TodayToday's edition
NVIDIA fixed the DGX Spark's display-triggered reboot loop on kernel 7.0.0-1024 on Nov. 3, but said on Nov. 5 it is still investigating reported freezes and GPU lockups; the display bug also hit ASUS GX10 owners.
Local AI DeskEdited by Ege Erten
4 min readAI-assisted
Follows our story: NVIDIA DGX Spark Review: 128GB GB10 Box Now Lists at $6,950
NVIDIA's product-page photo of the DGX Spark family: the gold Founders Edition box and GB10 partner systems. It is not a photo of any unit affected by the kernel reports.
Image: NVIDIA
Source · Press image from the NVIDIA newsroom, used for news reporting.
NVIDIA has not yet fixed the freezes, GPU lockups and watchdog resets that DGX Spark owners report on Linux kernel 7.0.0-1024-nvidia. Its last statement on the broader crashes, on Nov. 5, said it is still investigating, while a Nov. 3 package fixed only the display-triggered reboot loop.
The newest report, posted on Nov. 6 on NVIDIA's developer forum, extends the trouble to kernel 7.0.0-1025-nvidia on a machine that had package nvidia-spark-initcall-bl 1.0-1, not the 1.0-2 fix. One owner described three episodes of kernel memory corruption and freezes that began roughly 3.5 to 11 minutes after boot, with none seen after booting back to 7.0.0-1019 during a short observation period the owner called preliminary. That machine ran Ubuntu 26.04.1 with NVIDIA's platform repositories disabled and a DGX 7.6.0 release marker retained, not the stock DGX OS; an NVIDIA employee treated it as an Ubuntu 26.04 system. On Nov. 7 and Nov. 8 an NVIDIA employee replied that Ubuntu 26.04 systems do not receive the Nov. 3 fix automatically and need the power_wrap_init boot option added to GRUB by hand, persistently. The owner called the cause unproven. Our DGX Spark review covers the box's speed and price; this is the software side.
NVIDIA's Nov 3 package fixed the display-triggered reboot loop; other crashes on kernels 1024 and 1025 were still open as of NVIDIA's Nov 5 statement.
Illustration (AI-generated)
The reports began on Oct. 30, when owners said their units rebooted in a loop after an update moved them from kernel 7.0.0-1019 to 7.0.0-1024. Owners said units stayed up with no monitor connected; one said unplugging his USB-C monitor only delayed the reboots to about five minutes. One Founders Edition owner reported a reboot about every 30 seconds with an HDMI monitor and none when headless. An NVIDIA moderator said the same day that the company would look into the kernel issue. On Nov. 2 an NVIDIA employee asked testers to add power_wrap_init to the initcall_blacklist boot argument, next to the tegra234_cbb_init entry already there, and testers in the thread reported that it held.
On Nov. 3 an NVIDIA moderator said the fix for the reboot loop ships as the package nvidia-spark-initcall-bl version 1.0-2, which applies that same boot option. NVIDIA told owners to install it with apt dist-upgrade rather than apt upgrade, and on Nov. 4 said anyone updating through the DGX Dashboard receives it.
On Nov. 5 the moderator drew the line. The published workaround targets the display, HDMI and USB-C reset path only. NVIDIA is still working out whether the other reports share that root cause or form a separate regression, and it called booting a previously installed stable kernel a reasonable stopgap for systems too unstable to use.
The Nov. 3 fix is a kernel boot option, not a new kernel. The nvidia-spark-initcall-bl 1.0-2 package adds initcall_blacklist=power_wrap_init, and the Linux kernel documentation says that parameter tells the kernel not to run the listed start-up functions. NVIDIA said on Nov. 5 the option targets only the display, HDMI and USB-C reset path.
Systems that take NVIDIA's package get it automatically, an NVIDIA moderator said. An Ubuntu 26.04 install does not, and an NVIDIA employee said on Nov. 7 and Nov. 8 the option must be set persistently in GRUB. For a box meant to run unattended, that decides whether a kernel update is safe to take.
A user-compiled summary thread, opened on Nov. 3, lists idle reboots, reboot loops, hard resets by the hardware watchdog, GPU lockups logged as Xid 13, Xid 154 and Xid 109, kernel Oops messages and full freezes needing a power cycle. Several posters said rolling back to 7.0.0-1019 made their systems stable again. These are owners' reports. NVIDIA has acknowledged them but has not confirmed them as one bug or given a count of affected units.
