Today, we sit down with Dominic Jainy, a renowned expert in high-performance computer hardware, to dissect the latest revelations in the workstation CPU space. Leaked benchmarks for Intel’s upcoming Xeon 654 “Granite Rapids-WS” processor have stirred the market, painting a complex picture of impressive generational gains, a fierce rivalry with AMD, and a strategic new platform segmentation. We’ll delve into the architectural magic allowing fewer cores to outperform their predecessors, explore why Intel is still playing catch-up to AMD’s Zen 5, and discuss the practical implications of Intel’s new “Mainstream” and “Expert” workstation tiers.
The new 18-core Xeon 654 impressively matches the multi-threaded performance of the older 28-core Xeon W7-3465X. Can you walk us through the specific architectural improvements in Granite Rapids that allow for such a significant leap in per-core efficiency and overall performance?
It’s a testament to just how much progress can be made in core design from one generation to the next. Seeing an 18-core chip go toe-to-toe with a previous-generation 28-core model is genuinely exciting. The magic here is rooted in the Redwood Cove P-Core architecture. Each of these new cores is simply a much more powerful and efficient engine. We’re seeing a substantial increase in Instructions Per Clock (IPC), meaning each core accomplishes more work in every cycle. This is complemented by a much healthier cache configuration, with 72 MB of L3 and 36 MB of L2 cache, which keeps data flowing to the cores without delay. When you combine that fundamental efficiency with a strong boost clock, touching up to 4.8 GHz, you get this incredible result where 18 modern cores can outmuscle 28 older ones. It’s a classic case of working smarter, not just harder.
Despite its gains, the article notes the Xeon 654’s Redwood Cove architecture falls behind AMD’s 16-core Zen 5 Threadripper. Could you elaborate on the key technical reasons for this performance gap, detailing which metrics—like IPC, cache design, or frequency—are the primary differentiators?
This is where the competitive landscape gets really interesting. While Granite Rapids is a massive step forward for Intel’s workstation line, it’s competing against AMD’s very latest and greatest, the Zen 5 architecture. The performance gap we’re seeing isn’t a failure on Intel’s part but rather a reflection of Zen 5’s sheer strength. The primary differentiator is almost certainly IPC. AMD has made huge strides with Zen 5, and it appears to hold a decisive architectural advantage in terms of raw per-core throughput. So even though Intel’s 18-core Xeon 654 might have more cores than the 16-core Threadripper 9955WX, each of those Zen 5 cores is doing more work. It’s not just a single-threaded issue, either; that advantage scales up, which is why the Threadripper pulls ahead in multi-threaded benchmarks as well. It shows that in the high-end space, having the most advanced core architecture is the ultimate trump card.
Intel plans to segment Granite Rapids-WS into “mainstream” and “Expert” tiers, with major differences in memory channels and PCIe lanes. Can you share an anecdote or a specific user case that illustrates why a professional would absolutely need the 8-channel memory of the Expert line over the mainstream option?
Absolutely. Think of a data scientist I worked with who was building complex machine learning models for genomic sequencing. Their datasets were colossal, often hundreds of gigabytes that needed to be loaded and processed. On a mainstream system with 4-channel memory, the CPU, despite having dozens of cores, would frequently sit idle, waiting for data to be shuttled from RAM. It was like trying to fill a swimming pool with a garden hose. Migrating them to an 8-channel memory platform was transformative. The memory bandwidth effectively doubled, turning that garden hose into a fire hose. The CPU utilization shot up, and model training times were slashed by nearly 40%. For professionals like that, or for visual effects artists rendering massive 8K scenes with complex textures, the 8-channel memory and 128 PCIe lanes of the “Expert” tier aren’t a luxury; they are a fundamental requirement to prevent bottlenecks and keep their expensive, core-heavy processor fed and productive.
The content points out a generational lag, with the workstation line using Redwood Cove cores while newer desktop architectures exist. What is the strategic reason for this decision, and what steps can we expect Intel to take to achieve architectural parity between its workstation and desktop CPUs?
This is a classic strategic decision driven by the unique demands of the workstation and server markets. These platforms demand uncompromising stability and reliability, which necessitates a much longer and more rigorous validation cycle than consumer desktop parts. Intel likely made the calculated decision to launch Granite Rapids-WS on the proven and stable Redwood Cove architecture rather than risk delays by integrating the newer Lion Cove cores, which were probably still being finalized. This allows them to bring a competitive platform to market sooner. Looking ahead, closing this gap will be Intel’s top priority. We can expect them to work on synchronizing their development pipelines so that future workstation families can launch with core architectures that are on par with their desktop counterparts, perhaps with the planned Cougar Cove or Coyote Cove architectures in 2026. Achieving that architectural parity is crucial for them to truly challenge for the performance crown.
What is your forecast for the high-end workstation CPU market over the next two years, given the fierce competition between Intel’s revitalized Xeon-WS family and AMD’s powerful Threadripper lineup?
The next two years are going to be fantastic for consumers and incredibly intense for the manufacturers. My forecast is for a market defined by aggressive competition on two fronts: raw core counts and architectural efficiency. Intel’s Granite Rapids-WS, particularly the higher-end models with up to 86 cores, signals a powerful return to form and will put immense pressure on AMD in workloads that can scale to that many threads. However, AMD currently holds the architectural high ground with Zen 5, delivering superior per-core performance that excels in a huge range of professional applications. I predict we’ll see a price war and a feature race, with both companies pushing the boundaries of what’s possible in memory support, I/O, and platform capabilities. The ultimate winner will be the company that can not only deliver the most cores but also pair them with the most advanced architecture, and that race is now closer than it has been in years.