iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Cornelis Networks, the Intel spin-off behind Omni-Path, is positioning its 400-Gbps CN5000 fabric as an alternative to Nvidia InfiniBand and RoCE Ethernet for AI and HPC data centers. Its claimed edge is better price-performance through higher message rates, lower latency, and faster collective communication—not simply higher port bandwidth. The published evidence remains primarily vendor-originated.
Cornelis is reviving and commercializing technology that Intel separated into an independent company in 2020 after Intel stopped pursuing Omni-Path as a data-center product. Lisa Spelman, a former Intel executive, became Cornelis CEO in 2024. The company now sells a complete fabric rather than a standalone adapter: SuperNICs, switches, director-class switches, cabling, host software, management tools, and OpenFabrics support.
Intel Spin-Off: Our InfiniBand Alternative for AI Data Centers Has a ‘Devastatingly Good’ Edge
The phrase “devastatingly good” comes from Cornelis CEO Lisa Spelman’s positioning of the company’s price-performance advantage; it is not an independently established industry verdict. Cornelis says CN5000 can outperform comparable 400-Gbps Nvidia InfiniBand and selected RoCE configurations on message rate, latency, HPC workloads, and collective communication.
Free tools Windows power users keep installed
One-click scans. No signup required.
The distinction matters. A network can have an attractive microbenchmark result without shortening a real model-training run. Buyers need to test the complete stack—accelerators, adapters, switches, firmware, drivers, collective libraries, topology, storage, and applications—before treating Cornelis’s claims as a purchasing conclusion.
#1 Best Overall
- Item Package Quantity - 1
- Product Type - ELECTRONIC SWITCH
- Model Number - MSB7890-ES2F
- Accessories may not be original, but will be compatible and fully functional. Product may come in generic box.
Key takeaways
- Cornelis Networks spun out of Intel in 2020 and now develops Omni-Path independently for AI, HPC, government, and academic data centers.
- The CN5000 family provides up to 400 Gbps per port, with adapters, 48-port fixed switches, director-class switches, cabling, and software.
- Cornelis claims two times higher message rates, 35% lower latency, up to 30% faster selected HPC workloads, and up to six times faster collective communication than cited alternatives.
- The performance figures are Cornelis-attributed claims, not neutral benchmark results established by the available reporting.
- Nvidia’s current comparison point is the newer 800-Gbps Quantum-X800 InfiniBand platform, so 400-Gbps CN5000 and 800-Gbps Nvidia products are not equivalent generations.
- CN5000 is most compelling for buyers willing to qualify a newer fabric and seeking multi-vendor flexibility; Nvidia remains the lower-risk choice for customers prioritizing a mature, validated ecosystem.
What is Cornelis Networks?
Cornelis Networks is the independent company commercializing Intel’s former Omni-Path data-center interconnect technology. Intel separated the technology into Cornelis in 2020, and Cornelis continued developing the architecture after Intel halted its own data-center Omni-Path program. CRN’s 2025 report on the spin-off and CN5000 describes the company’s effort to position Omni-Path against InfiniBand and Ethernet for large distributed workloads.
Lisa Spelman, a former Intel executive, became Cornelis CEO in 2024. Cornelis focuses on high-performance computing, AI training and inference, government and academic supercomputing, and enterprise applications in which many servers or accelerators must communicate closely.
The company sells through direct customer engagement, server OEMs, systems integrators, and channel partners. The CRN report cited relationships with Dell, HPE, Lenovo, and Supermicro, along with channel partners including Ace Computers, Advanced Clustering Technologies, Colfax International, Microway, Penguin Solutions, and Aspen Systems.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
What exactly is the CN5000 family?
