Should Nvidia be worried? Plucky inference rival replaces 320 Nvidia GPUs with 16 reconfigurable dataflow units
When you buy through links on our articles, Future and its syndication partners may earn a commission.
SambaNova runs DeepSeek-R1 at 198 tokens/sec using 16 custom chips
The SN40L RDU chip is reportedly 3X faster, 5X more efficient than GPUs
5X speed boost is promised soon, with 100X capacity by year-end on cloud
Chinese AI upstart DeepSeek has very quickly made a name for itself in 2025, with its R1 large-scale open source language model, built for advanced reasoning tasks, showing performance on par with the industry's top models, while being more cost-efficient.
SambaNova Systems, an AI startup founded in 2017 by experts from Sun/Oracle and Stanford University, has now announced what it claims is the world's fastest deployment of the DeepSeek-R1 671B LLM to date.
The company says it has achieved 198 tokens per second, per user, using just 16 custom-built chips, replacing the 40 racks of 320 Nvidia GPUs that would typically be required.
'Powered by the SN40L RDU chip, SambaNova is the fastest platform running DeepSeek,' said Rodrigo Liang, CEO and co-founder of SambaNova. 'This will increase to 5X faster than the latest GPU speed on a single rack - and by year-end, we will offer 100X capacity for DeepSeek-R1.'
While Nvidia's GPUs have traditionally powered large AI workloads, SambaNova argues that its reconfigurable dataflow architecture offers a more efficient solution. The company claims its hardware delivers three times the speed and five times the efficiency of leading GPUs while maintaining the full reasoning power of DeepSeek-R1.
'DeepSeek-R1 is one of the most advanced frontier AI models available, but its full potential has been limited by the inefficiency of GPUs,' said Liang. 'That changes today. We're bringing the next major breakthrough - collapsing inference costs and reducing hardware requirements from 40 racks to just one - to offer DeepSeek-R1 at the fastest speeds, efficiently.'
George Cameron, co-founder of AI evaluating firm Artificial Analysis, said his company had 'independently benchmarked SambaNova's cloud deployment of the full 671 billion parameter DeepSeek-R1 Mixture of Experts model at over 195 output tokens/s, the fastest output speed we have ever measured for DeepSeek-R1. High output speeds are particularly important for reasoning models, as these models use reasoning output tokens to improve the quality of their responses. SambaNova's high output speeds will support the use of reasoning models in latency-sensitive use cases.'
DeepSeek-R1 671B is now available on SambaNova Cloud, with API access offered to select users. The company is scaling capacity rapidly, and says it hopes to reach 20,000 tokens per second of total rack throughput "in the near future".
Nvidia and AMD trade blows over who is faster on DeepSeek AI benchmarks
A look at the Nvidia GPU that powers DeepSeek's AI global ambition
AI phenomenon DeepSeek is officially growing faster than ChatGPT
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Yahoo
30 minutes ago
- Yahoo
After 4 Decades On Sunset Blvd., A Beloved Chinese Restaurant Is Leaving West Hollywood
Sunset Boulevard, a long stretch of road that runs through West Hollywood casually called the Sunset Strip, is home to some of the most iconic restaurants in Los Angeles. However, it will no longer be the home of a cherished restaurant staple — Chin Chin. The restaurant that has called West Hollywood home for 45 years, will be closing this summer, with its last night of service on July 27. In a statement made on the restaurant's Facebook page, it notes that the reasoning for the closing is unexpected, but that its other locations in Brentwood, Studio City, and Las Vegas will remain open. "Over the decades, we've celebrated countless milestones, shared unforgettable meals, and built lasting connections with our cherished guests," the statement said. "None of this would have been possible without your unwavering support, and for that, we are deeply grateful." The statement also notes that the restaurant has launched a GoFundMe page to help them look for a new home in the area, and to support the staff during the transition. Understandably, fans of Chin Chin had a lot of memories to share on social media about how much the restaurant has meant to them. One user noted that his first experience eating there was in 1987, and since then he has always ordered the famed Chinese Chicken Salad and Chicken Fried Rice for parties. Others expressed disappointment and disbelief with the sudden closing, asking fans to visit the other locations to show support. Read more: 13 Chinese Restaurant Chains, Ranked Worst To Best While Chin Chin is the latest business to have to shut its doors on Sunset Boulevard, it's not the only one. According to CBS News, the strip has gone through a tough time recently, with many of its most well-known businesses closing. Le Petit Four, a famous French bakery that had called West Hollywood home for 44 years, had to close in Marchm and is still looking for investors and a new location to call home. According to KTLA, Rock and Reilly's Irish pub, which had been open for nearly 15 years on the Sunset Strip, shut down unexpectedly just before St. Patrick's Day. Sunset Strip Liquor Store, which opened in 2020 and replaced Sun Bee Liquor Market, also announced its closure at the end of March. So why has Sunset Boulevard gone through such troubled times as of late? An Instagram post from Le Petits Four in March cited a litany of reasons, which while specific to this establishment, may also explain the rest of the shutdowns in the area. "Owner Alexandre Morgenthaler, who has lovingly run Le Petit Four since 1999, did everything possible to keep our doors open," the statement reads. "But with rising costs — including a 30% minimum wage increase since COVID and soaring rent — along with a decline in foot traffic, the decision became unavoidable." Read the original article on Tasting Table.