The reports are not limited to NVIDIA's own box. An owner of two ASUS Ascent GX10 units reported hard resets on 7.0.0-1024 on both, and later traced the trigger to a 4K TV over HDMI, with a 1080p monitor working on either port. Other posters described the reboot loop on a Gigabyte unit and on an MSI EdgeXpert. The ASUS Ascent GX10 review uses the same GB10 chip and DGX OS stack. In its Nov. 5 post NVIDIA addressed its update instruction to Founders Edition systems and asked anyone still crashing to say whether they run a Founders Edition or an OEM GB10 variant.
NVIDIA's own documents lag the forum. The DGX Spark release notes, last updated Nov. 4, list Canonical kernel 7.0.0-1019 as the current version, while owners received 1024 through updates and the Nov. 6 report concerns 1025. The notes say the table covers the Founders Edition only and that partner GB10 systems may not receive updates at the same time. The known-issues page, updated the same day, carries no entry on kernel 1024 or 1025.
For a machine bought to run large models such as gpt-oss-120b unattended (NVIDIA rates the box for up to 200B parameters), a kernel that can hang a few minutes after boot matters more than any benchmark. NVIDIA has not said whether a new kernel is coming, whether OEM GB10 systems from ASUS, Dell, HP, Gigabyte, MSI, Acer and Lenovo receive the 1.0-2 package on the same schedule, or when the release notes will list the problem. It has promised to update the Nov. 3 summary thread when it knows more.
| Product | Memory | Bandwidth | AI compute | Measured speed | Price |
|---|---|---|---|---|---|
| ASUS ExpertCenter Pro ET900N G3ASUS (NVIDIA DGX Station GB300 platform) | 748 GB coherent: 252 GB HBM3e (GPU) + 496 GB LPDDR5X (CPU)[1] | 7.1 TB/s (GPU HBM3e); 396 GB/s (CPU LPDDR5X)[1] | up to 20 PFLOPS AI performance (ASUS gives no precision or sparsity label) (sparse)[2] | GPT-OSS-120B MXFP4, vLLM, 512/512, 1-128 streams: 9,280 aggregate output tok/s at 128 streams; 5,023 on 8,192/1,024 (StorageReview, ASUS-supplied unit, no add-in GPU)[3] | $111,999.99 at Central Computers, US (ET900N G3-2SSD-V2, read Nov 8, 2026); no ASUS list price[4] |
| HP ZGX Fury AI StationHP | 252 GB HBM3e GPU + 496 GB LPDDR5x CPU (4x 128 GB SOCAMM); 748 GB unified[5] | 7.1 TB/s (GPU HBM3e); 396 GB/s (CPU LPDDR5x)[5] | up to 20 petaFLOPS FP4 (HP gives no sparsity label) (sparse)[5] | Two-unit cluster with an ASUS ET900N G3 over two 400G DACs: GLM-5.3 NVFP4 5,018 output tok/s at 256 streams, 512/512, vLLM 0.27.1 TP2 with MTP (StorageReview); no single-unit HP result[6] | Japan list price from 28,490,000 yen incl. tax (sales from Oct 27, 2026); no US list price published[7] |
| NVIDIA DGX StationNVIDIA (systems built by ASUS, Dell, GIGABYTE, HP, MSI, Supermicro and Exxact) | 748 GB coherent: 252 GB HBM3e (GPU) + 496 GB LPDDR5X (CPU)[8] | GPU HBM3e 7.1 TB/s; CPU LPDDR5X 396 GB/s; NVLink-C2C 900 GB/s between them[8] | 15 PFLOPS FP4 (without sparsity)[8] | gpt-oss-120b MXFP4, vLLM, 512/512 chat workload: 9,280 output tok/s at 128 streams (MSI WS300 9,294), single GB300 (StorageReview, ASUS ExpertCenter Pro ET900N G3)[3] | No NVIDIA list price; lowest US partner listing read Nov 4, 2026: Exxact Valence VWS-158270643 from $98,450 (Tom's Hardware found $94,930 in Aug); Dell Pro Precision with GB300 lists $160,195[9] |