CN5000 is a complete 400-Gbps Omni-Path fabric family, not a single chip or switch. The family combines host adapters, fixed-form-factor switches, director-class switches, copper and optical cabling, host and management software, open-source components, and OpenFabrics support. Cornelis describes the family in its official CN5000 overview.
| CN5000 component | Published specification | Why it matters |
|---|---|---|
| SuperNIC adapters | Up to 400 Gbps bidirectional; one- or two-port configurations; PCIe 5.0 | Connects servers and accelerator systems to the fabric |
| Adapter message rate | More than 1 billion MPI packets per second claimed | Targets workloads that exchange many relatively small messages |
| Adapter latency | As low as under 1 microsecond, according to Cornelis | Relevant to synchronization and tightly coupled workloads |
| Fixed switch | 48 × 400-Gbps ports; 38.4 Tbps aggregate bandwidth | Provides a building block for smaller and mid-sized clusters |
| Port subdivision | 32 ports can be subdivided into 2 × 200-Gbps ports | Adds topology and connectivity flexibility |
| Director-class switch | Up to 576 × 400-Gbps ports; up to 230.4 Tbps aggregate bandwidth | Targets very large cluster fabrics |
The adapter specifications are listed on the CN5000 Omni-Path adapter page. The fixed-switch figures appear in Cornelis’s CN5000 switch documentation, while the maximum director configuration appears on the CN5000 Director Class Switch page.
Cornelis materials describe CN5000 as supporting clusters exceeding 100,000 nodes. A 2025 CRN report also described a company claim of capability for a 500,000-node deployment. Those are scale claims, not evidence that every customer can deploy such a system with the same topology, software, failure characteristics, or application performance.
How does Omni-Path differ from InfiniBand and RoCE?
Omni-Path, InfiniBand, and RoCE are scale-out interconnect choices: they connect servers, GPUs, CPUs, and other accelerators across a cluster. They are not replacements for the internal GPU-to-GPU links inside a server.
| Fabric | Core strength | Main qualification issue |
|---|---|---|
| Cornelis Omni-Path | Purpose-built, lossless, congestion-managed fabric with an emphasis on message rate, efficiency, and multi-vendor systems | Smaller and newer ecosystem requires careful validation of hardware, software, and support |
| Nvidia InfiniBand | Mature, tightly integrated AI/HPC networking stack with established tools and validated configurations | Can increase dependence on Nvidia’s hardware and software ecosystem |
| RoCE Ethernet | Familiar Ethernet operations, broad supplier choice, and potential reuse of existing infrastructure | Predictable RDMA performance can require careful congestion, buffering, topology, and firmware tuning |
| Ultra Ethernet-based fabrics | Intended to extend Ethernet’s supplier and operational advantages to large AI systems | Standards, interoperable products, and production maturity must be distinguished from roadmap claims |
Cornelis positions Omni-Path as natively lossless, credit-flow-controlled, adaptive, and compatible with standard CPUs, GPUs, and accelerators. The company also emphasizes OpenFabrics Interface optimization and Verbs compatibility. Compatibility with those interfaces does not guarantee that every application will achieve the same performance as a vendor-validated InfiniBand configuration.
RoCE, or RDMA over Converged Ethernet, uses Ethernet infrastructure to provide remote direct memory access. Ethernet can be the better choice for organizations with strong Ethernet expertise, broad existing infrastructure, and a preference for component-level vendor choice. Ethernet is not inherently too slow for AI, and InfiniBand is not universally superior; topology, workload, congestion control, software, operations, and total cost determine the result.
Why does Cornelis claim CN5000 has an edge?
Cornelis’s technical argument is that a purpose-built, lossless fabric can spend less time recovering from congestion and can move synchronization traffic more efficiently. The argument centers on several architectural features.
Credit-based flow control
Credit-based flow control allows a sender to transmit according to the buffer capacity available at the receiver or downstream link. The design aims to prevent buffer overflow, reduce packet loss, and avoid the retransmission or recovery work associated with congestion.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Lossless, congestion-managed transport
Cornelis describes CN5000 as lossless and congestion-free, but the safer interpretation is that the fabric is designed to provide lossless, congestion-managed transport. Real clusters still experience contention, incast, hot spots, and uneven traffic. Flow control manages those conditions; it does not make traffic contention impossible.
Fine-grained adaptive routing
Dynamic adaptive routing can select paths based on current traffic conditions rather than relying entirely on a static route. Cornelis says fine-grained routing helps avoid hot spots and improves link utilization.
