Lenovo 4XG7A90288 Thinksystem SR665 V3 AMD Epyc 9734 112C 340W 2.2GHZ Processor W/o FAN

Lenovo 4XG7A90288 AMD EPYC 9734 112-Core Processor for ThinkSystem SR665 V3

Overview

The Lenovo 4XG7A90288 is a processor option kit pairing the AMD EPYC 9734 with Lenovo's ThinkSystem SR665 V3 server platform. At 112 cores and 224 threads running at a 2.2 GHz base clock — boosting to 3.0 GHz under load — this is a CPU designed for workloads that demand extreme thread density: large-scale AI inferencing, high-channel-count network video recorders, dense virtualization hosts, and HPC clusters where per-socket core count directly limits throughput. If you're sizing compute for a video analytics platform or a datacenter consolidation project, this processor warrants serious consideration. Note: the kit ships without a heatsink; a compatible fan/heatsink must be sourced separately for the SR665 V3 chassis.

Key Features

• 112 Cores / 224 Threads per Socket: The EPYC 9734 delivers more logical processors per socket than any mainstream x86 alternative in this class. For server platforms running concurrent AI video analytics streams or dense VM workloads, that thread count translates directly into how many simultaneous tasks you can assign without context-switching overhead eating into throughput.
• 2.2 GHz Base / 3.0 GHz Boost: The base clock sets a reliable floor for sustained, all-core workloads — important when you're running 100+ threads continuously rather than bursty single-threaded tasks. The 3.0 GHz boost gives responsive headroom for latency-sensitive operations that hit fewer cores at once.
• 340 W TDP with 320–400 W Configurable Range: The 340 W nominal TDP is substantial — this chip is a serious power draw. The configurable TDP-down to 320 W gives operators in power-constrained datacenters a way to trade a modest amount of peak performance for better per-rack power budget compliance. Conversely, the 400 W TDP-up envelope is available when the chassis cooling and PDU capacity supports it, squeezing additional boost headroom out of the silicon.
• 256 MB L3 Cache: With 256 MB of L3 cache spread across 112 cores, working datasets for analytics models, database hot-paths, or in-memory caches are far more likely to stay on-die rather than spilling to main memory — which matters most in latency-sensitive, high-throughput pipelines.
• 460.8 GB/s DDR5 Memory Bandwidth via Dodeca-Channel: Twelve DDR5 memory channels deliver 460.8 GB/s of aggregate bandwidth. For AI inference workloads, memory bandwidth is frequently the binding constraint — not compute. This bandwidth floor means large model weights and activation buffers move fast enough to keep all 112 cores fed. Pair with high-speed DDR5 DIMMs to realize this ceiling.
• Socket SP5 Platform: The SP5 socket is shared across AMD's EPYC Genoa/Bergamo processor family, which means future processor upgrade paths within the SR665 V3 platform don't require a motherboard swap — relevant for organizations planning a multi-year infrastructure roadmap.
• No Integrated Graphics: This is a headless compute processor — there is no on-board GPU or display output. Remote management via the SR665 V3's BMC/XCC interface is the expected access method. Plan accordingly if you're provisioning bare-metal with no IPMI/BMC infrastructure.

Integration & Compatibility

The 4XG7A90288 is validated for the Lenovo ThinkSystem SR665 V3 platform specifically. The SR665 V3 is a dual-socket 2U rack server; this processor option kit covers one socket — a second kit is required to populate both sockets in a dual-socket configuration. Verify your chassis firmware level and Bill of Materials with Lenovo's ServerProven compatibility tool before ordering to ensure the SR665 V3 system board revision supports this specific EPYC 9734 stepping. DDR5-SDRAM is the required memory type; DDR4 DIMMs are not compatible with the SP5 platform. For customers deploying this hardware in video surveillance infrastructure, integration with video management platforms and AI analytics engines benefits directly from the 224-thread compute density — particularly for concurrent multi-stream deep learning inference. Cooling note: this kit does not include a heatsink or fan module. Confirm the correct Lenovo heatsink option for your SR665 V3 airflow configuration (standard vs. high-performance) before deployment.

