Tag: Nvidia GB10

  • Asus Unveils Ascent GX10 Mini PC with Nvidia GB10 AI Superchip

    Asus Unveils Ascent GX10 Mini PC with Nvidia GB10 AI Superchip

    Key Takeaways

    1. Nvidia’s GB10 superchip features a 20-core ARM-based Grace CPU and a Blackwell GPU with 128 GB of unified memory, allowing both components to share resources.
    2. The GB10 chip offers high-performance capabilities, achieving around 1000 TOPS in AI performance, comparable to the Nvidia RTX 5070.
    3. The memory operates on a 256-bit bus with a speed of 273 GB/s, and it includes a dual-port ConnectX-7 network interface card for 200 GbE clustering.
    4. GB10 systems are designed for GPU-heavy local AI tasks, with general computing performance being less impressive compared to competitors like Apple’s M3 SoC.
    5. The pricing for systems like the DGX Spark is around $3,000, indicating that the upcoming Asus Ascent GX10 mini PC may have a similar price point.

    Nvidia’s GB10 superchip is a real powerhouse of a System on a Chip (SoC). It features an ARM-based Grace CPU with 20 cores, combined with a Blackwell GPU and 128 GB of unified memory that both components can use. Asus is set to unveil its Ascent GX10 mini PC, which is equipped with the GB10 superchip, on the 22nd of this month, as noticed by VideoCardz.

    Performance Across Vendors

    Various manufacturers have launched systems powered by the Nvidia GB10, and the performance differences among them are expected to be slight. The Nvidia DIGITS, also known as DGX Spark, includes the GB10 chip and reveals some fascinating information about it. The 128 GB of unified memory operates on a 256-bit bus, which allows for a memory speed of 273 GB/s. Additionally, the dual-port ConnectX-7 network interface card is designed for 200 GbE clustering, making it essential for many high-end users.

    AI Performance Metrics

    Nvidia claims that the GB10 can achieve around 1000 TOPS, or 1 petaflop of AI performance (FP4). This means it should perform similarly to systems equipped with the Nvidia RTX 5070, which has a peak performance of about 988 TOPS. However, the Nvidia RTX 5080, currently priced at $1,699 on Amazon, surpasses it with over 1400 TOPS. It’s worth noting that the RTX 5070 has much quicker memory, clocking in at 673 GB/s.

    Targeted Use Cases

    It’s important to remember that GB10-powered systems are specifically aimed at GPU-heavy local AI tasks, and their performance in general computing is lacking. Recent leaks of benchmarks show that the 20-core ARM CPU in the GB10 is roughly on par with Apple’s 8-core M3 SoC, which is acceptable but not particularly impressive.

    As for pricing, the DGX Spark is listed at $3,000, so it is likely that the Ascent GX10 from Asus will be in a similar price range. In comparison, Apple’s more expensive Mac Studio, which features the M3 Ultra SoC with an impressive 512 GB of unified memory at 800 GB/s, significantly outperforms the GB10 in terms of memory speed but fails to match it in raw AI capabilities.

    Source:
    Link

    Tags: , ,
  • Lenovo ThinkStation PGX Mini-PC Revealed with Nvidia GB10 Chip

    Lenovo ThinkStation PGX Mini-PC Revealed with Nvidia GB10 Chip

    Key Takeaways

    1. The Nvidia GB10 ‘superchip’ was recently spotted on Geekbench, generating excitement in the tech community.
    2. Lenovo’s ThinkStation PGX mini-PC features 128 GB of LPDDR5x RAM and runs on Nvidia’s DGX OS, but storage details remain unclear.
    3. The ThinkStation PGX will compete with Apple’s M3 Ultra and AMD’s Ryzen AI Max+ 395, with Apple currently leading in benchmark scores.
    4. Nvidia’s GB10 chip features a combination of 10 Cortex-X925 and 10 Cortex-A725 cores, positioning it between competitors.
    5. Anticipation is growing for more information on the ThinkStation PGX, especially with upcoming events like Computex.

    A few days back, a surprising Nvidia chip popped up on Geekbench, seemingly out of nowhere. It was the GB10 ‘superchip’ that was announced some time ago, and even though the listings have been removed now, news has already spread. Thanks to Lenovo, we now get our first glimpse at one of the initial GB10-powered mini-PCs: the Lenovo ThinkStation PGX.

