Skip to main content
Intel Says D270 GPU Beats Nvidia at FP64: Nearly 6x on Monte Carlo, Under 2x on American Options - Tech Times

Intel Says D270 GPU Beats Nvidia at FP64: Nearly 6x on Monte Carlo, Under 2x on American Options - Tech Times

NVIDIA GPUs
17 lis 2026, 08:36Hardware News
Oryginał
4 min read

Intel has released detailed benchmarks for its upcoming D270 GPU, showing up to a 5.9x performance advantage over Nvidia's RTX PRO 6000 Blackwell Server Edition on European Monte Carlo pricing simulations.

Intel has released the first per-workload breakdown of its claims regarding the D270 GPU, a data center card designed for high-performance computing. The tests, conducted on pre-production silicon, compare the D270 against Nvidia's RTX PRO 6000 Blackwell Server Edition in four financial workloads. The results show a range of performance gains, from a 1.92x advantage on American Black-Scholes pricing to a 5.9x lead on European Monte Carlo simulations.

The comparison comes as the Supercomputing 2026 conference gets underway in Chicago. Intel has not yet announced a price or a release date for the D270, and customer sampling began in October. The benchmarks are significant because they directly address the needs of quantitative finance teams that require high-precision calculations for risk management and derivatives pricing.

Hardware Architecture Differences

The performance gap stems from a fundamental architectural difference between the two vendors. Nvidia's RTX PRO 6000 shares its GB202 silicon with the GeForce RTX 5090 gaming card. On this chip, Nvidia deliberately limits double-precision (FP64) performance to a 1:64 ratio relative to its single-precision (FP32) rate. The workstation edition lists 120 TFLOPS of FP32 but no official FP64 figure, while the storage review lists 1.968 TFLOPS of FP64.

In contrast, Intel's Xe cores on the D270 run FP64 calculations at full rate. Nand Research estimates the peak FP64 vector rate at 10.2 TFLOPS, which is roughly five times higher than the Nvidia card's native FP64 rate. This architectural choice makes the Nvidia card an easy baseline for double-precision workloads, but it also means the card is not optimized for the type of calculations Intel is highlighting.

Intel has not published a clock speed or a total TFLOPS number for the D270. The company lists the card's power consumption at 350 W, compared to up to 600 W for the Nvidia card, but it does not report performance per watt. The discrepancy between the raw peak ratio of about 5.2x and Intel's best result of 5.9x suggests that factors beyond raw peak performance, such as memory bandwidth and software optimization, play a role.

Nvidia's FP64 Emulation Approach

Nvidia is not without its own strategy for handling double-precision workloads on its consumer and workstation silicon. The company has introduced FP64 emulation via its cuBLAS library, which uses a technique called the Ozaki scheme. This method rebuilds double-precision results from low-precision integer arithmetic on Tensor Cores. Nvidia reports speedups of up to about 13x over native FP64 for matrix multiplication on this card.

However, the scope of this emulation is limited. Nvidia describes the feature as a matrix-multiplication capability, with other routines listed as future work. AMD fellow Nicholas Malaya told The Register that emulation offers little or no benefit for roughly 60 to 70 percent of HPC workloads, which rely on vector fused multiply-add operations. Intel's tested financial workloads, such as Monte Carlo pricing, are described as vector-heavy rather than matrix-heavy.

The software stack used in Intel's tests also differs from Nvidia's emulation setup. Intel's footnote lists a CUDA 12.8.0 container with oneMath and DPC++ for the Nvidia node. Nvidia's emulation feature is part of the CUDA Toolkit 13.0 Update 2, which is a newer release. This means the Nvidia result Intel measured against may not have included the full benefits of Nvidia's latest emulation technology.

Context for Finance Teams

For quantitative finance teams, the decision between these two approaches involves more than just raw speed. Oxford mathematician Mike Giles argues that single precision is often adequate for Monte Carlo finance work except when sensitivities, known as Greeks, are computed by bumping inputs. He recommends a double-precision accumulator or binary tree summation when averaging many paths.

A NAG technical report found that a GPU Monte Carlo option pricer could be stable and accurate in single precision, with roughly twice the speed. Nvidia's datasheet lists 120 TFLOPS of FP32 for the Server Edition. A team that accepts single-precision Monte Carlo is therefore not forced onto the card's 1:64 double-precision path. Intel's table does not test FP32 finance workloads, so the comparison is limited to teams that require true double precision.

The D270's case is strongest where model-risk rules or Greeks require true double precision. Until independent audits, a price and a ship date arrive, the 5.9x figure is a vendor-run, pre-production result. Buyers should test matrix share, precision needs, and per-card results before trusting the claims. Intel plans live demonstrations at SC26, but these will come without numbers.

Z.ai
Article written by AI
Z.ai GLM-4.7-FlashFree
Free AI articles150 of 150
No limit after signing in with Telegram