# Aivres KRS8500V3

*Liquid-Cooled Exascale AI Rack Solution based on NVIDIA GB300 NVL72*

Machine-readable product context for AI assistants. Consumer page: https://aivres.com/product/krs8500v3/. JSON: https://aivres.upstation.io/p/krs8500v3.json.

KRS8500V3 is Aivres’ next-generation L11 rack-scale platform, integrating 36 NVIDIA Grace CPUs and 72 NVIDIA Blackwell Ultra GPUs into a fully liquid-cooled architecture. Designed for trillion-parameter LLM training and inference, it delivers unprecedented performance density, scalability, and energy efficiency, setting the foundation for large-scale AI, HPC, and data analytics in data centers worldwide.
Compared to the previous generation KRS8000V3 based on NVIDIA GB200 NVL72, KRS8500V3 achieves up to 1.5x faster training and ~1.4x higher inference throughput, establishing new MLPerf records and redefining the standard for hyperscale AI computing.

Supporting Extreme Performance Applications

## Highlights
- Ultra Tensor Cores: 2x attention-layer acceleration, 1.5x AI compute FLOPS uplift; Optimized for trillion-scale LLM and mixture-of-experts (MoE) workloads
- Expanded HBM3e Memory: 288GB per GPU, 1.5x larger vs. previous gen; Improved context length support and inference efficiency
- NVIDIA ConnectX-8 SuperNIC: Dual-device IO module delivering 800 Gb/s per GPU; Enables high-throughput, low-latency scale-out networking
- Rack-Scale Liquid Cooling: Modular CDU options (In-Row, In-Rack, Sidecar); Full lifecycle DLC service for deployment and operations
- 1.5x LLM Training Speed vs GB200
- 1.4x LLM Inference Throughput vs GB200
- AI Training & Inference — Optimized for trillion-scale models, transformer networks, and real-time inference.
- Big Data Analytics — Accelerates data pipelines, reduces storage cost, and improves query performance
- Scientific & Engineering Simulations — Enhances compute-intensive modeling, from CFD to circuit design.

## Identity

| Field | Value |
|---|---|
| Family | KRS8500 |
| Model numbers | see datasheet |
| CPU platform | NVIDIA Grace |
| GPU platform | NVIDIA GB300 NVL72 |
| Form factor | Rack-scale (NVIDIA GB300 NVL72, 18x 1U compute trays + 9x 1U NVLink switch trays) |
| Cooling | Liquid cooling (rack-scale; CPU/GPU/CX8 liquid-cooled, others air; CDU options L2L in-row / L2L in-rack / L2A sidecar) |
| Category | gpu |

## Specifications (aivres.com product page)

| Item | Specification |
|---|---|
| Configuration | 72x Blackwell Ultra GPUs, 36x Grace CPUs |
| NVLink Bandwidth | 130TB/s |
| Fast Memory | Up to 40 TB |
| GPU Memory \| Bandwidth | Up to 21 TB \| Up to 576 TB/s |
| CPU Memory \| Bandwidth | Up to 18 TB LPDDR5X \| Up to 14.3 TB/s |
| CPU Core Count | 2,592 Arm® Neoverse V2 cores |
| FP4 Tensor Core | 1,080–1,400 PFLOPS (rack-level) |
| FP8/16 Tensor Core | 720 PFLOPS |
| INT8 Tensor Core | 23 PFLOPS |
| FP16/BF16 Tensor Core | 360 PFLOPS |
| TF32 Tensor Core | 180 PFLOPS |
| FP32 Tensor Core | 6 PFLOPS |
| FP64 / FP64 Tensor Core | 100 TFLOPS |
| Rack Specifications: Dimensions (W x H x D) | 600mm x 2295mm x 1200mm (23.62” x 90.35” x 47.24”) |
| Rack Specifications: Weight | Weight 1,472–1,487 kg (3,245–3,279 lbs) |
| Rack Specifications: NVL Config | 72x 1 |
| Rack Specifications: NV OOB Switch | Option 1: 2x SN2201_M; Option 2: 3x SN2201_M; Option 3: 4x SN2201_M |
| Rack Specifications: NVL Cartridge | 4 |
| Rack Specifications: Rack Type (Per Rack) | 9x 1U NVlink Switch Trays; 18x 1U Compute Trays; 8x 1U Power Shelves |
| Rack Specifications: Power-Shelf | (6+2) x 33kW |
| Rack Specifications: Busbar | 1,400A |
| Rack Specifications: Rack Manifold | Option 1: 44RU, BF; Option 2: 44RU, TF |
| Rack Specifications: CDU | Option 1: L2L In-Row; Option 2: L2L In-Rack; Option 3: L2A Sidecar |
| Compute Tray: CPU/GPU | 2x Grace CPUs + 4x Blackwell Ultra GPUs |
| Compute Tray: Count | 18 per Rack |
| Compute Tray: Cooling | CPU/GPU/CX8 liquid-cooled, others air cooling |
| Compute Tray: Storage | 8x E1.S NVMe SSDs |
| Compute Tray: M.2 | 1 x M.2 NVMe SSD |
| Compute Tray: Front I/O | 1x USB 3.0, 1x Mgmt I/O , 1x RJ45, 1x mini display port |
| Compute Tray: N-S Networking | 1x FHFL PCIe Gen 5 X16 (BF3) |
| Compute Tray: E-W Networking | 2 x CX8 Mezzanine on board (2x CX8 per Mezzanine) for 4 x 800G OSFP Conn |
| Compute Tray: Fan | CPU region: 8x 12V 4056 hot-swap fans with N+1 redundancy |
| Compute Tray: Management | DC-SCM BMC management module |
| Compute Tray: TPM | Supports TPM 2.0 |
| Switch Tray: Type | NVLink N5500_LD |
| Switch Tray: Count | 9 per Rack |
| Switch Tray: Bandwidth | 14.4TB/s |
| Switch Tray: Cooling | 1U liquid cooled |
| Switch Tray: Front IO | 2x RJ45, 1x USB, 1x UART |

