Dual RTX 4090, 64 GB RAM
Like the dual 3090 PC with 64 GB: a little faster, and a lot more expensive.
Price
$5,910€5,470
Memory
48 GB graphics memory
2 cards × 24 GB
Plus 64 GB of normal memory (RAM)
Good for
Up to Llama 3.3 70B
The largest well-known model it runs well: a large model, at standard quality.
Comfortable~14 words/sAlso listed: Dual RTX 4090, 128 GB RAM, for about $7,631€7,330. Same cards, so models that fit in them run just as fast here. Its extra normal memory only helps with models too big for the cards, and those run slowly.
Check the graphics card price
This PC uses 2 of these cards. The other parts are sold separately. Our price counts used graphics cards. These shops mostly sell new ones, so their price can be higher.
EU shops · In the US? Switch the currency to USD.
US shops · In Europe? Switch the currency to EUR.
Waiting for prices to drop?
Which models run on it
Each AI model needs a certain amount of memory. This is what fits at standard quality with a normal-length conversation (about 12 pages of text), and how fast the answers would come.
| AI model | Runs? | Speed |
|---|---|---|
| Llama 3.2 3Bsmall | Runs | Fast~213 words/s |
| Mistral 7B v0.3small | Runs | Fast~116 words/s |
| Llama 3.1 8Bsmall | Runs | Fast~107 words/s |
| Gemma 2 9Bsmall | Runs | Fast~83 words/s |
| Qwen2.5 14Bmid-size | Runs | Fast~61 words/s |
| Phi-4 14Bmid-size | Runs | Fast~61 words/s |
| gpt-oss-20bmid-size | Runs | Not estimated |
| Mistral Small 3.2 24Bmid-size | Runs | Fast~41 words/s |
| OrcaCyber Nano 1.0mid-size | Runs | Fast~40 words/s |
| Gemma 2 27Bmid-size | Runs | Fast~34 words/s |
| Gemma 3 27Bmid-size | Runs | Fast~36 words/s |
| Qwen3.8-27Bmid-size | Runs | Fast~38 words/s |
| Muse Glimmer 30Bmid-size | Runs | Fast~38 words/s |
| Qwen3 Coder 30B A3Bmid-size | Runs | Not estimated |
| Gemma 4 31Bmid-size | Runs | Fast~32 words/s |
| Qwen3 32Bmid-size | Runs | Fast~29 words/s |
| Qwen2.5 32Bmid-size | Runs | Fast~29 words/s |
| Qwen3.6-35B-A3Bmid-size | Runs | Not estimated |
| Llama 3.3 70Blarge | Runs | Comfortable~14 words/s |
| DeepSeek-R1-Distill 70Blarge | Runs | Comfortable~14 words/s |
| Qwen2.5 72Blarge | Runs slowly | Slow~4.8 words/s |
| NVIDIA Nemotron 3.5 Bolt 74B-A9Blarge | Runs | Not estimated |
| Kolibri-1large | Runs slowly | Not estimated |
| Llama 4 Scoutvery large | Runs slowly | Not estimated |
| gpt-oss-120bvery large | Runs slowly | Not estimated |
| NVIDIA Nemotron 3.5 Super VLvery large | Runs slowly | Not estimated |
| Ornith-1.5-122Bvery large | Runs slowly | Not estimated |
| Qwen3.8-Flash-Nextvery large | Doesn't fit | – |
| Reflection Beam 1.0very large | Doesn't fit | – |
| DeepSeek V4.1 Flash 1017very large | Doesn't fit | – |
| Meta Muse Spark Openvery large | Doesn't fit | – |
Not estimated: these models use only part of themselves for each word. Our speed estimate can't model that yet, so we leave it out instead of guessing.
Other quality levels
Higher quality needs more memory. Standard quality is what most people run at home.
- Q4 · standard (smallest that keeps quality)
- Q5 · a bit sharper, a bit bigger
- Q8 · near full quality, almost twice Q4
- FP16 · full quality, over three times Q4
Swipe the table sideways to see every quality level.