    Details on the ThinkStation PGX

    Sadly, Lenovo hasn’t shared much about the ThinkStation PGX mini-PC, with the press release mentioning it comes equipped with Nvidia’s DGX OS and software suite. It boasts 128 GB of LPDDR5x RAM on a 256-bit bus, but the storage capacity is not specified. A quick look at Nvidia’s GB10 spec sheet shows that it can support up to 4 TB. Those eager to purchase a Lenovo ThinkStation PGX will have to be patient until Q3 of this year. Although the price hasn’t been disclosed, it’s likely to be on the expensive side.

    Competition in the Market

    The Nvidia GB10 powered ThinkStation PGX is set to compete mainly against Apple’s M3 Ultra and AMD’s Ryzen AI Max+ 395. From the Geekbench scores alone, Apple leads the field in both single and multi-core benchmarks. The GB10 combines 10x Cortex-X925 cores and 10x Cortex-A725 cores, placing it somewhere between the M3 Ultra and the Strix Halo parts. However, Nvidia’s real strength lies in its software ecosystem, which will surely be tailored for the GB10. Still, we expect to learn more during Computex, where it will hopefully be showcased in greater detail.

    Anticipation for More Information

    Lenovo’s ThinkStation PGX has generated quite a buzz, but the lack of details leaves potential buyers curious. As more information trickles in, especially with upcoming events like Computex, the anticipation is building for what this mini-PC can really do. With its unique specs and powerful Nvidia backing, it could change the game in the mini-PC market.

    Source:
    Link

    Tags: , ,
  • Nvidia GB10 Superchip Shows Promise in Early Benchmark Tests

    Nvidia GB10 Superchip Shows Promise in Early Benchmark Tests

    Key Takeaways

    1. Asus and Dell have launched devices using Nvidia’s GB10 Grace Blackwell ARM chipset, following Nvidia’s Project DIGITS AI supercomputer announcement.
    2. The GB10 chipset has shown varying benchmark results on Geekbench, labeled as ARMv8 despite using ARMv9 cores.
    3. Performance metrics for the GB10 are inconsistent; it competes with top ARM and x86 processors but should be viewed cautiously.
    4. The GB10’s Cortex-X925 cores can reach clock speeds of 3.9 GHz, but details on Cortex-X725 performance are lacking.
    5. Nvidia may introduce N1 and N1X alternatives for desktops and laptops at Computex 2025 later this month.

    Almost two months have gone since Asus and Dell introduced the first devices that use Nvidia’s GB10 Grace Blackwell ARM chipset. Prior to this, Nvidia showcased the Project DIGITS AI supercomputer at CES 2025, which it claimed would launch this month at a price of about $2,999.

    Benchmark Findings

    The so-called ‘superchip’ that supports these three devices has recently shown up on Geekbench with four different benchmark results. It’s important to mention that there are some peculiarities in these results. For instance, Geekbench labels the GB10 as an ARMv8 chipset, even though its ARM Cortex-X725 and Cortex-X925 CPU cores utilize the newer ARMv9 instruction set.

    Performance Insights

    Moreover, performance metrics are quite inconsistent at this point. To put it simply, we shouldn’t focus too much on these numbers, even though single-core performance competes with top-tier ARM and x86 processors. For context, Apple’s latest M4 generation chipsets are the leaders in this area, especially the M4 Max found in devices like the MacBook Pro 16, currently priced at $3,517 on Amazon.

    Geekbench notes that the GB10’s Cortex-X925 cores can clock in at 3.9 GHz. Sadly, there are no details regarding the performance of the Cortex-X725 cores. Even though the GB10 is designed for workstation use, there are hints that Nvidia might be heading to Computex 2025 later this month with N1 and N1X alternatives targeting desktops and laptops.

    Geekbench via @9550pro, Hardware Luxx – Image credit.

    Source:
    Link

    Nvidia GB10 Superchip Shows Promise in Early Benchmark TestsNvidia GB10 Superchip Shows Promise in Early Benchmark TestsNvidia GB10 Superchip Shows Promise in Early Benchmark TestsNvidia GB10 Superchip Shows Promise in Early Benchmark Tests
    Tags: , ,