Gigabyte AI TOP ATOM Review: $7,312 at CDW, Up From $3,999.99 at Launch
CDW advertises Gigabyte's 4TB GB10 box at $7,312, 83% above launch and above NVIDIA's DGX Spark. StorageReview's vLLM runs and ServeTheHome's power data show what that buys.
Local AI DeskEdited by Ege Erten
7 min readAI-assisted

CDW advertised $7,312 for the 128 GB, 4 TB Gigabyte AI TOP ATOM on Nov. 2, 2026; Newegg's reported launch price for the 4TB PCIe 5.0 model was $3,999.99.
Illustration (AI-generated)
CDW advertised Gigabyte's AI TOP ATOM with 128 GB of memory and a 4 TB SSD at $7,312 on Nov. 2, 2026, against a $7,999.99 list price and 83% above its $3,999.99 launch price.

Gigabyte AI TOP ATOM measured 649.62 tok/s at batch 64, the lowest of five GB10 boxes in separate StorageReview reviews; precision not stated.
Illustration (AI-generated)
That price is the news for buyers, because the machine itself has been measured. StorageReview ran the AI TOP ATOM through its vLLM serving tests in March, and ServeTheHome measured its wall power the same week. Both found NVIDIA's GB10 reference design in a black Gigabyte chassis, with speed that tracks the other GB10 boxes. What changed since then is the cost: in March, ServeTheHome described Gigabyte's GB10 boxes as about $700 cheaper than NVIDIA's own DGX Spark. On CDW's Nov. 2 listing, the 4 TB model now costs $362 more than the $6,950 NVIDIA showed for its 4 TB DGX Spark on Oct. 31. Voltgrid did not test the unit; every figure below names the outlet that measured it.
What's new in the Gigabyte AI TOP ATOM
The chip is not new. Gigabyte's Oct. 13, 2025 release puts the AI TOP ATOM on NVIDIA's GB10 Grace Blackwell Superchip with 128 GB of unified memory, rates it at up to 1 petaFLOP of FP4 compute, and says it runs models of up to 200 billion parameters, or 405 billion with two units linked through the built-in ConnectX-7 network card. NVIDIA states the same 1 PFLOP FP4 figure for GB10 using its sparsity feature. Gigabyte said the box would go on sale worldwide on Oct. 15, 2025.
Unified memory is the design choice that matters here. The CPU and GPU read one 128 GB pool of LPDDR5x over a 256-bit bus at 273 GB/s, StorageReview's spec table shows, so the GPU can hold a 120-billion-parameter model such as gpt-oss-120b in one piece. The same 273 GB/s caps how fast each new token can be generated, because every token reads the model's active weights from that memory.
What Gigabyte controls is the box around the board. ServeTheHome found no front ports or buttons, a power button on the back, two 3 W Delta fans on a large heatsink, and an M.2 2242 SSD slot wired to PCIe 5.0 x4. Gigabyte sells a 1 TB model and two 4 TB models, and only the top 4 TB version ships a PCIe 5.0 drive, ServeTheHome said; the others use PCIe 4.0. StorageReview's 4 TB unit carried Samsung's PM9E1 Gen5 drive.
Software comes from NVIDIA: DGX OS, a version of Ubuntu, plus NVIDIA Sync, the DGX Dashboard, AI Workbench and the DGX Spark Playbooks, ServeTheHome reported. Gigabyte adds its AI TOP Utility for fine-tuning and inference, according to its release. In the US, ServeTheHome said, Gigabyte works with AVADirect to offer next-day on-site repair.
What it solves
The AI TOP ATOM lets a team run a 120-billion-parameter model on a desk without a cloud account or a server room. StorageReview served OpenAI's gpt-oss-120b with the vLLM online serving benchmark, at equal input and output lengths, and measured 60.71 tokens per second at batch 1, rising to 649.62 tokens per second in aggregate at batch 64. The review does not state the model's precision. With prompt-heavy requests the total reached 2,771.38 tokens per second at batch 64; with output-heavy requests it ran from 47.01 to 299.41.
Smaller models scale further on the same test. StorageReview measured gpt-oss-20b from 641.73 to 1,587.62 tokens per second, Qwen3 Coder 30B A3B in FP8 from 101.27 to 1,251.84, and Llama 3.1 8B Instruct in FP4 from 69.84 to 2,756.91, each from batch 1 to batch 64. The shape matters more than any one number: one user gets about 60 tokens per second from gpt-oss-120b, while a team sharing the box gets roughly ten times that in total.
Storage is the other problem it solves. With the PM9E1, StorageReview measured NVIDIA's GPU Direct Storage reads, which move data from the SSD to the GPU without passing through the CPU, at 11.03 to 11.12 GiB/s from 8 to 64 threads, and 1 MB writes at about 12.2 GiB/s. It called these some of the strongest results in its group of GB10 systems.
Pros and Cons
The speed figures come from separate StorageReview reviews that used the same vLLM benchmark: Gigabyte, Dell, ASUS and Acer in March, and HP's ZGX Nano in an April review. StorageReview put the AI TOP ATOM in the middle of the field and said separation between GB10 boxes is narrow. ServeTheHome's own runs agree: Geekbench 6 placed the AI TOP ATOM a few percent behind ASUS's Ascent GX10, and its AI tests showed a spread the outlet put down to normal run-to-run variation. After more than an hour of heat soak, the Gigabyte box came out a few percent ahead of NVIDIA's DGX Spark in most of ServeTheHome's eight AI tests, winning three and tying one.