| Supermicro Super AI Station (ARS-511GD-NB-LCC)Supermicro (NVIDIA DGX Station platform) | 748 GB total coherent: 252 GB HBM3e GPU + 496 GB LPDDR5X system[10] | 7.1 TB/s (GPU HBM3E); 396 GB/s (CPU LPDDR5X)[11] | up to 20 PFLOPS FP4 (Supermicro gives no sparsity label; ServeTheHome describes it as sparse FP4)[10] | Qwen3.8 Flash Next NVFP4, vLLM, tool-loop 4,096-in/512-out: several scenarios exceeded 3,000 output tok/s at 32 concurrent users, still rising (ServeTheHome; RTX PRO 2000 installed)[12] | — |
| NVIDIA RTX PRO 6000 Blackwell Workstation EditionNVIDIA (sold through PNY, TD SYNNEX and workstation makers) | 96 GB[13] | 1,792 GB/s[13] | 4,000 AI TOPS (FP4, with sparsity) (sparse)[13] | gpt-oss-120b MXFP4, stock vLLM, 512/512 chat workload: 3,064 output tok/s at 128 streams (1,169 on 8,192/1,024 prompts); Server Edition at 600W (StorageReview)[14] | $16,000 on NVIDIA's US marketplace (Workstation Edition), listed out of stock[15] |
| Apple Mac Studio (M5 Ultra)Apple | up to 512 GB unified (96 GB on 30-core M5 Ultra; 256 GB or 512 GB on 36-core/80-core M5 Ultra)[16] | 1.2 TB/s[16] | — | Qwen 3.8 27B 4-bit MLX in LM Studio: about 55 tok/s on M5 Ultra (36-core, 256 GB) vs about 40 on M3 Ultra (BGR)[17] | $22,299 minimum with 512 GB ($6,799 36-core CPU/80-core GPU M5 Ultra config plus $15,500 memory upgrade; Oct 28, 2026)[18] |
| NVIDIA RTX Spark (N1X) PCsNVIDIA (MediaTek collaborated on the custom CPU design) | Up to 128 GB LPDDR5X (20-core CPU / 6,144-core GPU version); up to 64 GB (18-core / 5,120-core laptop version)[19] | Up to 300 GB/s: LPDDR5X-9400 on a 256-bit bus (NVIDIA's product page lists no bandwidth)[20] | — | gpt-oss:20b via Ollama in a Codex CLI coding agent, 64K context: 58.5 tok/s generation, 3 of 4 tasks passed (Dell XPS 16, 20-core N1X, 64 GB; StorageReview)[21] | Laptops from $2,599; pricing and configurations vary by region[22] |
| AMD Ryzen AI Halo (Ryzen AI Max+ PRO 495, 192 GB)AMD | 192 GB LPDDR5X (256-bit)[23] | 273 GB/s[23] | — | — | $7,299.99 for the Linux version (released the week of Oct 12, 2026); Windows version coming soon[23] |
| ASUS Ascent GX10ASUS | 64 GB or 128 GB LPDDR5X-8533 unified[24] | 273 GB/s (8533 MT/s), as listed by StorageReview; ASUS's spec page states no bandwidth[25] | Up to 1 PFLOP FP4 (GB10; NVIDIA's footnote: theoretical, using the sparsity feature). ASUS's pages say 1 petaFLOP / FP4 with no sparsity label (sparse)[26] | gpt-oss-120b, vLLM online serving, equal ISL/OSL: 70 tok/s at batch 1 rising to 680 tok/s aggregate at batch 64 (StorageReview)[25] | $5,999 (128 GB, 1 TB, GX10-GG0010BN, ASUS eShop US, Oct 30, 2026)[27] |
| Dell Pro Precision with GB10 (model FCM1253; launched as Dell Pro Max with GB10)Dell | 128 GB LPDDR5x up to 8533 MT/s onboard (64 GB option also listed)[28] | 273 GB/s (128 GB LPDDR5x, 256-bit)[29] | — | gpt-oss-120b, vLLM online serving, equal ISL/OSL: 68.57 tok/s at batch 1 rising to 670.83 tok/s aggregate at batch 64 (StorageReview)[30] | $7,901 (128 GB, 2 TB Gen4 QLC SSD default config FCM1253, Dell.com US, in stock, read Nov 8, 2026 15:14 UTC); 64GB LPDDR5x option listed at a $2,150 difference[28] |
| Framework Desktop (Ryzen AI Max+ PRO 495, 192 GB)Framework | 192 GB LPDDR5X-8533 soldered[31] | 273 GB/s (256-bit)[31] | — | — | $6,799 (Max+ PRO 495 192 GB DIY Edition pre-order, Nov 3, 2026; storage not included)[32] |
| GIGABYTE AI TOP ATOMGIGABYTE | 128 GB unified system memory[33] | 273 GB/s (LPDDR5x-8533, 256-bit)[34] | — | gpt-oss-120b, vLLM online serving, equal ISL/OSL: 60.71 tok/s at batch 1 rising to 649.62 tok/s aggregate at batch 64 (StorageReview)[34] | $7,499.99 (128 GB, 4 TB PCIe 4.0) and $9,405.63 (128 GB, 4 TB PCIe 5.0; more options from $7,999.99); 128 GB from $5,999.99 (1 TB) and $6,999.99 (2 TB) at Newegg US, read Nov 9, 2026[35] |
