In this article
Short answer
This AI PC build under 3 lakh centres on an RTX 5060 Ti 16GB with a Ryzen 5 9600X, 32 GB of DDR5 and a 2 TB NVMe drive. It keeps 16 GB of VRAM, the figure that decides which models load, and spends modestly elsewhere. It suits learners and hobbyists running local models and image generation.
- 16 GB of VRAM runs 14B-class models at 4-bit, which an 8 GB card cannot do without slowing sharply.
- A 650 W supply covers the card's single 8-pin connector and the 550 W that ASUS recommends.
- The mATX board has two memory slots, so 32 GB now and a 64 GB replacement kit later.
- The budget band is a ceiling; the parts list sits comfortably inside it.
- CPUAMD Ryzen 5 9600X
- GPUASUS Dual GeForce RTX 5060 Ti 16GB GDDR7 OC
The parts list#
| Component | Part | Why |
|---|---|---|
| CPU | AMD Ryzen 5 9600X | 6 cores and 12 threads at 65 W, with PCIe 5.0 and DDR5 support |
| Motherboard | MSI B650M Gaming WiFi mATX Motherboard | PCIe 4.0 x16 slot from the CPU, two M.2 slots, Wi-Fi 6E |
| GPU | ASUS Dual GeForce RTX 5060 Ti 16GB GDDR7 OC | 16 GB of GDDR7, one 8-pin connector, 229 mm long |
| RAM | Patriot Viper Venom 32GB (2x16GB) DDR5-6000 CL30 Kit | 32 GB at 6000 MT/s with EXPO and XMP support |
| Storage | WD_BLACK SN850X 2TB PCIe 4.0 NVMe M.2 SSD | 7300 MB/s reads and 1200 TBW endurance |
| PSU | Cooler Master MWE Bronze 650 V3 650W ATX 3.1 Power Supply | Four 8-pin PCIe cables, above the card's recommended wattage |
| CPU cooler | Thermalright Assassin X 120 Refined SE Air CPU Cooler | 148 mm single tower with AM5 support |
| Case | Corsair Frame 4000D RS ARGB Black Mid-Tower Case | Three pre-installed fans and room for 430 mm cards |
Why these parts#
CPU
The Ryzen 5 9600X (opens in a new tab) has 6 cores and 12 threads, a 5.4 GHz maximum boost clock and a 65 W default TDP, according to AMD. Six fast cores are enough to keep one mid-range GPU busy in training, inference and image generation. The graphics card does the heavy work.
AMD lists PCIe 5.0 with 24 usable lanes and DDR5 memory with a 256 GB maximum. It does not bundle a cooler with this chip, so the build adds one. The AM5 socket also gives a clear upgrade path: the same board accepts the Ryzen 7 9700X used in our deep learning build if you want eight cores later. For more options at this tier, see the best budget CPU for an AI PC build.
Motherboard
The MSI B650M Gaming WiFi (opens in a new tab) is a micro-ATX board on AMD's B650 chipset. MSI lists support for Ryzen 9000, 8000 and 7000 series processors, so confirm the board's BIOS revision with the seller if you buy it for a Ryzen 9000 chip. It has two DDR5 slots with a 128 GB maximum, one PCIe 4.0 x16 slot from the CPU, two PCIe 4.0 x4 M.2 slots and four SATA ports.
The slot is PCIe 4.0, not 5.0, and that is fine here. PNY lists the RTX 5060 Ti as an x8 card, and ComputingForGeeks notes that once weights sit in VRAM the link is nearly idle for language-model inference. MSI also lists 2.5 Gb Ethernet and Wi-Fi 6E.
GPU
The ASUS Dual GeForce RTX 5060 Ti 16GB OC Edition (opens in a new tab) pairs 16 GB of GDDR7 with a 128-bit bus. NVIDIA gives the chip 4,608 CUDA cores and a 180 W graphics power rating. ASUS lists the card at 229 mm long and 2.5 slots wide, with one 8-pin power connector and a 550 W recommended supply.
