GB10 Boxes Compared: Similar gpt-oss-120b Speed, Prices $4,759 to $7,967
StorageReview's separate vLLM reviews put six NVIDIA GB10 boxes close together on gpt-oss-120b. Listed US prices run $4,759 to $7,967 across configurations and dates.
Local AI DeskEdited by Ege Erten
6 min readAI-assisted
Follows our story: Dell Pro Precision GB10 Review: Dell.com Lists 128GB Box at $8,103

NVIDIA's DGX Spark family of GB10 boxes. The image shows the Spark and partner systems, not each box reviewed here.
Image: NVIDIA
Source · Press image from the NVIDIA product page, used for news reporting.
Six NVIDIA GB10 boxes sold in the US land close together on gpt-oss-120b in StorageReview's vLLM serving tests: four boxes from one March series sit within 9.8% at batch 64, and two later runs fall in the same range. Their listed US prices, for different configurations and dates, run from $4,759 to $7,967. Speed does not separate them. Price, stock, storage and cooling do.
The boxes are NVIDIA's DGX Spark, the ASUS Ascent GX10, the Acer Veriton GN100, the Gigabyte AI TOP ATOM, the Dell Pro Precision with GB10 and the HP ZGX Nano. This piece builds on our reviews of the Dell Pro Precision with GB10, the Gigabyte AI TOP ATOM, the HP ZGX Nano and the DGX Spark. Since Oct. 30, Dell.com has listed its 128 GB model at three different prices, ASUS's US store showed the GX10 out of stock on Nov. 2, and CDW lists the Gigabyte ATOM at $7,312, below its $7,999.99 list price.
Why GB10 Boxes Perform Alike
Every box here is built on the same GB10 Grace Blackwell Superchip with 128 GB of LPDDR5X memory on a 256-bit bus. NVIDIA specifies 273 GB/s of bandwidth, and the CPU and GPU share that one pool. When a model writes text, each new token requires reading the model's active weights from that memory, so bandwidth sets the ceiling on speed. All six makers share that ceiling. What they change is the case, the fan curve, the SSD and the price tag. NVIDIA says one 128 GB unit runs models of up to 200 billion parameters. A 120B model such as gpt-oss-120b therefore fits on any of them.
gpt-oss-120b Speed: One Lab, One Test, Three Dates
StorageReview ran the same vLLM online serving benchmark on each box. It used gpt-oss-120b with equal input and output lengths (Equal ISL/OSL) and swept the batch from 1 to 64 concurrent requests. The batch-64 figure is total throughput across all requests, not the speed one user sees.
Four of the boxes were tested in one series, published Mar. 3, 2026. At batch 64, the Gigabyte ATOM reached 649.62 tok/s, the Dell 670.83, the ASUS 680 and the Acer 713.18. At batch 1, a single user's case, they ran 60.71, 68.57, 70 and 69.65 tok/s. The Acer finished 9.8% ahead of the Gigabyte. In the text of those reviews, StorageReview named the Founders Edition Spark as a reference but did not state its numbers.

StorageReview's batch-64 figures come from three separate runs, Mar. 3, Apr. 24 and Jul. 6, 2026, and are not one table.
Illustration (AI-generated)
The HP ZGX Nano came later, in a review published Apr. 24. It ran the same three workloads and climbed from 69.6 tok/s at batch 1 to 722.9 tok/s at batch 64. That is the top figure in the set, 11.3% above the Gigabyte. Because it comes from a separate run a month and a half later, part of that lead may reflect software rather than hardware. StorageReview did not say.
The DGX Spark's figure is from a third date. In its July 6 review of AMD's Ryzen AI Halo, StorageReview measured the Spark at 701 tok/s at batch 64, with 256 input and 256 output tokens stated, against 222 tok/s for the Halo. We covered that gap in our Halo review. The model and tool match the March series, but the software stack was four months newer. Treat it as the same ballpark, not a ranked result.
On the March figures alone, the spread is under 10%. Adding HP's April run and the Spark's July run widens it to 11.3%, and nothing in the data shows any box pulling away. A single user should also expect less than these totals. Notebookcheck, using its own test rather than StorageReview's, saw gpt-oss 120B on the HP drop below 5 tok/s as the context filled.
GB10 Price Spread: $3,208 for the Same Chip
Dell.com listed the Dell Pro Precision with GB10, with 128 GB and its default 2 TB Gen4 QLC drive, at $7,967 at 16:40 UTC on Nov. 2. It read $8,103 earlier that day and $6,593 on Oct. 30. That makes it the most expensive box here and the one whose price has moved most.