| GMKtec EVO-X5 Pro AI Mini PCGMKtec | 192 GB LPDDR5X-8533 onboard (8 x 24 GB, non-upgradeable); up to 160 GB allocatable as VRAM[36] | Up to 273 GB/s[37] | — | GMKtec Laboratory table: GLM-5.3-Flash UD-IQ4_XS (146.05 GB) decode 17.79 tok/s, prefill 70.18, TTFT 14.71 s; Qwen3.8-Flash-Next ROCmFP4-FAST (93.48 GB) decode 45.08 tok/s[36] | $6,799 (192 GB/2 TB) and $7,099 (192 GB/4 TB), GMKtec US store, Nov 2, 2026[36] |
| HP ZGX Nano AI Station (G1n)HP | 128 GB coherent unified system memory[38] | 273 GB/s (128 GB LPDDR5x)[39] | GB10: up to 1 PFLOP FP4, theoretical, using the sparsity feature; HP and reviewers list 1,000 TOPS FP4 (sparse)[26] | gpt-oss-120b, vLLM online serving, equal ISL/OSL: 69.6 tok/s at batch 1 rising to 722.9 tok/s aggregate at batch 64 (StorageReview)[40] | $7,399.00 at the HP Store for the 4 TB model; $4,759 on Amazon (Notebookcheck, review updated Oct 5, 2026)[41] |
| Lenovo ThinkStation PGXLenovo | 128 GB LPDDR5X unified (256-bit)[42] | 273 GB/s[42] | — | — | about $4,100 for the 1 TB configuration (Mar 2026)[43] |
| MSI EdgeXpert (MS-C931)MSI | 64 GB or 128 GB LPDDR5x unified (256-bit)[44] | 273 GB/s[44] | 1 PFLOPS FP4 (sparse)[44] | — | No list price; MSI sells via quote (Get a quote)[44] |
| NVIDIA DGX SparkNVIDIA (partner GB10 systems from Acer, ASUS, Dell, GIGABYTE, HP, Lenovo and MSI) | 128 GB; a 64 GB configuration is sold only through OEM partners[26] | 273 GB/s[26] | Up to 1 PFLOP FP4 (theoretical, using the sparsity feature) (sparse)[26] | Qwen 3.8 27B (dense) with multi-token prediction: decode of about 20 tokens/s, with time-to-first-token staying interactive out to the full 262K context.[45] | $6,950.00 (128 GB, 4 TB) on NVIDIA's enterprise marketplace with an Add To Cart button in a copy cached 07:59 UTC on 2026-10-31; stock unverified (listed out of stock on 2026-10-21); 64 GB partner systems from $4,999[46] |
| NVIDIA DGX Spark 64 GBNVIDIA (sold by Acer, ASUS, Dell, Gigabyte, HP and MSI) | 64 GB LPDDR5X unified (two units pool 128 GB over ConnectX-7)[47] | 273 GB/s (256-bit; NVIDIA spec table covers both 64 GB and 128 GB)[26] | 1 PFLOPS FP4 (sparse; NVIDIA spec table covers both memory configurations)[26] | — | from $4,999 (partner systems, Oct 2026; no Founders Edition)[47] |
| Sipeed MaixStationSipeed | 128 GB LPDDR5X (256-bit)[23] | 273 GB/s (Sipeed claim)[23] | 256 TOPS FP4 (native FP4, no sparsity; Axera/Sipeed claim)[23] | — | about $3,999 expected for 128 GB, with Kickstarter launch discounts (Sipeed to Tom's Hardware, Oct 2026)[23] |
| AMD Ryzen AI Halo Developer PlatformAMD | 128 GB LPDDR5x-8000 unified[48] | 256 GB/s[48] | — | gpt-oss-120b, vLLM online serving, equal ISL/OSL 256/256: 222 tok/s aggregate at batch 64 vs 701 on DGX Spark (StorageReview)[49] | $3,999 with 2 TB SSD (Jul 2026)[49] |
| Framework Desktop (Ryzen AI Max+ 395)Framework | 128 GB LPDDR5x-8000 soldered (32 GB and 64 GB options)[50] | 256 GB/s[51] | — | gpt-oss-120b MXFP4 in LM Studio (out-of-the-box setup): roughly 45 tok/s (ServeTheHome; Ryzen AI Max+ 395, 128 GB, Windows 11 Pro)[52] | $3,449 (DIY Edition, Max+ 395 with 128 GB, system only; out of stock, Nov 3, 2026)[53] |