Bandwidth on this card is 448 GB/s, according to ComputingForGeeks, which is half of the RTX 5070 Ti. That is the trade-off at this tier. The memory size is the same 16 GB, so the same models load, but they generate tokens more slowly. ComputingForGeeks also notes that running a 14B model on an 8 GB card needs CPU offloading and cuts throughput by 5 to 10 times, which is why this build does not drop to the 8 GB version.
RAM
The Patriot Viper Venom (opens in a new tab) kit is 32 GB as two 16 GB modules, rated at 6000 MT/s with CL30 timings at 1.35 V. Patriot states XMP 3.0 and AMD EXPO support for the line and asks buyers to confirm EXPO on specific models with its technical support. Check that before you enable a profile on AM5.
AMD lists DDR5-5600 as the top official speed with two modules on the 9600X, so the kit's 6000 profile runs as an overclock setting. 32 GB covers a language model's working memory and a ComfyUI session. ComputingForGeeks suggests at least 32 GB for FLUX-class image models, so this is the floor, not the ceiling. Our guide to how much RAM for a local LLM explains when to go higher.
Storage
The WD_BLACK SN850X 2TB (opens in a new tab) is a PCIe 4.0 x4 NVMe drive. Western Digital lists 7300 MB/s sequential reads, 6600 MB/s writes and 1200 TBW of endurance with a five-year warranty. The drive is sold bare at 80 x 22 x 2.38 mm, and a heatsink version is sold separately, so use the heatsink on the motherboard's M.2 socket.
A 2 TB drive is a deliberate choice. Model files are large, and ComputingForGeeks lists a 17.2 GB FLUX fp8 checkpoint, so a 1 TB drive fills quickly once you hold several models side by side.
Power supply
ASUS recommends a 550 W power supply for the card and a single 8-pin connector, and NVIDIA's own figure for the chip is a 600 W system. The Cooler Master MWE Bronze 650 V3 (opens in a new tab) is a 650 W ATX 3.1 unit with 80 Plus Bronze efficiency and a five-year warranty. It is non-modular.
Cooler Master lists four PCIe 6+2-pin connectors and two EPS connectors, and does not list a 12V-2x6 cable, which is not needed for this card. The unit is 140 mm long and uses a 120 mm fan. If you plan to move to a higher-power card, check that card's connector and recommended wattage first, since this supply has no 16-pin cable.
CPU cooler
The Thermalright Assassin X 120 Refined SE (opens in a new tab) is a single-tower cooler with four 6 mm heat pipes and one 120 mm PWM fan rated up to 1550 RPM. Thermalright lists AM5 and LGA1851 support. It stands 148 mm tall and is only 46 mm wide, so it leaves space around the memory slots.
Thermalright does not publish a TDP rating. A 65 W processor is a modest load for this size of cooler. The Corsair case below allows 170 mm of cooler height, so there is a margin.
Case
The Corsair Frame 4000D RS ARGB (opens in a new tab) supports ATX, micro-ATX and mini-ITX boards, and Corsair lists room for 430 mm graphics cards, 170 mm coolers and 220 mm power supplies. This build's card is 229 mm, the cooler 148 mm and the power supply 140 mm, so nothing is close to the limit.
Corsair ships it with three ARGB fans pre-installed and a steel front built for airflow. It has seven horizontal expansion slots. A micro-ATX board in an ATX case leaves empty space, which suits a first build and gives room for fingers, cables and later swaps.
Expected performance#
ComputingForGeeks ran Ollama on both the RTX 5070 Ti and the RTX 5060 Ti 16GB, using Llama 3.1 8B and Qwen2.5 14B at Q4_K_M. The figures below are for the GPU chip rather than this board partner's card.
Ollama token generation, Llama 3.1 8B Q4_K_M: token generation (tokens/s, higher is better)
2 products. Highest: ASUS Prime GeForce RTX 5070 Ti 16GB GDDR7 OC, 125 tokens/s. Lowest: ASUS Dual GeForce RTX 5060 Ti 16GB GDDR7 OC, 75.1 tokens/s.
| Product | tokens/s |
|---|---|
| ASUS Prime GeForce RTX 5070 Ti 16GB GDDR7 OC | 125 |
| ASUS Dual GeForce RTX 5060 Ti 16GB GDDR7 OC | 75.1 |
About this data: figures are copied from the named source, not measured by us, and the test system and settings are those the source states above. See all benchmark sources and credits.
