News
- OpenAI published the first Jalapeño benchmarks at Hot Chips, claiming 1.5 to 1.9x more throughput per kilowatt and up to 3.6x lower latency than Nvidia's GB200 and GB300 rack systems
- The chip's purported gains are at a reported 700W power draw and center around efficiency, with Nvidia's GB300 still leading on absolute throughput per package by 20-25 percent.
- The numbers are currently self-reported, and volume production of the chip is not expected to ramp up until 2027
OpenAI took the stage at Hot Chips on August 25 with the first published performance figures for Jalapeño, the inference accelerator it co-developed with Broadcom, and the numbers are designed to be read the way it flatters the former: efficiency.
The company reported 1.5 to 1.9 times more throughput per kilowatt and 1.7 to 3.6 times lower end-to-end latency across three open models than the Nvidia rack systems it tested against.
OpenAI's Jalapeño is currently rated at 700W, versus comparable Nvidia silicon that was rated for 1200W and 1400W, a key differentiator in a benchmark that already has limited information available to researchers looking to pick a clear winner.
AI efficiency takes center stage?OpenAI's Jalapeño is an ASIC, or an Application-specific integrated circuit, which essentially means that it focuses primarily on and is great for very specific AI workloads; in this case, OpenAI's inference needs, which allow it to run multiple models with significant efficiency gains in tow.
Richard Ho, who runs OpenAI's hardware program, told reporters on a press call that the results show "a very, very significant performance advance over state-of-the-art."
OpenAI ran SemiAnalysis's public InferenceX suite on GPT-OSS 120B, DeepSeek R1 670B, and Moonshot AI's trillion-parameter Kimi K2.5, at a nominal 8,000-token input and 1,000-token output. It scored a 1.9x efficiency win vs. GB200 on its own GPT-OSS model, and 1.7x and 1.5x on DeepSeek and Moonshot's offerings against a GB300 system. Interestingly, the figures for GPT-OSS 120B running on a GB300 compared to OpenAI's Jalapeño are not published.
It is important to point out here that there is a certain degree of cherry-picking to flatter OpenAI's own results: by choosing to compare ratios on a per-kilowatt basis versus a per-chip basis, while efficiency remains consistent, it does paint Jalapeño as potentially a much more potent competitor to Nvidia's last-generation offerings than it actually is.
Nvidia's GB300 might not have an efficiency win versus Jalapeño, but it remains a much more powerful chip in all tests, at a time when Nvidia is already rolling out Vera Rubin. Curiously, OpenAI has not published any information about how its chip stacks up against an HBM4-touting Vera Rubin, which Nvidia has presented as an efficiency juggernaut.
It is important to point out that Jalapeño's results also use single-token prediction throughout, with no speculative decoding and no prefill-decode disaggregation. Nvidia's production deployments commonly use multi-token prediction, and when OpenAI pits Jalapeño against a GB300 configured that way, the peak efficiency lead drops to roughly 1.5x.
OpenAI says it built the chip to keep model state local, writing that it "designed Jalapeño to minimize data movement and communication delays." Cores and HBM are divided into slices, each core slice holding a low-latency view of its own memory, with synchronization pushed onto a dedicated collective network. Each package pairs a compute die with six HBM4 stacks for 216 GiB at 15.4 TB/s, which is less capacity than the GB300's 288GB of HBM3E but considerably more bandwidth, and roughly 50 percent more memory per watt of rated power.
Every number published currently comes from A0 silicon, also known as the initial, first-run version of the chip. A more efficient B0 stepping is already in the fab with roughly 25 percent better performance per watt, and production is scheduled to ramp gradually across 2027.
Although it may not hold a candle to Nvidia's Vera Rubin, it might not have to; Alexander Harrowell of Omdia told CNBC that "this is the biggest competitive threat to NVIDIA," noting that roughly half of AI infrastructure capital spending comes from firms that already run custom chip programs or could. Yole Group's Adrien Sanchez framed it as pressure on Nvidia's inference margins specifically, the fastest-growing part of the business.
The timing of this announcement is both awkward and potentially suspect at a time when Nvidia just agreed to backstop $105 billion in financing for OpenAI's upcoming Ohio facility, even as Richard Ho told Bloomberg that the company continues to need a lot of Nvidia's chips, with the latter being "a really good partner." This is hardly surprising, given Jalapeño's limitations as an ASIC useful mainly for inference, while Nvidia's chips are key to training new AI models.
From Nvidia's perspective, this can be seen as a blow to its plans for a high-margin inference industry where one of the two major buyers of inference hardware is not only looking for a homegrown alternative, but is already well on its way to getting the job done.
- HUR identified 140 components linked to the upgraded Iskander missile system
- American manufacturers supplied 63 of the missile's identified electronic components
- Only one-quarter of all components came from Russian domestic manufacturers
Ukraine’s Defense Intelligence (HUR) has released details about foreign-made parts it says were found inside Russia’s upgraded Iskander-1000 ballistic missile.
The disclosure said American companies, including Xilinx, part of AMD, Atmel, part of Micron Technology, DM&P Electronics, Winbond Electronics, and Analog Devices, supplied critical parts for the missile's guidance and processing systems.
The agency also identified 140 components linked to the missile and said only a limited share came from Russian manufacturers.
Foreign chips remain inside Russia’s upgraded missile systemThe disclosure points to continued foreign technology use in a weapon system developed by Moscow despite efforts to reduce outside dependence.
According to Ukrainian intelligence, 35 companies are connected to the production network supporting Russia’s Iskander-M missile family.
The agency said the updated 9M723-2 variant, also called Iskander-1000, has appeared during recent attacks and forms part of the Bora-M missile system.
The newer missile version reportedly uses a larger engine and modified launcher design, allowing it to reach distances of up to 550km.
This compares with the 390km range attributed by Ukrainian intelligence to the earlier 9M723-1 missile model.
HUR stated that Russia continues to use imported electronic parts in key missile areas despite attempts to replace foreign technology with domestic alternatives.
The agency said only about one-quarter of the identified components were produced by Russian firms.
The intelligence report lists 63 components linked to manufacturers from the United States, while Switzerland and Belarus each account for eight components.
Germany and Taiwan were each linked to four parts, while China supplied three components according to the information released by HUR.
“The most critical components—microprocessors, FPGAs, memory, digital-to-analog and analog-to-digital converters—come from American and Taiwanese manufacturers…” said Ukraine’s Defense Intelligence.
The missile’s guidance system was described as one of the areas with significant foreign dependence.
HUR claims only five of 43 parts inside a flight information processing unit connected to the missile’s laser gyroscope were Russian-made.
Sanctions face challenges as supply chains remain complexThe identified components include field programmable gate arrays, memory chips, microprocessors and conversion devices used for handling digital and analogue signals.
Such parts are common in many electronic systems, making their movement across global markets difficult to monitor completely.
Ukraine’s intelligence service has called for stronger restrictions against companies, organizations and facilities it believes are linked to Russia’s missile production network.
The agency said international cooperation would be necessary to limit access to important technologies.
However, controlling the movement of advanced electronic components remains difficult because many parts have civilian uses and can pass through several countries before reaching final users.
The presence of international suppliers does not alone establish whether companies directly supply military programs.
The report comes as Ukraine continues efforts to disrupt Russia’s missile operations through sanctions pressure and attacks against launch equipment.
Ukrainian officials have also reported strikes against Russian missile launch platforms before they could be used.
“To counter the aggressor’s plans, it is important to consolidate efforts aimed at restricting Russia’s program for the development, production, and modernization of ballistic weapons,” HUR said.
Via United24media