Frequently Asked Questions

Q: Does the Lenovo 4XG7A90288 include a heatsink or fan?

A: No. The 4XG7A90288 is explicitly listed as a processor option without a fan. You must order a compatible heatsink and fan module separately for the ThinkSystem SR665 V3 chassis. Confirm the correct part number with Lenovo's configurator based on your chassis airflow tier.

Q: How many of these processors does the ThinkSystem SR665 V3 support?

A: The SR665 V3 is a dual-socket platform. This kit populates one socket. A second 4XG7A90288 (or compatible EPYC SP5 processor) is required to fully populate both sockets for maximum core density.

Q: What memory type is required with the EPYC 9734?

A: The EPYC 9734 requires DDR5-SDRAM. It supports dodeca-channel (12-channel) memory for up to 460.8 GB/s aggregate bandwidth. DDR4 is not compatible with the SP5 socket platform.

Q: What is the thermal design power of the 4XG7A90288?

A: The nominal TDP is 340 W. It is configurable: TDP-down to 320 W for power-constrained environments, or TDP-up to 400 W when chassis cooling and power delivery support the higher envelope.

Q: Is this processor compatible with other AMD EPYC server platforms?

A: The EPYC 9734 uses the SP5 socket, which is shared across AMD's Genoa/Bergamo family. However, the 4XG7A90288 is a Lenovo option kit validated specifically for the ThinkSystem SR665 V3. Using this kit in non-Lenovo platforms voids Lenovo's validation and support coverage. Always verify platform compatibility via Lenovo ServerProven before deployment.

Q: What is the processor's boost frequency?

A: The EPYC 9734 boosts to 3.0 GHz from a 2.2 GHz base clock. Boost frequency applies to lightly-threaded workloads hitting a subset of the 112 available cores; sustained all-core workloads run closer to the base clock floor.

Eden Phillips

The 4XG7A90288 is one of the highest core-density processor options available for the SR665 V3 platform, and that 112-core / 224-thread count is the primary reason you'd choose it over a lower-core-count EPYC variant. I want to be direct about what that means operationally: at 340 W nominal TDP, you're committing to serious rack power and cooling infrastructure before the first workload runs. If your facility is running 30A circuits per rack with standard 2U airflow, plan your power draw carefully before ordering multiples of this kit.

Technical Highlights:

• 112 Cores / 224 Threads: For parallel AI inferencing or high-density VM hosting, this is the spec that drives the purchase decision — more threads per socket means more concurrent workloads without adding rack space or licensing overhead.
• 460.8 GB/s DDR5 Bandwidth: Twelve DDR5 channels at this aggregate bandwidth level keeps memory from becoming the bottleneck in large model inference pipelines — a genuine constraint on prior-generation EPYC and Intel Xeon platforms running comparable workloads.
• Configurable TDP 320–400 W: The 80 W configurability window gives operators meaningful flexibility — run at 320 W TDP-down in dense racks with tight power budgets, or open up to 400 W when the chassis fans and PDU headroom support it for maximum boost headroom.

Deployment Considerations:

• This kit ships without a heatsink — don't assume the SR665 V3 chassis comes with one pre-installed for the second socket. Confirm the correct Lenovo heatsink part number for your airflow configuration before the server ships to site, or you'll be waiting on a separate order before the system can power on safely.
• At 256 MB of L3 cache across 112 cores, the per-core cache allocation is relatively modest compared to lower-core-count EPYC variants. Workloads that are cache-sensitive and single-threaded (certain database query patterns) may perform better on a lower-core, higher-clock EPYC option — the 9734 is optimized for throughput at scale, not for latency at low concurrency.

This processor is a strong fit for organizations deploying large-scale AI video analytics infrastructure — specifically environments running 100+ concurrent deep learning inference streams per node, where maximizing thread density per 2U chassis slot directly reduces the number of servers (and licenses) required to hit throughput targets.