Calculate the TPS of the CMP 90HX on local AI models
Every model in our catalogue assessed against this card at the context length and minimum quality you choose. Speed is an estimate for a single request, calculated from this card's memory bandwidth and the size of each model once compressed.
Calculated for this card
Largest model it holds
Ling-lite-1.5 ("Bailing")
16.8B · Q3_K_M · 51.7 tok/s
Fastest model
Gemma 3 QAT 1B
322 tok/s · 1B
What AI models can a CMP 90HX run?
Set the inputs, read the answer
More context means more memory for the conversation cache. Speed is for a fresh conversation and does not change with this setting.
Hides models that would only fit by being compressed below this point.
381 models match
Calculating| Quantisation | Fit | ||||||
|---|---|---|---|---|---|---|---|
|
322
tok/s
274–386 |
Gemma 3 1B | 1B | Mar 2025 | 1.8 GB | 33k tokens | Q8_0 | Comfortable |
|
322
tok/s
274–386 |
Gemma 3 QAT 1B | 1B | Apr 2025 | 1.8 GB | 33k tokens | Q8_0 | Comfortable |
|
322
tok/s
193–515 · low confidence |
HGRN 1B (WT 103) ≈ | 1B | Nov 2023 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
322
tok/s
193–515 · low confidence |
LLama 3..2 Typhoon 2 1B ≈ | 1B | Dec 2024 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
322
tok/s
193–515 · low confidence |
OLMo-1B ≈ | 1B | Feb 2024 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
322
tok/s
193–515 · low confidence |
Pythia-1b ≈ | 1B | Apr 2023 | 1.8 GB | 131k tokens ? | Q8_0 | Comfortable |
|
302
tok/s
181–483 · low confidence |
DeepSeekMoE-16B | 16B | Jan 2024 | 8.1 GB | 4k tokens | Q3_K_M | Tight |
|
298
tok/s
179–477 · low confidence |
OpenELM-1.1B ≈ | 1.1B | May 2024 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
293
tok/s
176–468 · low confidence |
DeciCoder-1B ≈ | 1.1B | Aug 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
293
tok/s
176–468 · low confidence |
SantaCoder ≈ | 1.1B | Jan 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
293
tok/s
176–468 · low confidence |
TinyLlama-1.1B (1T token checkpoint) ≈ | 1.1B | Oct 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
293
tok/s
176–468 · low confidence |
TinyLlama-1.1B (3T token checkpoint) ≈ | 1.1B | Oct 2023 | 1.9 GB | 131k tokens ? | Q8_0 | Comfortable |
|
268
tok/s
161–429 · low confidence |
EXAONE 4.0 (1.2B) ≈ | 1.2B | Jul 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
268
tok/s
161–429 · low confidence |
MinerU2.5 ≈ | 1.2B | Sep 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
268
tok/s
161–429 · low confidence |
Pleias 1.0 1.2B ≈ | 1.2B | Dec 2024 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
268
tok/s
161–429 · low confidence |
Pleias-RAG-1B ≈ | 1.2B | Apr 2025 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
262
tok/s
223–314 |
Llama 3.2 1B | 1.2B | Sep 2024 | 2.2 GB | 131k tokens | Q8_0 | Comfortable |
|
258
tok/s
155–413 · low confidence |
MiniCPM-1.2B ≈ | 1.2B | Jun 2024 | 2.0 GB | 131k tokens ? | Q8_0 | Comfortable |
|
248
tok/s
149–396 · low confidence |
DeepSeek Coder 1.3B ≈ | 1.3B | Jan 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
248
tok/s
149–396 · low confidence |
DeepSeek-VL-1.3B ≈ | 1.3B | Mar 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
248
tok/s
149–396 · low confidence |
DigiRL ≈ | 1.3B | Jun 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
248
tok/s
149–396 · low confidence |
GLA Transformer 1.3B ≈ | 1.3B | Aug 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
248
tok/s
149–396 · low confidence |
Janus 1.3B ≈ | 1.3B | Oct 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
248
tok/s
149–396 · low confidence |
Kosmos-2.5 ≈ | 1.3B | Aug 2024 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
|
248
tok/s
149–396 · low confidence |
Otter ≈ | 1.3B | May 2023 | 2.1 GB | 131k tokens ? | Q8_0 | Comfortable |
Speeds are estimates for a single request — one conversation at a time — calculated from memory bandwidth, model size and quantisation. Real throughput varies with the inference runtime and its version. Figures published by hardware vendors measure many simultaneous requests and are much higher.
On record
CMP 90HX full specification
Everything on record for this board, ordered by how much it bears on running a language model rather than by how a spec sheet would list it. Memory comes first because it decides the outcome; the rest is context.
Memory
The two specifications that decide what this card can run and how quickly. Capacity sets which models fit; bandwidth sets how many tokens per second they produce once they do.
- Memory size
- 10 GB
- Memory bandwidth
- 760 GB/s
- Memory type
- GDDR6X
- Memory bus width
- 320 bit
- Memory clock
- 1.19 GHz
The chip
Which processor is on the board and how it was manufactured. A smaller process size generally means more performance for the same power.