The heat comes with that small edge. StorageReview logged a 90°C CPU peak and an 81°C GPU peak during prompt-heavy bursts, toward the upper bound of its group, with GPU power peaking at 75.54 W, at the top end of the GB10 units it tested. Temperatures levelled off under sustained output-heavy load, and StorageReview read the peaks as a power-allocation choice, not a cooling failure. The NVMe drive stayed at or below 59.8°C. The idle figure has a cause: ServeTheHome said unplugging the ConnectX-7 link saves about 18 W. At 36.5 W around the clock, an idle box draws about 320 kWh a year.
Who the Gigabyte AI TOP ATOM is for
It fits a small team that shares one machine for a model in the 120B class. StorageReview's 649.62 tokens per second at batch 64 on gpt-oss-120b is aggregate capacity across up to 64 concurrent requests, at a sustained draw that ServeTheHome measured at 102 to 158 W.
It fits a developer building for NVIDIA's data-center stack. The software is NVIDIA's DGX OS, and the two 200 Gbps ports let two units pool memory for models up to 405 billion parameters, by Gigabyte's claim.
It fits data-heavy work on the top 4 TB model, where StorageReview's GPU Direct Storage results put it near the front of the GB10 group.
It does not fit a single user chasing tokens per second on a budget. All GB10 boxes share the 273 GB/s memory that limits output speed, and StorageReview measured just 60.71 tokens per second at batch 1 on gpt-oss-120b for this unit. At CDW's $7,312, the same chip is available for less: ASUS listed its 1 TB Ascent GX10 at $5,999 on Oct. 30, though its US store showed it out of stock on Nov. 2, and NVIDIA lists 64 GB partner systems from $4,999.
How it compares with DGX Spark, HP ZGX Nano and Dell
Every rival here uses the same GB10 chip, the same 128 GB of memory and the same 273 GB/s, so price and the chassis decide. On StorageReview's vLLM test with gpt-oss-120b at batch 64, the AI TOP ATOM's 649.62 tokens per second sits below Dell's Pro Max with GB10 (now listed as Pro Precision) at 670.83, ASUS at 680, Acer's Veriton GN100 at 713.18 and HP's ZGX Nano at 722.9. Each comes from a separate StorageReview review on the same gpt-oss-120b vLLM test, with precision unstated, and the HP figure is from an April review.
NVIDIA's DGX Spark, as Voltgrid reported, lists at $6,950 for 128 GB and 4 TB on NVIDIA's marketplace, though its stock was unverified on Oct. 31. That makes the Gigabyte box at CDW $362 dearer, reversing the gap ServeTheHome described in March, when it priced Gigabyte's 4 TB model at about $4,000 against $4,699 for the Spark.
HP's ZGX Nano is the cooler-running alternative: Notebookcheck reported $7,399 at the HP Store for 4 TB. Dell's Pro Precision with GB10 read $7,967 on Dell.com later on Nov. 2 for a 2 TB Gen4 QLC model, after $8,103 earlier that day, and StorageReview logged its CPU near 87.7°C, below Gigabyte's 90°C. On storage, the Gigabyte and NVIDIA boxes both use the PM9E1 Gen5 drive on their top models.
The table under this story shows each catalog value with its source. What it says, read across: StorageReview's separate reviews put the boxes close together on speed, while listed 4 TB prices read $6,950 (NVIDIA, Oct. 31) to $7,399 (HP Store, per Notebookcheck, Oct. 5), with CDW's Gigabyte at $7,312 on Nov. 2.
What to watch
Prices first. ServeTheHome warned in March that GB10 prices were being reset for DRAM costs, and CDW's Nov. 2 figure shows how far that went: $7,312, up from $3,999.99 for the 4 TB Gen5 model at launch on Newegg, as CNX Software reported in October 2025. Whether CDW's advertised price holds through November 2026 is the first test. CDW's listing does not say whether model ATAGB10-9000 carries the Gen5 or the Gen4 drive, and Gigabyte's US site could not be read to confirm a list price.
NVIDIA now lists 64 GB partner systems from $4,999, and no Gigabyte 64 GB listing has appeared yet. NVIDIA reports its third-quarter fiscal 2027 results on Nov. 17, 2026, the next chance for word on GB10 supply and memory costs. Independent single-user numbers from llama.cpp or Ollama on this unit are still missing.
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 ATOMGIGABYTEThis story | 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
- 03Gigabyte AI TOP ATOM NVIDIA GB10 Review A Neat Little BoxServeTheHome · servethehome.com
- 04GIGABYTE AI TOP ATOM ReviewStorageReview · storagereview.com
- 05GIGABYTE Announces its Personal AI Supercomputer AI TOP ATOM Will be Available Globally on October 15GIGABYTE (PR Newswire) · tools.prnewswire.com
- 06Gigabyte AI TOP ATOM - compact PC - AI PC - GB10 Grace Blackwell Superchip - 128 GB - SSD 4 TBCDW · cdw.com







