$22,000 for 192 GB of card memory
The red flash has six R9700s and a 32-core EPYC. The red kimi has four on a newer PCIe 5 platform for $30,000. Both are made to order and paid by wire.
tinybox for local AI, compared
The tinybox is tiny corp’s multi-GPU AI machine: a water-cooled AMD EPYC machine with four or six Radeon AI PRO R9700 cards, the same 32 GB card a Lucebox has, built to order by the company behind tinygrad.
100B-class models fit on the cards
The red flash’s 192 GB of card memory holds DeepSeek V4 Flash (101.5 GB) and Qwen3.8-Flash-Next (106.3 GB) with room for long prompts. Split across cards, a model reads several cards at once.
No published run on a tinybox yet
One of its cards, an R9700, wrote Qwen3.8-27B at 303.38 tok/s with drafting on a short code prompt, 36.7 without. Six cards working for one user do not write six times as fast.
tinybox red, green and pro: specs of each version
As of 3 November 2026, tiny corp sells two preset tinybox versions with GPUs, the red flash and the red kimi, both built on AMD Radeon AI PRO R9700 cards. The tinybox green, green v2 and original red are earlier models that tinygrad.org no longer offers, and we found no tinybox pro on any tinygrad.org page.
| Spec | tinybox red flash | tinybox red kimi | Lucebox Zero 495 |
|---|---|---|---|
| Chip | 32-core AMD EPYC (water cooled) and 6x Radeon AI PRO R9700, held to 225 W each | AMD EPYC Genoa and 4x Radeon AI PRO R9700 | AMD Ryzen AI MAX+ PRO 495 and a Radeon AI PRO R9700 card |
| Memory | 192 GB on the cards, 32 GB DDR4 | 128 GB on the cards, 32 GB DDR5 | 32 GB on the card, plus 128 GB or 192 GB LPDDR5X-8533 |
| Memory speed | 640 GB/s per card, 3,840 GB/s across six | 640 GB/s per card, 2,560 GB/s across four | 640 GB/s on the card, 273 GB/s unified |
| Networking | Not stated | Not stated | 10 GbE, Wi-Fi 7, optional 100 GbE cluster kit |
| Ports and expansion | Not stated | PCIe 5.0, 16 TB of RAID storage | 2× USB4, 2× USB-A 3.2 Gen 2, 2× USB-A 2.0, HDMI 2.1, DisplayPort 1.4, plus 4 display outputs on the GPU |
| Power | 1,700 W power supply | 1,700 W power supply | 1,000 W power supply, SFX form factor |
| Noise | Under 60 dB, per tiny corp | Not stated | Not stated |
| Size and weight | Not stated | Not stated | 340 × 110 × 320 mm, 11.97 L, 6 kg |
| Ships with | Not stated | Not stated | Ubuntu Server with the Lucebox engine and models installed |
| Warranty | Not stated | Not stated | 1 year, parts and labor |
| Sources | tiny corp | tiny corp | Lucebox |
tinybox red flash: six R9700s for $22,000
The red flash is tiny corp’s G4 platform, a 32-core AMD EPYC on PCIe 4 with 32 GB of DDR4, carrying six R9700 cards held to 225 W each. That gives 192 GB of card memory and 3,840 GB/s across the cards, with a 1 TB boot SSD and a 1,700 W power supply.
tinybox red kimi: four R9700s and a 16 TB RAID
The red kimi uses the G5 platform, a 32-core EPYC Genoa on PCIe 5 with DDR5 and two 10 GbE ports, and four R9700 cards: 128 GB of card memory at 2,560 GB/s. Its RAID of sixteen 1 TB drives reads at 238.4 GB/s, and the whole box is $30,000.
tinybox green and green v2: the NVIDIA models
tiny corp’s docs list the tinybox green with six RTX 4090 cards and the green v2 with four RTX 5090s (128 GB of card memory in all), each running on two 1,600 W power supplies. Neither is in the tinygrad.org configurator today, and the docs give no price. For four 5090s’ worth of CUDA now, see what an RTX 5090 build for LLMs costs.
The original tinybox red, and the tinybox pro
The original tinybox red, still on the docs page, had six Radeon RX 7900 XTX cards and two 1,600 W supplies; today’s red flash keeps six AMD cards and uses the R9700. We found no tinybox pro on tinygrad.org or in its docs on 3 November 2026, so this page lists no specs for one.
tinybox base and the 9070XT option
A tinybox base, the G4 platform with no GPUs, is $7,000, and the G5 platform starts at $12,000. Four AMD 9070XT cards (16 GB each, 64 GB in all) add $4,000, four R9700s add $10,000 and six add $15,000.
Which LLMs fit in a tinybox?