CDW advertised the Gigabyte AI TOP ATOM with 128 GB and 4 TB at $7,312 on Nov. 2, against a $7,999.99 list price. NVIDIA's enterprise marketplace showed the DGX Spark with 128 GB and 4 TB at $6,950 in a copy cached on Oct. 31. Its stock is unverified, and the same listing read out of stock on Oct. 21. The Spark launched at $3,999 in October 2025.
The ASUS eShop in the US lists the Ascent GX10 with 128 GB and 1 TB at $5,999. On Oct. 30 and again on Nov. 2, it showed the unit as out of stock, with a notify-me button only.
HP's own product page shows no price for the ZGX Nano. Notebookcheck, in a review updated Oct. 5, reported $7,399 at the HP Store for the 4 TB model and $4,759 on Amazon. That Amazon price is the low end of the range, and it is a month old. Acer announced the Veriton GN100 from $3,999 in North America in September 2025. We found no current US listing to confirm that it still applies.
Since Oct. 23, NVIDIA's partners have also sold a 64 GB GB10 configuration from $4,999, as Voltgrid reported. NVIDIA rates it for models of up to 100 billion parameters, so gpt-oss-120b is out of reach on one unit. No independent lab has published tests of it yet.
Cooling, Storage and Power Compared
The boxes differ more on thermals than on speed. In StorageReview's one-hour, three-stage load ramp, the Acer peaked at 74.7°C on the CPU and 69°C on the GPU, the coolest of the group. The HP peaked at 77.3°C and 69°C in its April run. The ASUS reached 87.3°C and 82°C, and the Dell about 87.7°C and 80°C. The Gigabyte ran hottest at 90°C and 81°C, with GPU power peaking at 75.54 W, the top of the group. StorageReview found that all of them settled once the burst phase passed.
Storage is where the configurations split. StorageReview put the ASUS's 1 TB Gen4 SSD near the bottom of its storage tests. The Dell ships a QLC drive on its 2 TB model, offers no Gen5 option, and logged the warmest NVMe drive in the group. The Gigabyte supports up to a 4 TB Gen5 SSD and posted some of the best GPU Direct Storage results. Dell is also the only one with a 280 W adapter. NVIDIA, ASUS, Gigabyte and HP ship 240 W supplies.
Which GB10 Box to Buy
For the lowest outlay on a 128 GB box with a full independent review, buy the HP ZGX Nano if the $4,759 Amazon price Notebookcheck reported is still available. It ran cool, and it posted the top batch-64 figure in StorageReview's tests. At HP's own $7,399, the case for it is weaker.
For a quiet desk or a warm office, the Acer Veriton GN100 ran coolest and matched the field on speed. Confirm a current price first, because Acer's $3,999 is from September 2025.
For a team that wants NVIDIA's own software image and support, the DGX Spark at $6,950 sits mid-range for 4 TB. Its stock was unconfirmed when we checked.
For large local datasets, the Gigabyte AI TOP ATOM's 4 TB Gen5 option and its GPU Direct Storage results justify its $7,312 CDW price. It runs hottest under bursts, so give it room to breathe.
The ASUS Ascent GX10 is the value pick at $5,999 once it is back in stock. Its 1 TB drive is the trade-off.
The Dell Pro Precision with GB10 at $7,967 measured no faster than the $5,999 ASUS in the same StorageReview series. Buy it for Dell procurement and support, and wait for its price to settle.
Skip the 64 GB GB10 systems at $4,999 if gpt-oss-120b is the target, and wait for the first independent tests. If vLLM serving is the job, AMD's $3,999 Ryzen AI Halo is not a substitute: it reached 222 tok/s against the Spark's 701 on the same test.