| AI model | Q4 | Q5 | Q8 | FP16 |
|---|---|---|---|---|
| Llama 3.2 3B | Runs | Runs | Runs | Runs |
| Mistral 7B v0.3 | Runs | Runs | Runs | Runs |
| Llama 3.1 8B | Runs | Runs | Runs | Runs |
| Gemma 2 9B | Runs | Runs | Runs | Runs |
| Qwen2.5 14B | Runs | Runs | Runs | Runs |
| Phi-4 14B | Runs | Runs | Runs | Runs |
| gpt-oss-20b | Runs | Runs | Runs | Runs |
| Mistral Small 3.2 24B | Runs | Runs | Runs | Runs slowly |
| OrcaCyber Nano 1.0 | Runs | Runs | Runs | Runs slowly |
| Gemma 2 27B | Runs | Runs | Runs | Runs slowly |
| Gemma 3 27B | Runs | Runs | Runs | Runs slowly |
| Qwen3.8-27B | Runs | Runs | Runs | Runs slowly |
| Muse Glimmer 30B | Runs | Runs | Runs | Runs slowly |
| Qwen3 Coder 30B A3B | Runs | Runs | Runs | Runs slowly |
| Gemma 4 31B | Runs | Runs | Runs | Runs slowly |
| Qwen3 32B | Runs | Runs | Runs | Runs slowly |
| Qwen2.5 32B | Runs | Runs | Runs | Runs slowly |
| Qwen3.6-35B-A3B | Runs | Runs | Runs | Runs slowly |
| Llama 3.3 70B | Runs | Runs slowly | Runs slowly | Doesn't fit |
| DeepSeek-R1-Distill 70B | Runs | Runs slowly | Runs slowly | Doesn't fit |
| Qwen2.5 72B | Runs slowly | Runs slowly | Runs slowly | Doesn't fit |
| NVIDIA Nemotron 3.5 Bolt 74B-A9B | Runs | Runs slowly | Runs slowly | Doesn't fit |
| Kolibri-1 | Runs slowly | Runs slowly | Runs slowly | Doesn't fit |
| Llama 4 Scout | Runs slowly | Runs slowly | Doesn't fit | Doesn't fit |
| gpt-oss-120b | Runs slowly | Runs slowly | Runs slowly | Runs slowly |
| NVIDIA Nemotron 3.5 Super VL | Runs slowly | Runs slowly | Doesn't fit | Doesn't fit |
| Ornith-1.5-122B | Runs slowly | Runs slowly | Doesn't fit | Doesn't fit |
| Qwen3.8-Flash-Next | Doesn't fit | Doesn't fit | Doesn't fit | Doesn't fit |
| Reflection Beam 1.0 | Doesn't fit | Doesn't fit | Doesn't fit | Doesn't fit |
| DeepSeek V4.1 Flash 1017 | Doesn't fit | Doesn't fit | Doesn't fit | Doesn't fit |
| Meta Muse Spark Open | Doesn't fit | Doesn't fit | Doesn't fit | Doesn't fit |
Speeds and memory are our estimate from the hardware specs, not a measured test. Prices are dated and may change. How we calculate
Price details and parts
- Graphics cards (2×)$4,892€4,370
- Memory, 64 GB DDR4 (estimate)$555€600
- Rest of the PC: processor, board, case, power supply, storage (estimate)$463€500
The card is the GeForce RTX 4090 24GB. Used price, the typical price of second-hand cards that actually sold. Source: pcprice.watch, checked 16 Oct 2026.
The total is our estimate from these parts, not the price of a ready-made PC in a shop.
DDR5 was not separately priced, so this total uses the price we found for 64 GB DDR4 memory.
Suggested parts
- 2× GeForce RTX 4090 24GB, used
- Ryzen 9 7950X or similar
- 1600 W power supply
Prices exclude shipping. Euro prices include VAT. Dollar prices are converted from our euro prices without EU VAT and before US sales tax, so they are an estimate.
Technical specs
- Graphics memory
- 48 GB graphics memory (2 cards × 24 GB)
- Usable for AI models (our estimate)
- About 46.8 GB. A little is kept free for the graphics driver.
- Normal memory (RAM)
- 64 GB. Our estimate only counts it when a model is too big for the graphics memory: that part then runs from normal memory, slowly. We keep 8 GB of it free for the system.
- Memory speed (bandwidth)
- 2,016 GB/s. How fast the chip can read the model; higher means faster answers. From the maker's spec sheet.
- Graphics card power (maximum)
- 900 W.
- In short
- Same memory as two 3090s, faster memory, a lot more money. 64 GB of normal memory instead of 128 GB keeps the price down.
Common questions
- Can the Dual RTX 4090, 64 GB RAM run Llama 3.3 70B?
- Yes, by our estimate, at about 14 words per second (Comfortable). At standard quality and with a normal-length conversation, Llama 3.3 70B needs about 46 GB of memory, and the Dual RTX 4090, 64 GB RAM has about 46.8 GB it can use.
- Which AI models run fully on the Dual RTX 4090, 64 GB RAM?
- By our estimate, 21 of the 31 models we track run fully in its memory at standard quality: Llama 3.2 3B, Mistral 7B v0.3, Llama 3.1 8B, Gemma 2 9B, Qwen2.5 14B, Phi-4 14B, gpt-oss-20b, Mistral Small 3.2 24B, OrcaCyber Nano 1.0, Gemma 2 27B, Gemma 3 27B, Qwen3.8-27B, Muse Glimmer 30B, Qwen3 Coder 30B A3B, Gemma 4 31B, Qwen3 32B, Qwen2.5 32B, Qwen3.6-35B-A3B, Llama 3.3 70B, DeepSeek-R1-Distill 70B, NVIDIA Nemotron 3.5 Bolt 74B-A9B.
- How fast is the Dual RTX 4090, 64 GB RAM with AI models?
- With Llama 3.3 70B, the largest well-known model it runs well, about 14 words per second at standard quality (our estimate, not a measured test). Comfortable: About as fast as you read. Fine for chat.