Intel has published a set of benchmarks for its new D270 GPU, a 350W accelerator based on the Xe3P architecture, and the company is making bold claims. In its own comparisons against NVIDIA's 600W RTX PRO 6000 Blackwell, Intel says the D270 delivers up to 5.9x higher throughput in FP64 workloads. The gains are much smaller in AI inference tests.
Intel Puts the D270 Against NVIDIA Blackwell
The benchmarks pit Intel's Xe3P-based D270 against one of NVIDIA's most powerful professional graphics cards. The RTX PRO 6000 Blackwell is aimed at workstation, simulation, and AI users who need large memory capacity and strong compute performance in a single card.
Intel's choice of comparison is deliberate. By targeting NVIDIA's flagship professional Blackwell product, the company is signaling that it wants to compete in the high-end accelerator market rather than only in entry-level or midrange segments.
The Headline Number: Up to 5.9x in FP64
The most eye-catching result is the 5.9x advantage in FP64, or double-precision, testing. Double-precision math is central to scientific computing, engineering simulation, computational fluid dynamics, and other workloads where numerical accuracy matters more than raw AI throughput.
The wording "up to" is important. It suggests the 5.9x figure is the best case among the tests Intel selected, not an average across all workloads. Readers should treat it as a peak result.
The size of the gap is also unsurprising. NVIDIA's professional Blackwell cards are widely understood to prioritize graphics, AI, and lower-precision compute over FP64. A GPU designed with stronger double-precision hardware can therefore show a very large lead in that one area.
Why Power Efficiency Matters in This Comparison
Intel is also emphasizing the power difference between the two cards. The D270 is rated at 350W, while the RTX PRO 6000 Blackwell is rated at 600W. Delivering a large performance lead at far lower power would translate into better performance per watt.
For data centers and research labs, that matters as much as raw speed. Power delivery, cooling, and rack density all limit how many accelerators can be deployed, and a 250W difference per card adds up quickly across a cluster.
Even so, power ratings are not the same as measured power draw during a given workload. Real-world efficiency depends on how each card behaves under sustained load, and Intel's numbers will need independent testing to confirm.
AI Inference Gains Are Noticeably More Modest
The picture changes when the focus shifts to AI inference. According to the benchmark results, the D270's advantage is much smaller there than in FP64 tests. This suggests the card's strengths lie in traditional high-performance computing more than in machine learning deployment.
That is a meaningful caveat given how much of the current market is driven by AI. NVIDIA has built a large ecosystem around inference and training, and its Blackwell architecture is heavily optimized for those workloads.
Intel's smaller lead in this category shows that the D270 is not a blanket replacement for NVIDIA's hardware. Its appeal will depend on what customers actually run.
What the Xe3P Architecture Brings to the Table
The D270 is built on Xe3P, Intel's next step in its GPU architecture roadmap. The benchmarks suggest the company has focused on compute capability, particularly in areas where competitors have de-emphasized double-precision performance.
If that focus holds up in independent testing, Xe3P could give Intel a distinctive position. Rather than chasing NVIDIA on every metric, it could carve out a niche in scientific and technical computing.
Who Could Benefit From the D270
The most obvious audience is researchers and engineers who rely on FP64. This includes fields like weather modeling, physics simulation, molecular dynamics, and structural analysis, where double-precision math is still the standard.
Organizations running mixed workloads may find the picture less clear-cut. If AI inference is the main task, the smaller performance gap means other factors, such as Software support, memory capacity, and price, will carry more weight.
Software Support and Ecosystem Challenges
Hardware performance is only part of the story. NVIDIA's CUDA ecosystem remains deeply entrenched, with years of optimized libraries, tools, and developer familiarity. Any challenger must convince customers that migrating or supporting another platform is worth the effort.
Intel has been investing in its software stack, including oneAPI and related tools, but adoption takes time. Buyers will want to see that the applications they depend on run well on the D270 without extensive rework.
Why Vendor Benchmarks Deserve Caution
As with any manufacturer-published results, Intel's numbers should be read carefully. Test configurations, software versions, and workload selection can all influence outcomes, and companies typically highlight scenarios in which their products perform best.
Independent reviews and third-party benchmarks will be needed to confirm whether the 5.9x claim holds in practice and how the D270 behaves across a broader range of applications. Until then, the results are best seen as a promising but unverified indicator.
What This Means for the Competitive GPU Market
If the D270 performs as Intel says, it could add some competitive pressure in the professional and HPC accelerator space. NVIDIA still dominates overall, but a credible alternative with strong FP64 performance and lower power draw would give buyers more options.
For now, the announcement shows that Intel is positioning its GPU efforts around specific strengths rather than trying to match NVIDIA across the board. Whether that strategy succeeds will depend on independent validation, software maturity, pricing, and availability in the months ahead.
If you have any questions, please don't hesitate to Contact Us
โ Back to Technology News