How it compares
Full table| Product | Memory | Bandwidth | AI compute | Measured speed | Price |
|---|---|---|---|---|---|
| HP ZGX Fury AI StationHP | 252 GB HBM3e GPU + 496 GB LPDDR5x CPU (4x 128 GB SOCAMM); 748 GB unified[1] | 7.1 TB/s (GPU HBM3e); 396 GB/s (CPU LPDDR5x)[1] | — | 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[2] | Japan list price from 28,490,000 yen incl. tax (sales from Oct 27, 2026); no US list price published[3] |
| 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)[4] | GPU HBM3e 7.1 TB/s; CPU LPDDR5X 396 GB/s; NVLink-C2C 900 GB/s between them[4] | 15 PFLOPS FP4 (without sparsity)[4] | Over 26,000 output tokens/s aggregate on Qwen3.6-35B-A3B NVFP4 at 512 concurrent users (24-30-token input, 1,024-token output), Gigabyte W775-V10-L01; idle never below 250 W.[5] | No NVIDIA list price; first public US OEM list price is Dell Pro Precision with GB300 at $160,195 default config (Dell.com, read Nov 1, 2026); ServeTheHome had heard about $100K[6] |
| NVIDIA RTX PRO 6000 Blackwell Workstation EditionNVIDIA (sold through PNY, TD SYNNEX and workstation makers) | 96 GB[7] | 1,792 GB/s[7] | 4,000 AI TOPS (FP4, with sparsity) (sparse)[7] | LM Studio: Mistral Small (26 GB) at 42.4 tokens/s vs 17 on an RTX 5090; Gemma 3 27B at 29 vs 5; tied with the RTX 5090 at about 81 tokens/s on DeepSeek Llama 8B Distil, where VRAM does not limit.[8] | $16,000 on NVIDIA's US marketplace (Workstation Edition), listed out of stock[9] |
| 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)[10] | 1.2 TB/s[10] | — | 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)[11] | $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)[12] |
| 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)[13] | Up to 300 GB/s: LPDDR5X-9400 on a 256-bit bus (NVIDIA's product page lists no bandwidth)[14] | — | 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)[15] | Laptops from $2,599; pricing and configurations vary by region[16] |
| 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)[17] | 273 GB/s (128 GB LPDDR5x, 256-bit)[18] | — | 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)[19] | $7,967 (128 GB, 2 TB Gen4 QLC SSD default config, Dell.com US, read Nov 2, 2026 16:40 UTC)[17] |
| Framework Desktop (Ryzen AI Max+ PRO 495, 192 GB)Framework | 192 GB LPDDR5X-8533 soldered[20] | 273 GB/s (256-bit)[20] | — | — | from $6,799 (DIY Edition pre-order, Oct 30, 2026)[21] |
| GIGABYTE AI TOP ATOMGIGABYTE | 128 GB unified system memory[22] | 273 GB/s (LPDDR5x-8533, 256-bit)[23] | — | 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)[23] | $7,312.00 advertised at CDW US against a $7,999.99 list (128 GB, 4 TB, ATAGB10-9000; read Nov 2, 2026)[24] |
| 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[25] | Up to 273 GB/s[26] | — | 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[25] | $6,799 (192 GB/2 TB) and $7,099 (192 GB/4 TB), GMKtec US store, Nov 2, 2026[25] |
| HP ZGX Nano AI Station (G1n)HP | 128 GB coherent unified system memory[27] | 273 GB/s (128 GB LPDDR5x)[28] | GB10: up to 1 PFLOP FP4, theoretical, using the sparsity feature; HP and reviewers list 1,000 TOPS FP4 (sparse)[29] | 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)[30] | $7,399.00 at the HP Store for the 4 TB model; $4,759 on Amazon (Notebookcheck, review updated Oct 5, 2026)[31] |
| Lenovo ThinkStation PGXLenovo | 128 GB LPDDR5X unified (256-bit)[32] | 273 GB/s[32] | — | — | about $4,100 for the 1 TB configuration (Mar 2026)[33] |
| MSI EdgeXpert (MS-C931)MSI | 64 GB or 128 GB LPDDR5x unified (256-bit)[34] | 273 GB/s[34] | 1 PFLOPS FP4 (sparse)[34] | — | No list price; MSI sells via quote (Get a quote)[34] |
| 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[29] | 273 GB/s[29] | Up to 1 PFLOP FP4 (theoretical, using the sparsity feature) (sparse)[29] | 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.[35] | $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[36] |
| 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)[37] | 273 GB/s (256-bit; NVIDIA spec table covers both 64 GB and 128 GB)[29] | 1 PFLOPS FP4 (sparse; NVIDIA spec table covers both memory configurations)[29] | — | from $4,999 (partner systems, Oct 2026; no Founders Edition)[37] |