Traffic categorization and quality of service
CN5000 switches include traffic controls intended to separate or prioritize traffic patterns, including incast-prone applications. Quality-of-service controls can matter when collective operations, storage traffic, management traffic, and ordinary application messages share infrastructure.
Telemetry and error handling
Cornelis describes telemetry engines and visibility tools for identifying bottlenecks and optimizing flows. Published product material also describes bit-error recovery, optional error correction, and packet-integrity protections. The CN5000 switch page and the CN5000 product brief provide the relevant vendor specifications.
What performance advantage does Cornelis claim?
Cornelis claims that CN5000 improves communication efficiency compared with selected alternatives. The available source does not establish these results as independent industry benchmarks, and the exact test methodology is essential.
| Metric | Cornelis-attributed claim | Stated comparison | How to interpret it |
|---|---|---|---|
| Message rate | Two times higher | Nvidia 400-Gbps InfiniBand NDR comparison | Potentially important for small-message and synchronization-heavy workloads |
| Latency | 35% lower | Nvidia 400-Gbps InfiniBand NDR comparison | Requires exact message size, topology, firmware, and percentile definition |
| HPC performance | Up to 30% faster | Selected workloads including computational fluid dynamics | Workload-specific; not a universal claim about all HPC applications |
| Collective communication | Up to six times faster | Cited RoCE solutions | Highly sensitive to collective library, congestion settings, topology, and node count |
According to the 2025 CRN interview with Cornelis CEO Lisa Spelman, Cornelis reported those comparisons as part of its case for higher accelerator utilization and lower total cost of ownership. The claims should be written as “Cornelis says its testing showed” rather than “CN5000 is 30% faster.”
The comparison also needs a fair baseline. Buyers should ask which Nvidia adapters and switches were tested, which RoCE implementation and congestion-control settings were used, what packet and message sizes were selected, how many nodes were connected, which topology was used, and whether the comparison matched port count, power, hardware cost, and nominal bandwidth.
Rank #3
- Performance
- 40 X HDR 200Gb/s ports in a 1U switch
- 80 X HDR100 100Gb/s ports (using splitter cables)
- 16Tb/s aggregate switch throughput
- Sub-90ns switch latency
AI results deserve additional caution. A point-to-point message-rate test or collective microbenchmark may not predict end-to-end training time. A credible proof of concept should use the customer’s actual model architecture, batch sizes, node count, collective patterns, storage path, software versions, and production topology.
Why does GPU utilization matter to the business case?
GPU utilization matters because expensive accelerators can sit idle while they wait for data, synchronization, or collective communication. Cornelis’s economic thesis is that higher message rates, lower communication latency, and faster collectives can reduce those stalls.
- More efficient communication can shorten synchronization phases.
- Shorter synchronization phases can increase useful accelerator time.
- Higher accelerator utilization can increase throughput on the same hardware.
- A customer may then complete a target workload with fewer accelerators or complete more work in the same time.
The inference is reasonable at the systems level, but the benefit depends on the workload. Gains may be small when a model is communication-light, local computation dominates, storage or data loading is the bottleneck, the application does not use optimized collectives, or the cluster is too small for fabric contention to matter.
How does CN5000 compare with Nvidia’s current InfiniBand platform?
CN5000 should be compared with 400-Gbps Nvidia InfiniBand NDR when evaluating Cornelis’s original performance claims, but Nvidia’s current official comparison point is the newer 800-Gbps Quantum-X800 platform.
| Comparison point | Cornelis CN5000 | Nvidia Quantum-X800 direction |
|---|---|---|
| Published port bandwidth | Up to 400 Gbps per port | 800 Gbps InfiniBand connectivity |
| Switch scale cited in dossier | Fixed switch: 48 × 400 Gbps; director: up to 576 × 400 Gbps | Quantum-X800 switch: 144 × 800-Gbps ports |
| Accelerator-side claim | One- or two-port 400-Gbps adapters | ConnectX SuperNICs offering up to 1.6 Tbps per GPU |
| Network features | Credit-based flow control, adaptive routing, telemetry, traffic controls, and lossless transport design | SHARP v4 in-network computing, adaptive routing, telemetry-based congestion control, and large-scale AI support |
| Evidence and maturity | Vendor claims and product specifications; buyers must qualify the complete stack | Established Nvidia ecosystem and published validated configurations |
The Nvidia Quantum-X800 product page lists the 800-Gbps platform, while Nvidia’s validated-configuration documentation provides tested firmware and software combinations for supported clusters. That validation can materially reduce deployment risk even when a competing fabric presents an attractive benchmark.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →A current buyer should separate three questions: whether CN5000 is faster than a particular 400-Gbps NDR configuration, whether CN5000 is cheaper or more efficient than Nvidia’s complete current platform, and whether Cornelis’s future 800-Gbps products will match Nvidia’s current ecosystem maturity. Those questions do not have the same answer.