On Llama 3.1 8B the 5060 Ti generated about 75 tokens per second, against about 125 for the 5070 Ti. That is comfortable for chat and for a coding assistant.
Ollama token generation, Qwen2.5 14B Q4_K_M: token generation (tokens/s, higher is better)
2 products. Highest: ASUS Prime GeForce RTX 5070 Ti 16GB GDDR7 OC, 73.2 tokens/s. Lowest: ASUS Dual GeForce RTX 5060 Ti 16GB GDDR7 OC, 42.3 tokens/s.
| Product | tokens/s |
|---|---|
| ASUS Prime GeForce RTX 5070 Ti 16GB GDDR7 OC | 73.2 |
| ASUS Dual GeForce RTX 5060 Ti 16GB GDDR7 OC | 42.3 |
About this data: figures are copied from the named source, not measured by us, and the test system and settings are those the source states above. See all benchmark sources and credits.
On the larger Qwen2.5 14B the 5060 Ti managed about 42 tokens per second, against about 73 for the 5070 Ti. ComputingForGeeks measured the 14B model at 4-bit using about 9.0 GB, so it sits well inside 16 GB. A 32B model at 4-bit needs about 20 GB and does not fit.
Image generation is slower but works. In ComfyUI, ComputingForGeeks measured an SDXL image at 1024 x 1024 and 30 steps in about 11.3 seconds on the 5060 Ti, against about 4.8 seconds on the RTX 5090.
ComfyUI SDXL 1024x1024, 30 steps: time per image (seconds, lower is better)
2 products. Lowest: INNO3D GeForce RTX 5090 X3 OC 32GB, 4.84 seconds. Highest: ASUS Dual GeForce RTX 5060 Ti 16GB GDDR7 OC, 11.34 seconds.
| Product | seconds |
|---|---|
| INNO3D GeForce RTX 5090 X3 OC 32GB | 4.84 |
| ASUS Dual GeForce RTX 5060 Ti 16GB GDDR7 OC | 11.34 |
About this data: figures are copied from the named source, not measured by us, and the test system and settings are those the source states above. See all benchmark sources and credits.
FLUX.1-dev at fp8 took about 37 seconds per image on the same card in that test, against about 8.8 seconds on the RTX 5090. ComputingForGeeks adds that FLUX at fp8 peaked near 15.4 GiB on a 16 GB card, so it runs but with little spare memory.
What does this mean in daily use? For text work, the card is quick enough that waiting is rarely the issue. A reply from an 8B model arrives faster than most people read. A 14B model is slower but still interactive. For image work, a batch of SDXL images is a task you start and return to, not one you watch. The card's real limit is not speed but the 16 GB ceiling. When a model or a video workflow needs more than that, no amount of patience helps, and the next step is a card with more memory.
Memory bandwidth explains most of the gap to the 5070 Ti. Puget Systems states that token generation depends more on memory bandwidth, while prompt processing follows tensor-core throughput. The 5060 Ti has half the bandwidth of the 5070 Ti, and ComputingForGeeks measured the faster card at 1.66 to 1.85 times the speed in Ollama, close to that ratio.
Upgrade path#
The cheapest upgrade is memory. The board has two slots and a 128 GB maximum, so a 64 GB kit replaces the 32 GB one. Our best RAM for AI and machine learning guide lists kits by capacity.
The graphics card is the next step. A move to an RTX 5070 Ti brings 750 W recommended by ASUS and a 16-pin connector, so the power supply would change as well. The case and cooler already take the larger card. The deep learning PC build shows a platform sized for it.
The processor can move up within AM5, for example to the Ryzen 7 9700X. The board's PCIe slot would then be the main limit, since it is PCIe 4.0.
Alternatives by budget band#
If you need more VRAM, the local LLM PC build goes up to a 32 GB card. If image generation is the main job, the Stable Diffusion PC build moves to a faster card with a larger case and 64 GB of memory.
Assembly and setup notes#
These notes come from the makers' own documentation. Read each manual for the steps.
- BIOS: MSI lists Ryzen 9000 support for the board. Check the revision on the box and update the BIOS if the manual says a newer one is needed.