| AMD Ryzen AI Halo Developer PlatformAMD | 128 GB LPDDR5x-8000 unified[38] | 256 GB/s[38] | — | gpt-oss-120b, vLLM online serving, equal ISL/OSL 256/256: 222 tok/s aggregate at batch 64 vs 701 on DGX Spark (StorageReview)[39] | $3,999 with 2 TB SSD (Jul 2026)[39] |
| Framework Desktop (Ryzen AI Max+ 395)Framework | 128 GB LPDDR5x-8000 soldered (32 GB and 64 GB options)[40] | 256 GB/s[41] | — | — | $3,449 (DIY Edition, Max+ 395 with 128 GB, Oct 30, 2026)[21] |
| Acer Veriton GN100 AI Mini WorkstationAcer | 128 GB LPDDR5x coherent unified[42] | — | — | gpt-oss-120b, vLLM online serving, equal ISL/OSL: 69.65 tok/s at batch 1 rising to 713.18 tok/s aggregate at batch 64 (StorageReview)[43] | from $3,999 in North America (announced Sep 3, 2025; prices vary by region)[42] |
| ASRock Industrial AI BOX-A395ASRock Industrial | 128 GB LPDDR5X-8000 soldered[44] | — | — | 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[45] | — |
| ASUS Ascent GX10ASUS | 64 GB or 128 GB LPDDR5X-8533 unified[46] | — | — | 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)[47] | $5,999 (128 GB, 1 TB, GX10-GG0010BN, ASUS eShop US, Oct 30, 2026)[48] |
| Dell Pro Precision with GB300 (model FCT6263; announced as Dell Pro Max with GB300)Dell | 252 GB HBM3e GPU + 496 GB LPDDR5X CPU[6] | — | — | — | $160,195 (Dell.com US default config with RTX PRO 2000 display GPU and 16 TB SSD, read Nov 1, 2026)[6] |
| GMKtec EVO-X2 AI Mini PCGMKtec | 64 GB or 128 GB onboard LPDDR5X-8000 (non-upgradeable)[49] | — | — | — | $3,499.99 (128 GB/1 TB, GMKtec US store, Oct 30, 2026; listed against $3,999.99)[49] |
| HP Z2 Mini G1a WorkstationHP | up to 128 GB unified, up to 96 GB assignable to the GPU, up to 8000 MT/s[50] | — | — | gpt-oss-120b GGUF (63.39 GB, full GPU offload) in LM Studio: close to 40 tok/s (StorageReview)[51] | — |
| Microsoft Surface RTX Spark Dev BoxMicrosoft | 128 GB unified memory[52] | — | up to 1 PFLOP of AI compute (Microsoft; precision and sparsity not stated on the page) (sparse)[52] | — | From $5,999.99 (Microsoft Store US, pre-order, read Nov 1, 2026)[52] |
| MSI XpertStation WS300MSI (NVIDIA DGX Station platform) | 252 GB HBM3e GPU + 496 GB LPDDR5X CPU (748 GB coherent)[53] | — | — | GPT-OSS-120B MXFP4, vLLM, 512/512 ISL/OSL, concurrency sweep to 128: peak 9,294 output tok/s vs 3,384 on an RTX PRO 6000 and 500 on DGX Spark (StorageReview, MSI-hosted unit, no add-in GPU)[53] | ServeTheHome calls it a $100K+ system; no MSI list price given[54] |
| Tenstorrent TT-QuietBox 2 (Blackhole)Tenstorrent | 128 GB GDDR6 on the accelerators + 256 GB DDR5-5600 system memory[55] | — | — | — | $9,999 (Oct 30, 2026)[55] |
| 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[56] | — | — | — | $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)[56] |
| 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[56] | — | — | — | $9,999 list (WiCi shop, Oct 2026); a $600-off code (WICIWELCOME) is applied at checkout[56] |
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.
Sources (56)
- [1] HP
- [2] StorageReview
- [3] HP Japan
- [4] NVIDIA
- [5] ServeTheHome
- [6] Dell
- [7] NVIDIA
- [8] Gamers Nexus
- [9] NVIDIA
- [10] Apple
- [11] BGR
- [12] MacRumors
- [13] NVIDIA
- [14] ServeTheHome
- [15] StorageReview
- [16] NVIDIA
- [17] Dell
- [18] HotHardware
- [19] StorageReview
- [20] Framework Blog
- [21] Framework
- [22] GIGABYTE (PR Newswire)
- [23] StorageReview
- [24] CDW
- [25] GMKtec
- [26] GMKtec
- [27] HP
- [28] HotHardware
- [29] NVIDIA
- [30] StorageReview
- [31] Notebookcheck
- [32] Lenovo PSREF
- [33] ServeTheHome
- [34] MSI
- [35] Tom's Hardware
- [36] NVIDIA Marketplace
- [37] NVIDIA Blog
- [38] AMD
- [39] StorageReview
- [40] Framework
- [41] Framework Blog
- [42] Acer
- [43] StorageReview
- [44] ServeTheHome
- [45] ServeTheHome
- [46] ASUS
- [47] StorageReview
- [48] ASUS eShop
- [49] GMKtec
- [50] HP
- [51] StorageReview
- [52] Microsoft
- [53] StorageReview
- [54] ServeTheHome
- [55] Tenstorrent
- [56] WiCi
Sources
- 01Acer Veriton GN100 Review: A Standout in the NVIDIA Spark EcosystemStorageReview · storagereview.com
- 02ASUS Ascent GX10 ReviewStorageReview · storagereview.com
- 03ASUS Ascent GX10 (GX10-GG0010BN)ASUS · eshop.asus.com
- 04GIGABYTE AI TOP ATOM ReviewStorageReview · storagereview.com
- 05Gigabyte AI TOP ATOM - compact PC - AI PC - GB10 Grace Blackwell Superchip - 128 GB - SSD 4 TBCDW · cdw.com