Is RoCE Ethernet a better alternative?
RoCE Ethernet can be the better alternative when an organization values existing Ethernet skills, broad supplier availability, and the ability to reuse operational practices and infrastructure. RoCE also offers flexibility in selecting switches, NICs, optics, network operating systems, and monitoring tools.
The trade-off is configuration and qualification effort. Predictable RDMA performance may depend on priority flow control, explicit congestion notification, buffer sizing, queue configuration, topology, firmware, and careful tuning. Performance can vary substantially between implementations. The lowest-cost Ethernet component is not necessarily the lowest-cost working AI fabric once engineering, validation, support, and troubleshooting are included.
Cornelis’s response is to add Ethernet protocols to later products. As of August 16, 2026, Cornelis’s official listing describes CN6000 as an 800-Gbps multi-protocol SuperNIC supporting Omni-Path, RoCEv2, and Ultra Ethernet on PCIe 6.0. The listing establishes how Cornelis describes the product; buyers should separately confirm shipping status, ordering availability, supported software, interoperability, and production references.
Rank #4
![Mellanox InfiniBand MSB7700-ES2F [SB7700] 36-Port QSFP28 100GbE EDR Network Switch w/ Dual PSU [No Rails Kit] (Renewed)](https://m.media-amazon.com/images/I/218as6o1fnL._SL160_.jpg)
- Includes: 2x Power Supply Units, 4x Fans, 2x Power Cords, 1x Console Cable, 1x Rack Kit
- NO RAIL KIT
What is the CN6000 roadmap status?
The original 2025 story described CN6000 as a future 800-Gbps Ethernet-compatible generation planned for 2026 and CN7000 as a 1.6-Tbps generation planned for 2027. The roadmap needs updating: as of August 16, 2026, Cornelis’s official materials list CN6000 as an 800-Gbps multi-protocol SuperNIC with Omni-Path, RoCEv2, and Ultra Ethernet support on PCIe 6.0.
| Product or claim | Status supported by the dossier | Buyer caution |
|---|---|---|
| CN5000 | Current 400-Gbps product family with adapters and switches | Validate exact availability, software release, and supported system configuration |
| CN6000 | Officially listed as an 800-Gbps multi-protocol SuperNIC | “Listed” does not by itself prove universal shipping, interoperability, or production maturity |
| CN7000 | Earlier reporting described a 1.6-Tbps 2027 generation | Treat as roadmap information, not a present purchasing option |
A roadmap can reduce adoption risk only when the vendor provides a credible migration path: compatible software interfaces, upgradeable management, clear cable and optics plans, performance targets, support commitments, and commercially available hardware. A buyer should not purchase CN5000 solely on the assumption that future Ethernet or Ultra Ethernet compatibility will arrive on schedule.
Who should consider CN5000?
CN5000 is most worth evaluating when network communication is a meaningful part of the workload and the buyer can run a serious proof of concept.
- HPC centers: MPI- and SHMEM-heavy applications can benefit from high message rates, low latency, and tightly managed congestion.
- Government and academic supercomputing: These organizations may value an alternative supplier and may have the engineering expertise to qualify a newer fabric.
- AI training clusters: CN5000 deserves testing where all-reduce, all-gather, or other collectives materially limit accelerator utilization.
- Server builders and integrators: Multi-vendor CPU, GPU, and accelerator systems may offer more component flexibility than a fixed integrated platform.
