factor
- PCIe, dual-slot air-cooled or single-slot liquid-cooled
- GPU memory
- 80 GB HBM2e
- GPU memory bandwidth
- 1,935 GB/s
- Max TDP
- 300 W
- Interconnect
- NVLink bridge for 2 GPUs: 600 GB/s; PCIe Gen4 64 GB/s
- MIG
- Up to 7 instances @ 10 GB
- Host platform
- Partner/Certified systems with 1-8 GPUs
Specifications verified against NVIDIA A100.
Fits Standard PCIe Gen4 x16 slot. Lower memory bandwidth than SXM4 and no HGX baseboard — the two forms are not interchangeable.
Best for Adding 80 GB accelerators to an existing PCIe server, or two-GPU NVLink pairs in one chassis.
Get a quote for 80 GB PCIe80 GB SXM4 SXM4 module · requires an HGX A100 baseboard
- Form factor
- SXM4
- GPU memory
- 80 GB HBM2e
- GPU memory bandwidth
- 2,039 GB/s
- Max TDP
- 400 W (standard configuration)
- Interconnect
- NVLink 600 GB/s; PCIe Gen4 64 GB/s
- MIG
- Up to 7 instances @ 10 GB
- Host platform
- HGX A100 partner/Certified systems with 4, 8 or 16 GPUs
Specifications verified against NVIDIA A100.
Fits Mounts on an HGX A100 baseboard in a 4-, 8- or 16-GPU server. It is a module, not a PCIe card.
Best for Dense multi-GPU training nodes where the higher SXM4 memory bandwidth and NVLink fabric are needed.
Get a quote for 80 GB SXM4This is the range we accept enquiries for, not a stock list. Add-in cards and SXM modules are quoted as separate versions and are not interchangeable.
Get a quote
Pick the product and the version if you know it. Quantity and delivery destination are enough for us to start.
Pricing and purchasing conditions
How quantity, condition and destination affect a quotation — and what is agreed per order rather than published site-wide.
- Pricing
- Quotation based. Prices are quoted for your quantity and condition.
- Quantity and price
- Priced by quantity. No site-wide discount tiers are published.
- Minimum order
- Confirmed per model and lot.
- First response
- We respond to complete RFQs within one business day.
- Formal quotation
- Issued after supply, version, destination and trade terms are verified.
- Quote validity
- Stated in the quotation. Not a fixed site-wide period.
- Stock hold
- Holding stock requires written confirmation. Submitting an RFQ does not reserve it.
What you can verify
Entity and registration
Company registration, tax identification and corporate documents, released per counterparty.
Entity & KYC packTest and grading documentation
Condition and test documentation for the units we supply, confirmed per model and lot.
Compliance & end-useDelivery, payment and after-sales
Each item below is confirmed in the written quotation for your order.
- Payment
- T/T or another confirmed method. Deposit and balance are agreed per order.
- Delivery terms
- Confirmed in the quotation. DAP or DDP is used only when it matches the actual arrangement.
- Lead time
- We state whether it is stock, dispatch or an estimated arrival, and what it runs from.
- Warranty and DOA
- Agreed per condition, product and lot.
- RMA
- The contact, the required materials, the steps and the freight responsibility are stated per order.
- Trade documents
- Invoice, test, end-user or other documents are confirmed per product, use and destination.
How a bulk purchase runs
- 01
Submit RFQ
Send the model, the quantity you need and the delivery destination.
- 02
Confirm supply and quote
We confirm the available supply route, the exact version and the quotation terms.
- 03
Agree terms and payment
Delivery terms, payment method and the deposit and balance split are agreed per order.
- 04
Dispatch and delivery
Dispatch and delivery follow the terms confirmed in writing for that order.
Buyer questions
What is the minimum order quantity?
MOQ is confirmed per model and lot. If you are not sure of your quantity yet, submit the quantity you need and we will confirm.
Are these units in stock or sourced on request?
Supply status is held per model, not as one site-wide stock label. Where a lot is confirmed we say so and show the check date; otherwise a model is sourced on request or its availability is confirmed on enquiry.
How do I get a bulk quote?
Send the model, the quantity and the delivery destination. Use the RFQ form on this site, or email the same three details to felix.mm@seraya.systems.
Can you deliver to my country?
We confirm the destinations we can serve. Other destinations are assessed per request.
What payment arrangements are available?
The payment method, and the deposit and balance split, are confirmed in the quotation for that order.
What warranty and DOA terms apply?
Warranty and DOA responsibility are agreed per product, condition and lot, and stated in the quotation. They are not applied as one site-wide term.
Can I place a trial order or request staged delivery?
You can ask. Whether we accept, and on what terms, is confirmed against actual supply and purchasing conditions.
How long is a quote valid, and can stock be reserved?
Validity is written into each quotation rather than a fixed period. Submitting an RFQ does not reserve stock; holding stock requires written confirmation.
Need a complete system?
Cards are quoted here. Complete systems are configured and quoted on the Systems line.
felix.mm@seraya.systems
Get a quote
Model, version, quantity and delivery destination are enough to start. We respond to complete RFQs within one business day.