## Documents

- Datasheet, parsed (2 pp): [Markdown](https://aivres.upstation.io/docs/krs8500v3/datasheet.md) · [HTML](https://aivres.upstation.io/docs/krs8500v3/datasheet.html) · [original PDF](https://aivres.com/wp-content/uploads/KRS8500V3-Datasheet.pdf)
- Product page: https://aivres.com/product/krs8500v3/ · family page: https://aivres.com/product/krs8500v3/

## Images

Hotlinkable from aivres.com; figure crops hosted here. For presentations about this product only.

- hero: https://aivres.com/wp-content/uploads/hero-KRS8500V3-0-scaled-1.webp
- product: https://aivres.com/wp-content/uploads/KRS8500V3-Showcase.webp
- other: https://aivres.com/wp-content/uploads/krs8500-features-360px.webp
- figure from the datasheet: https://aivres.upstation.io/docs/krs8500v3/images/reducto/p001_figure_0003.jpg
- figure from the datasheet: https://aivres.upstation.io/docs/krs8500v3/images/reducto/p001_figure_0007.jpg
- figure from the datasheet: https://aivres.upstation.io/docs/krs8500v3/images/reducto/p001_figure_0028.jpg

## Lead time

- Not yet confirmed with Aivres; quoted with the price.

## Pricing

Pricing is being confirmed with Aivres and is not shown yet. Request a quote below.

## Terms

- One bundled price per configured system; quotes are valid 7 days; the price locks when the order is processed
- 25% deposit within 3 days of an NCNR purchase order, 75% before shipment; DAP delivery, continental US; 3-year standard warranty
- NVIDIA Inception discounts apply only when the approval code is on file before the order is processed
- US-sited deployments are handled by Aivres US; buyers outside North America are routed to an Aivres regional desk or handled through UpStation

## How to buy

Standard NDA → New-customer account form, W-9 or resale certificate, NVIDIA end-user statement, business licence → Quotation table (the intake fields below) → Quote with lead time, then purchase order and sales agreement → Pro forma, 25% deposit, compliance check (purchaser / end user / ship-to / HQ / parent / payer / destination / installation) → Ship.

## Get a firm quote

Email alec@upstation.io with the fields below, or open this prepared email: mailto:alec@upstation.io?subject=Quote%20request%3A%20Aivres%20KRS8500V3%20%28via%20aivres.upstation.io%29&body=End%20user%20or%20reseller%3A%20%0AEnd-user%20name%20and%20address%3A%20%0AShip-to%20address%3A%20%0APlatform%20%28for%20example%20KR9288-X3%20Intel%20or%20KR9288-E3%20AMD%29%20and%20configuration%3A%20%0AQuantity%3A%20%0AService%3A%203-year%20or%205-year%2C%20NBD%20on-site%20yes/no%2C%20white-glove%20yes/no%3A%20%0AEstimated%20PO%20date%20and%20required%20ship%20date%3A%20%0ATarget%20price%3A%20%0AConfidence%20%28%25%29%3A%20%0A%0ASource%3A%20https%3A//aivres.upstation.io/p/krs8500v3%0A

- End user or reseller
- End-user name and address
- Ship-to address
- Platform (for example KR9288-X3 Intel or KR9288-E3 AMD) and configuration
- Quantity
- Service: 3-year or 5-year, NBD on-site yes/no, white-glove yes/no
- Estimated PO date and required ship date
- Target price
- Confidence (%)

Data as of 2026-10-15.