A tinybox red flash’s 192 GB of card memory holds every model in this table but one: DeepSeek V4.1 Flash takes about 190 GB on its own, which leaves almost no room for the prompt.
| Model and size | tinybox red kimi128 GB on 4 cards | tinybox red flash192 GB on 6 cards | RTX PRO 6000 workstation96 GB on the card | Lucebox Zero 495, 192 GB32 GB card + 192 GB |
|---|---|---|---|---|
| Qwen3.8-27B15.2 GB | Yes, on the card | Yes, on the card | Yes, on the card | Yes, on the card |
| DeepSeek V4 Flash101.5 GB | Yes, on the card | Yes, on the card | No | Yes, card and unified |
| Qwen3.8-Flash-Next106.3 GB | Yes, on the card | Yes, on the card | No | Yes, card and unified |
| Qwen3.8-27B and DeepSeek V4 Flash at once116.7 GB | No | Yes, on the card | No | Yes, card and unified |
| DeepSeek V4.1 Flash190.8 GB | No | No | No | Yes, card and unified |
tinybox speed: one R9700 card, measured
No one has published a run of these models on a tinybox. These are single cards: one R9700 like the tinybox has, an RTX 5090 and a Strix Halo box.
Model
Measure
R9700 (Lucebox card) writes 2.9x as fast as Strix Halo with a 27-token prompt, same engine and file. The other runs used a different setup. Drafting speeds some prompts far more than others.
tinybox speed test: watch them write
Race
Task
Same engine, file and prompt, drafting on: the Radeon AI PRO R9700 card a Lucebox uses, against a Strix Halo box.
R9700, the Lucebox card0.0 s
writing at 303.38 tok/s
Strix Halo box0.0 s
writing at 104.32 tok/s
tinybox price
The tinybox price on tinygrad.org on 4 November 2026 was $22,000 for the red flash (six R9700s, 192 GB) and $30,000 for the red kimi (four R9700s, 128 GB, 16 TB RAID), each made to order, shipped in 1 to 8 weeks and paid by wire.
| Machine | Memory | Bandwidth, GB/s | Software | US price |
|---|---|---|---|---|
| NVIDIA RTX build | ||||
| RTX 5090 card The card alone; the PC around it costs extra. | 32 GB | 1,792 | CUDA | $7,681 in stock |
| RTX PRO 6000 workstation A complete workstation; the card alone was $17,063 in stock. | 96 GB | 1,792 | CUDA | $21,460 |
| tiny corp tinybox | ||||
| tinybox red flash Made to order, ships in 1 to 8 weeks. | 192 GB on 6 cards | 3,840 | ROCm, Vulkan | $22,000 |
| tinybox red kimi Made to order, ships in 1 to 8 weeks. | 128 GB on 4 cards | 2,560 | ROCm, Vulkan | $30,000 |
| Lucebox | ||||
| Lucebox Zero 495, 128 GB | 32 GB on the GPU + 128 GB | 640 / 273 | ROCm, Vulkan | $5,999 |
| Lucebox Zero 495, 192 GB | 32 GB on the GPU + 192 GB | 640 / 273 | ROCm, Vulkan | $8,199 |
US prices read on 30 October 2026 and 4 November 2026.
tinybox price per GB of card memory
The red flash works out to about $115 per GB of card memory and the red kimi to about $234, because the kimi also pays for the PCIe 5 platform and the 16 TB RAID. An RTX PRO 6000 workstation, at $21,460 for 96 GB, comes to about $224 per GB.
What the price includes
Both presets come with water cooling, a 1,700 W power supply, a 1 TB boot SSD and 32 GB of system RAM. tiny corp builds each tinybox to order and takes payment by wire only.
What else $22,000 buys for local AI
Model
Rank by
Our pick
Lucebox Zero 495, 128 GB
- Holds DeepSeek V4 Flash (101.5 GB) across the 32 GB card and 128 GB of unified memory.
- Writes it at 51.1 tok/s on our earlier 395 build, drafting on (our lab).
- $5,999, $16,001 under your budget.
Also consider
- Holds DeepSeek V4 Flash (101.5 GB) in 128 GB of unified memory.
- Writes it at 41.9 tok/s on another box with the same chip, drafting on (our lab).
- $3,500, $18,500 under your budget.
We make the Lucebox. It is ranked by the same rule as every other row; a run marked “on our earlier 395 build” comes from the previous Lucebox, with the same R9700 card.