- Flexible HGX-based deployments: Cornelis CEO Lisa Spelman identified Nvidia HGX systems as a stronger opportunity than fixed Nvidia DGX systems because HGX customers have more flexibility over additional components, according to CRN’s report.
Who should probably avoid CN5000?
CN5000 is a weaker fit for buyers who value ecosystem maturity more than fabric experimentation or who cannot support a detailed qualification process.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minutePC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11- Fixed Nvidia DGX environments: Cornelis’s CEO identified DGX systems as outside the company’s target area, and the exact platform may not permit the desired fabric substitution.
- Small clusters: The communication gains may not offset engineering and support costs when the cluster is too small for network contention to dominate.
- Organizations without fabric specialists: Driver, firmware, topology, collective-library, telemetry, and failure-recovery validation require appropriate expertise.
- Buyers needing the broadest established tooling: Nvidia has a larger installed base, more familiar administrative patterns, extensive documentation, and validated configurations.
- Cloud or multi-tenant operators seeking a globally standardized platform: A newer specialized fabric may have less operational precedent and fewer established integrations.
What should buyers validate before choosing CN5000?
Buyers should treat CN5000 as a system-level procurement decision, not a port-speed comparison. The following checklist identifies the minimum technical, commercial, and operational questions.
Technical validation
- Measure required bandwidth per accelerator and per node, including oversubscription and the planned topology.
- Measure average and tail latency, not only a single latency number.
- Benchmark all-reduce, all-gather, broadcast, reduce-scatter, and the collectives used by the target applications.
- Test the actual GPU, CPU, accelerator, PCIe generation, operating system, kernel, driver, firmware, MPI, OFI, Verbs, and collective-library versions.
- Confirm GPU-direct or equivalent direct data paths rather than assuming that GPU compatibility guarantees optimized transfers.
- Measure real training time, tokens per second, HPC application completion time, GPU utilization, and power per completed job.
- Test mixed traffic, incast, congestion, link failures, switch failures, cable faults, and recovery behavior.
- Confirm interoperability with storage, parallel file systems, schedulers, cluster provisioning, monitoring, and management networks.
Commercial validation
- Request a complete-fabric quote covering adapters, switches, director hardware, optics, copper and optical cables, software, support, installation, warranty, and replacement parts.
- Request a comparable Nvidia quote covering Quantum-X800 or the relevant InfiniBand generation, ConnectX SuperNICs, cables, optics, software, support, and services.
- Compare three-year total cost of ownership rather than NIC or switch prices alone.
- Include power, cooling, rack density, staff training, engineering time, qualification, migration, and failure-replacement costs.
- Ask for independent or reproducible benchmark evidence using the buyer’s intended application and topology.
- Confirm firmware lifecycle, driver support, security-update procedures, response times, and vendor financial and roadmap commitments.
Proof-of-concept test plan
- Build equivalent CN5000, Nvidia InfiniBand, and—where relevant—RoCE test configurations.
- Match GPU and CPU hardware, node count, port count, topology, cable quality, software versions, firmware, power limits, and tuning effort.
- Run microbenchmarks for point-to-point bandwidth, message rate, average latency, and tail latency.
- Run the actual AI models and HPC applications at the intended scale.
- Record end-to-end job time, tokens per second, collective-operation time, GPU utilization, power per job, and failure-recovery time.
- Repeat tests under mixed traffic and realistic storage conditions.
- Compare the measured result with the complete installed cost and the operational effort required to achieve it.
What are the main adoption risks?
The largest trade-off is specialized performance versus ecosystem maturity. Cornelis may offer attractive performance and price-performance, but Nvidia InfiniBand has a larger established ecosystem around adapters, switches, software, validated configurations, training, documentation, and support.
Open interfaces can reduce dependence on one hardware supplier, but open interoperability increases the buyer’s qualification responsibility. A system may support standard CPUs, GPUs, OFI, and Verbs while still requiring vendor-specific drivers, firmware, management tools, and collective-library tuning. “Works with Nvidia GPUs” is therefore not the same as “delivers the same production performance as an Nvidia-validated networking stack.”