- Memory: Patriot advises confirming EXPO support on the specific model with its technical support, so enable the profile in the BIOS and test stability.
- Graphics card: ASUS lists one 8-pin connector. Cooler Master lists four PCIe 6+2-pin cables on the power supply.
- SSD: Western Digital sells the drive bare, so fit the board's M.2 heatsink.
- Cooler: Thermalright supplies the mounting hardware for AM5. Follow its manual for the bracket.
Verdict#
This is the build to choose when the budget is firm but you still want 16 GB of VRAM. The RTX 5060 Ti 16GB loads the same models as faster cards, the Ryzen 5 9600X is enough to feed it, and the supporting parts have margin under every maker limit we checked.
Pick a different build if you need speed on 14B models, in which case the RTX 5070 Ti build is better, or if you plan two cards.
For a build in this budget, buy from Indian retailers that issue a GST invoice, because it supports warranty claims and keeps every part's guarantee valid with the maker's local service network.
Prices in India shift by city and season, so the 3 lakh budget works as a target to check against current Indian retailer listings, not a fixed total.
Frequently asked questions#
Is 16 GB of VRAM worth it at this tier?
Yes. ComputingForGeeks found a 14B model needs CPU offloading on an 8 GB card, which cuts throughput by 5 to 10 times. With 16 GB the same model sits entirely in VRAM.
Can this build run 32B models?
Not fully in VRAM. A 32B model at 4-bit needs about 20 GB, according to ComputingForGeeks. It would need layers placed in system memory, which is much slower.
Is 32 GB of RAM enough?
For a single user, yes. ComputingForGeeks suggests at least 32 GB for FLUX-class image models. If you work with large datasets, replace the kit with 64 GB.
Why a micro-ATX board in an ATX case?
The MSI board suits the budget and has the slots this build needs. The case supports micro-ATX, so it gives room to work in and a straightforward move to a larger board later.
Does PCIe 4.0 slow the card?
PNY lists the RTX 5060 Ti as a PCIe x8 card, and ComputingForGeeks says the link makes little difference to language-model inference once the weights are in VRAM.
Can I use an AMD graphics card instead?
You can, but most AI frameworks target NVIDIA's CUDA first, so some tools need extra steps on AMD. Our guide to the best 16 GB GPU for AI covers the options and the trade-offs.
Sources & credits#
We have not tested these products ourselves. Specs come from the makers; benchmark figures come from the sources we name.
- AMD, Ryzen 5 9600X product page (opens in a new tab)
- MSI, B650M Gaming WiFi specifications (opens in a new tab)
- ASUS India, Dual GeForce RTX 5060 Ti 16GB OC Edition (opens in a new tab)
- NVIDIA India, GeForce RTX 50 series, RTX 5060 family (opens in a new tab)
- PNY, RTX 5060 Ti 16GB Dual Fan brochure (opens in a new tab)
- Patriot Memory, Viper Venom DDR5 (opens in a new tab)
- Western Digital India, WD_BLACK SN850X (opens in a new tab)
- Cooler Master, MWE Bronze 650 V3 (opens in a new tab)
- Thermalright, Assassin X 120 Refined SE (opens in a new tab)
- Corsair, Frame 4000 series cases (opens in a new tab)
- ComputingForGeeks, RTX 5060 Ti vs 5070 Ti: Local AI Benchmarks (opens in a new tab)
- ComputingForGeeks, Best GPU for ComfyUI & Stable Diffusion by VRAM (opens in a new tab)
Facts and figures in this article come from these sources. Articles are compiled from the makers' published specifications and the benchmarks named above, and checked by the NextAIGrid Editorial Team.
Disclaimer
This page is for information only. Specifications and availability change, so check the maker's page before you buy. Found an error? Tell us.
How was this pc build put together?
It draws on 10 named sources: maker specification pages and published third-party benchmarks, listed in the article. We have not tested these products ourselves. Specs come from the makers; benchmark figures come from the sources we name. Specifications and sources were last checked on 17 November 2026. Warranty terms and availability differ by brand and seller in India, so check the maker's India page for the warranty period and buy from an authorised seller. Read how we research and our editorial policy.
Spotted an error? Tell us. Confirmed fixes are listed on the corrections page.