| Acer Veriton GN100 AI Mini WorkstationAcer | 128 GB LPDDR5x coherent unified[54] | — | — | gpt-oss-120b, vLLM, total token throughput: prefill heavy 303.61 to 2,777.56 tok/s; decode heavy 38.38 to 292.41 tok/s (batch 1 to 64, StorageReview)[55] | from $3,999 in North America (announced Sep 3, 2025; prices vary by region)[54] |
| ASRock Industrial AI BOX-A395ASRock Industrial | 128 GB LPDDR5X-8000 soldered[56] | — | — | AgentSTH V7, 32 threads: 15% ahead of the next-closest Ryzen AI Max+ 395 box (single-core within 1%); estimated SPEC CPU2026 int rate: top of a 44% spread across four 395 systems; measured by ServeTheHome, Ubuntu 26.04[57] | — |
| Dell Pro Precision with GB300 (model FCT6263; announced as Dell Pro Max with GB300)Dell | 252 GB HBM3e GPU + 496 GB LPDDR5X CPU[58] | — | — | — | $160,195.00 (Dell.com US default config FCT6263, read Nov 8, 2026)[58] |
| GMKtec EVO-X2 AI Mini PCGMKtec | 64 GB or 128 GB onboard LPDDR5X-8000 (non-upgradeable)[59] | — | — | Llama 3.3 70B Q6_K Instruct: 3.7-3.8 tok/s generation fully on the Radeon 8060S (about 2 tok/s CPU-only); Qwen3-32B 10.4-10.5 tok/s; 128 GB unit, short prompt, tool not named (ServeTheHome)[60] | $3,499.99 (128 GB/1 TB) and $3,649.99 (128 GB/2 TB), each against $3,999.99; $2,199.99 for 64 GB/1 TB against $2,599.99 (GMKtec US store, read Nov 8, 2026)[59] |
| HP Z2 Mini G1a WorkstationHP | up to 128 GB unified, up to 96 GB assignable to the GPU, up to 8000 MT/s[61] | — | up to 50 NPU TOPS[61] | gpt-oss-120b GGUF (63.39 GB, full GPU offload) in LM Studio: close to 40 tok/s (StorageReview)[62] | $4,781 list for StorageReview's review configuration; about $3,300 street (B&H top spec with 2 TB SSD at $3,342.65, Aug 2025)[62] |
| Microsoft Surface RTX Spark Dev BoxMicrosoft | 128 GB unified memory[63] | — | up to 1 PFLOP of AI compute (Microsoft; precision and sparsity not stated on the page) (sparse)[63] | — | From $5,999.99 (Microsoft Store US, pre-order, read Nov 1, 2026)[63] |
| MSI XpertStation WS300MSI (NVIDIA DGX Station platform) | 252 GB HBM3e GPU + 496 GB LPDDR5X CPU (748 GB coherent)[64] | — | — | StorageReview vLLM: gpt-oss-120b MXFP4 9,294 output tok/s at 128 streams (512/512), 5,038 on 8,192/1,024; MiniMax M3 NVFP4 1,041 at 32 streams with EAGLE3-GQA at a forced acceptance length of 3 (best-case)[64] | $99,999.00 'Start From' at Newegg US (model WS300T60L, 2x 2 TB NVMe RAID1), sold by Newegg as a request-a-quote listing (read Nov 4, 2026)[65] |
| Tenstorrent TT-QuietBox 2 (Blackhole)Tenstorrent | 128 GB GDDR6 on the accelerators + 256 GB DDR5-5600 system memory[66] | — | — | — | $9,999 (Oct 30, 2026)[66] |
| WiCi One (RTX 5060 Ti edition)WiCi | 16 GB GDDR7 VRAM; 1 TB NVMe 4 SSD for TurboStream weight paging (WiCi's claim: NVMe used as working VRAM). System RAM not stated[67] | — | — | — | $2,599 list (WiCi shop, Oct 2026); a $600-off code (WICIWELCOME) is applied at checkout. Early-access price was $1,999 (Notebookcheck, Engadget, Sep 2026)[67] |
| WiCi One (RTX 5090 edition)WiCi | 32 GB GDDR7 VRAM; 4 TB NVMe 5 SSD for TurboStream weight paging (WiCi's claim: NVMe used as working VRAM). System RAM not stated[67] | — | — | — | $9,999 list (WiCi shop, Oct 2026); a $600-off code (WICIWELCOME) is applied at checkout[67] |
Italic values are reported by an outlet or measured by an independent lab; the rest come from the maker. Measured speeds use different models, quantizations and tools, so read each with its source rather than against the others. Voltgrid News does not test hardware itself.
Local AIProduct file