Yahoo
30 minutes ago
- Yahoo
China's Chip Dreams Just Hit a Wall--But a New Tech Power Could Rise From the Wreckage
Xiaomi (XIACY) just unveiled a big leap in its chip journeya 3nm self-designed processor called XRING O1, built by TSMC in Taiwan. But it might be the last of its kind for a while. A new directive from the US government now restricts the sale of advanced electronic design automation (EDA) software to Chinese companies, hitting the very tools that helped make that chip possible. Sources familiar with the matter say Xiaomi, along with companies like Lenovo and Bitmain, could be first in line to feel the impact. These firms have been quietly investing years into developing their own silicon, while relying on EDA software from US players like Synopsys and Cadence. Now, with access to future updates and tech support potentially cut off, that roadmap just got a lot bumpier. Warning! GuruFocus has detected 3 Warning Signs with XIACY. The restrictions don't revoke existing licenses, but they do freeze the futureno upgrades, no fixes, no help. That's a problem, because keeping chips manufacturing-ready at TSMC demands constant fine-tuning with the latest software patches. So far, chips for smartphones and tablets appear exempt from the AI-level restrictions, but the writing is on the wall. The move signals a wider push to choke off China's long-term access to bleeding-edge chip development, not just hardware. While giants like Alibaba and Baidu also design their own processors, the exact fallout for them remains murky. But one thing is increasingly clear: China's chip strategy can't depend on foreign tools forever. That might be where the real story begins. Empyrean TechnologyChina's leading homegrown EDA providerhas been quietly building an alternative software stack. It's not yet on par with US offerings, but insiders say it's already usable for 7nm chips and above. Empyrean, along with firms like Primarius and Semitronix, saw their shares surge after news of the US restrictions broke. There's also a darker undercurrent: analysts point out that hacked versions of US EDA tools are already being used inside China. As one expert noted, It's very easy to reverse-engineer what you need. That reality helps explain why Synopsys and Cadence have seen lagging China revenues despite growing chip demand. Whether the US ban slows China downor pushes it faster toward self-reliancemay depend on how fast its EDA ecosystem matures. Either way, this fight just shifted from factories to source code. This article first appeared on GuruFocus. Sign in to access your portfolio
Yahoo
37 minutes ago
- Yahoo
Nvidia Stock Jumps as Jefferies Sees Margins 'Could Skyrocket to 80%'
June 3 - Shares of Nvidia (NASDAQ:NVDA) climbed about 3% Tuesday morning after Jefferies named the chipmaker a top stock pick and projected profit margins could hit as high as 80% this year. Warning! GuruFocus has detected 4 Warning Signs with NVDA. Jefferies analyst Blayne Curtis added Nvidia to the firm's highest conviction list, citing the ramp-up of next-generation Blackwell chips. In his view, mounting demand for Blackwell is poised to drive a significant margin expansion. Currently, Nvidia's gross margins sit near 61%, but Jefferies predicts they could leap toward 80% as Blackwell volumes accelerate. Margins at that level are rare in hardware, underscoring Nvidia's robust pricing power and market position. The firm says Nvidia has evolved beyond selling standalone chips, emerging as a comprehensive AI infrastructure provider. Customers ranging from hyperscale datacenters to hedge funds are lining up for its combined hardware, software, and systems offerings. Continuous software licensing and large-scale deployments are expected to bolster profitability as AI workloads proliferate. This article first appeared on GuruFocus. Error in retrieving data Sign in to access your portfolio Error in retrieving data Error in retrieving data Error in retrieving data Error in retrieving data