- Lucebox Zero 495, 128 GB32 GB card + 128 GBCard + unified 63% used273 GB/s51.1 tok/s drafting on, our lab, on our earlier 395 build
- Lucebox Zero 495, 192 GB32 GB card + 192 GBCard + unified 45% used273 GB/s51.1 tok/s drafting on, our lab, on our earlier 395 build
- GMKtec EVO-X2, 128 GB128 GB unifiedFits 79% used256 GB/s41.9 tok/s drafting on, our lab, on another box with the same chip
- NVIDIA DGX Spark128 GB unifiedFits 79% used273 GB/s35.3 tok/s drafting on, public benchmark
- Mac Studio (M5 Max), 128 GB128 GB unifiedFits 79% used614 GB/s39.35 tok/s no drafting, public benchmark
- tinybox red flash192 GB on 6 cardsOn the card 53% used3,840 GB/sno run yet
- Mac Studio (M5 Ultra), 256 GB256 GB unifiedFits 40% used1,228 GB/sno run yet
- Framework Desktop (Ryzen AI MAX+ PRO 495)192 GB unifiedFits 53% used273 GB/sno run yet
- GMKtec EVO-X2, 64 GB64 GB unifiedDoes not fit needs 159%256 GB/sno run yet
- Mac Studio (M5 Max), 48 GB48 GB unifiedDoes not fit needs 211%614 GB/sno run yet
- GB10 box with 64 GB (Dell, ASUS and others)64 GB unifiedDoes not fit needs 159%273 GB/sno run yet
- Mac Studio (M5 Ultra), 96 GB96 GB unifiedDoes not fit needs 106%1,228 GB/sno run yet
- RTX 5090 card32 GB on the cardDoes not fit needs 317%1,792 GB/sno run yet
- RTX PRO 6000 workstation96 GB on the cardDoes not fit needs 106%1,792 GB/sno run yet
- Mac Studio (M5 Ultra), 512 GB512 GB unified$299 over budgetFits 20% used1,228 GB/sno run yet
- tinybox red kimi128 GB on 4 cards$8,000 over budgetOn the card 79% used2,560 GB/sno run yet
The pick holds the model with room for the prompt, fits your budget and has the fastest run we can link to. Runs with drafting rank against each other first, then runs without it, then machines with no run, by memory speed. The amber band is the Lucebox's 32 GB card.
Who should buy a tinybox?
Buy one if
- You serve several people or fine-tune, and want many cards in one case.
- You want 128 or 192 GB of card memory without NVIDIA prices.
- You are fine with a 1,700 W supply and paying by wire.
Look elsewhere if
- You are one person running one model: a $5,999 Lucebox has one of the same cards.
- You need CUDA: the current tinybox models are all AMD.
- You want a quiet box for a desk.
tinybox or Lucebox: where each fits
We make the Lucebox Zero 495, so read this box as ours.
Where it fits
- One R9700, the same card, with 128 or 192 GB of unified memory beside it, for $5,999 or $8,199.
- DeepSeek V4.1 Flash on one box with 192 GB: 33 tok/s at a 32K prompt.
- Two models loaded together: Qwen3.8-27B on the R9700 and DeepSeek V4 Flash beside it in unified memory.
Where it doesn't
- You serve a team or fine-tune: four or six cards do more at once.
- Your models need more than 32 GB of card memory to run fast.
- You want your machine within weeks: tiny corp quotes 1 to 8 weeks.
From $5,999, delivered from January 2027. See the hardware Configure and order
More local AI hardware comparisons
NVIDIA DGX Spark
AMD Ryzen AI MAX+ boxes
- DGX Spark vs Strix Halo Close on memory speed, about half the price, CUDA only on the Spark.
- Framework Desktop The same chip as the cheaper boxes, in a small case you can build.
- GMKtec EVO-X2 AI mini PC One of the cheapest 128 GB Strix Halo boxes, with 64 GB as an option.
- Minisforum MS-S1 MAX Strix Halo with a PCIe slot, and a 192 GB Gorgon Halo version.
- Beelink GTR9 Pro A 128 GB Strix Halo box with two 10 GbE ports.
- Corsair AI Workstation 300 Strix Halo from a big US brand, in stock.
Apple Mac Studio
tinybox FAQ
How much does a tinybox cost?
On tinygrad.org on 4 November 2026, the tinybox red flash with six R9700 cards was $22,000 and the red kimi with four was $30,000, both shipping in 1 to 8 weeks. A tinybox base with no GPUs is $7,000.
Is a tinybox worth it for running LLMs at home?
A tinybox is worth it at home if several people share it or you fine-tune models: the red flash gives 192 GB of card memory for $22,000. For one person chatting with one model it is a lot of machine, with a 1,700 W power supply, noise that tiny corp rates under 60 dB, and extra cards that add memory more than speed.
Is the tinybox green v2 still sold?
The tinybox green v2 is not in the tinygrad.org configurator on 3 November 2026. tiny corp’s docs still describe it, with four RTX 5090 cards and two 1,600 W power supplies, but give no price.
Which GPUs does the tinybox use now?
tiny corp’s current tinybox presets use AMD Radeon AI PRO R9700 cards with 32 GB each, the card the Lucebox uses, and four AMD 9070XT cards are offered as a cheaper option. The NVIDIA-based green (RTX 4090) and green v2 (RTX 5090) appear only on tiny corp’s docs page.
tinybox or Lucebox?
The tinybox has four or six R9700 cards for teams and training, at $22,000 or $30,000 with the cards. The Lucebox has one R9700 and up to 192 GB of unified memory for one person or a few people sharing it, from $5,999.
Can a tinybox run DeepSeek V4 Flash?
Yes: at 101.5 GB, DeepSeek V4 Flash fits across the cards of both the tinybox red flash (192 GB) and red kimi (128 GB). We have seen no published run of it on a tinybox.