Another risk is a misleading baseline. A vendor comparison may look favorable if the competing system is poorly configured. Equal port speed is not enough: the comparison must match switch generation, optics, topology, CPU and GPU hardware, software, firmware, node count, tuning, power, and cost.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitchesFinally, a microbenchmark can succeed while the application disappoints. Storage, data loading, local computation, model architecture, batch size, collective implementation, and cluster size can overwhelm a fabric advantage. Buyers should demand application-level evidence.
How much does CN5000 cost?
No public list price was found in the reviewed Cornelis product pages as of August 16, 2026. Cornelis uses an “Enquire Now” buying path, so buyers should expect quote-based enterprise procurement through Cornelis, an OEM, or a systems integrator. Nvidia also does not publish list pricing on the reviewed pages and directs buyers to partners, distributors, its marketplace, and sales channels through its official networking purchasing page.
Because neither side supplies a comparable public price in the reviewed sources, no honest conclusion can say that CN5000 is cheaper in absolute terms. The relevant comparison is complete installed cost: adapters, switches, optics, cabling, software, support, power, cooling, installation, engineering, qualification, and the value of faster or more reliable completed workloads.
What is the defensible verdict on Cornelis CN5000?
Cornelis has a technically credible alternative architecture and a potentially compelling economic argument, but “devastatingly good” remains a vendor claim until independent, application-level testing confirms the advantage across representative AI workloads.
Recommended Free Tools
CN5000 makes the strongest case where communication is a major bottleneck, the buyer wants an alternative to Nvidia’s vertically integrated networking stack, and the organization can qualify hardware and software at production scale. Nvidia remains the lower-risk purchase for customers that prioritize mature tooling, broad support, validated configurations, and a tightly integrated InfiniBand platform.
The practical decision is not “Omni-Path or InfiniBand by headline bandwidth.” The practical decision is whether CN5000 delivers better completed-job economics on the buyer’s actual models and applications after qualification, operations, support, migration, and roadmap risk are included.
Frequently Asked Questions
Is Cornelis CN5000 a replacement for Nvidia InfiniBand?
Cornelis CN5000 is an alternative scale-out fabric to Nvidia InfiniBand, but it is not a universal drop-in replacement. Buyers must validate adapters, GPU-direct paths, drivers, collective libraries, MPI, topology, scheduler integration, storage behavior, monitoring, and application performance.
Are Cornelis’s two-times, 35%, 30%, and six-times performance figures independently verified?
The available figures are claims attributed to Cornelis, not neutral benchmark results established by the reviewed reporting. Buyers should request test methodology and reproduce the results with equivalent hardware, firmware, software, topology, node count, and real workloads.
Recommended Free Tools
Is CN5000 faster than RoCE Ethernet?
CN5000 may outperform selected RoCE configurations on message rate, latency, or collective communication, according to Cornelis, but CN5000 is not universally faster than every RoCE network. RoCE performance depends heavily on switch configuration, congestion control, buffering, firmware, topology, libraries, and tuning.
Is Cornelis CN6000 available now?
As of August 16, 2026, Cornelis’s official materials list CN6000 as an 800-Gbps multi-protocol SuperNIC supporting Omni-Path, RoCEv2, and Ultra Ethernet on PCIe 6.0. The listing alone does not establish universal ordering, shipping, interoperability, or production maturity, so buyers should confirm availability directly.
Who is the best fit for Cornelis CN5000?
CN5000 is best suited for HPC centers, government and academic supercomputing, AI training clusters with communication bottlenecks, and flexible server or HGX-based systems where buyers can run a detailed proof of concept. Fixed DGX environments, small clusters, and teams without fabric expertise may face higher adoption risk.
The Bottom Line
Bottom line: Cornelis CN5000 is a serious 400-Gbps Omni-Path alternative for AI and HPC—not a proven universal “InfiniBand killer.” Its claimed advantage is network efficiency and price-performance, supported by an architecture built around lossless flow control, adaptive routing, congestion management, and high message rates. The purchasing decision should rest on independent, application-level testing and complete-fabric cost, not vendor headline percentages.
Free tools Windows power users keep installed
One-click scans. No signup required.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