- Graphics processor
- GA102
- Architecture
- Ampere
- Generation
- Mining GPUs
- Foundry
- Samsung
- Process size
- 8 nm
- Transistors
- 28.3 billion
- Transistor density
- 45,100 K/mm²
- Die size
- 628 mm²
- Package
- BGA-3328
- Released
- 28 July 2021
Clock speeds
How fast the processor runs. Worth far less here than on a gaming benchmark: generating text is limited by memory bandwidth, so a higher clock barely moves the result.
- Base clock
- 1.5 GHz
- Boost clock
- 1.71 GHz
Processing units
What the chip contains. These drive graphics performance and matter mainly for processing a long prompt rather than for producing the answer.
- Shading units
- 6,400
- Texture mapping units
- 200
- Render output units
- 80
- Streaming multiprocessors
- 50
- Tensor cores
- 200
- Ray tracing cores
- 50
- L1 cache
- 128 KB
- L2 cache
- 5 MB
Theoretical performance
Peak arithmetic rates published for the board. These are ceilings that no real workload reaches, and generating text reaches a small fraction of them because it is limited by memory rather than arithmetic.
- Half precision (FP16)
- 21.9 TFLOPS
- Single precision (FP32)
- 21.9 TFLOPS
- Double precision (FP64)
- 342 GFLOPS
- Pixel rate
- 137 GPixel/s
- Texture rate
- 342 GTexel/s
The board
What it takes to physically install and power the card — the practical constraints that decide whether it fits the machine you already own.
- Power draw (TDP)
- 320 W
- Suggested power supply
- 700 W
- Power connectors
- 2x 8-pin
- Bus interface
- PCIe 1.0 x4
- Slot width
- Dual-slot
- Dimensions
- 285 mm
Software support
Which graphics and compute interfaces the card supports. CUDA compute capability is the one that bears on inference: below 7.0 there are no tensor cores, and modern inference software falls back to slower code paths.
- CUDA compute capability
- 8.6
- DirectX
- 12.2
- OpenGL
- 4.6
- Vulkan
- 1.4
- OpenCL
- 3.0
- Shader model
- 6.8
Listings
Where to buy a CMP 90HX
No vendor is currently listing this card. Listings come from vendors who publish them here directly — browse the vendor directory to see who is selling what.
What the numbers mean
Capacity and bandwidth
Memory
10 GB
Bandwidth
760 GB/s
Largest model
Ling-lite-1.5 ("Bailing")
At 10 GB of GDDR6X the CMP 90HX is limited to the smaller end of the catalogue. About 9 GB is actually available to a runtime, and a model has to fit entirely inside it before generating anything at all.
The memory bus moves 760 GB/s across a 320-bit bus. That is the number that governs generation speed — arithmetic per byte read is small enough that the bus, not the cores, is what everything waits on.
The figure is the memory clock — 1.19 GHz here — multiplied by the bus width. It is why core counts predict generation speed so poorly.
In practice that combination tops out at Ling-lite-1.5 ("Bailing") — 16.8B, compressed to Q3_K_M, generating around 51.7 tokens per second.
The chip and how it was built
The CMP 90HX is built on the GA102 graphics processor, using NVIDIA's Ampere architecture, as part of the Mining GPUs generation.
The chip is manufactured by Samsung, on a 8 nm process, with a die measuring 628 mm², holding 28.3 billion transistors. A smaller process generally means more performance for the same power, though for language models it matters far less than the memory subsystem.
It was released in July 2021, roughly 5 years ago. Inference software support tends to follow hardware by a year or two, so a card of this age generally has mature, well-optimised code paths available to it.
Compute throughput, and why it matters less than it looks
FP16
21.9 TFLOPS
FP64
342 GFLOPS
Tensor cores
200
On paper the CMP 90HX reaches 21.9 TFLOPS at half precision and 21.9 TFLOPS at single precision. These are peak figures no real workload sustains, and generating text reaches only a small fraction of them — decoding is limited by memory rather than arithmetic, which is why a card can look enormously powerful here and still produce tokens at an ordinary rate.
Double-precision throughput is 342 GFLOPS. It has no bearing on running a language model — no inference runtime uses it — but it separates datacentre parts from consumer ones, since the latter deliberately restrict it.
The card carries 200 tensor cores across 50 streaming multiprocessors. These accelerate the matrix arithmetic at the heart of a transformer, and they are what make prompt processing — reading a long document before answering — dramatically faster than it would otherwise be.
Clocks run from 1.5 GHz at base to 1.71 GHz boosted. Worth far less here than on a gaming benchmark: raising the clock speeds up the arithmetic, and the arithmetic is not what generation is waiting on.
Cache and processing units
The CMP 90HX has 128 KB of L1 cache, backed by 5 MB of L2. Cache absorbs a share of the memory traffic that would otherwise hit the main bus, which is the one place on this page where a number other than bandwidth quietly affects generation speed — a large L2 lets more of the working set stay close to the cores.
There are 6,400 shading units, 200 texture mapping units, and 80 render output units. These drive graphics workloads and contribute to prompt processing, but they sit idle for much of the time a model spends generating a reply.
Power, size and installation
Power draw
320 W
The CMP 90HX is rated at 320 W, with a 700 W power supply suggested for the whole system. Running a language model keeps a card busy in bursts rather than continuously — it draws hard while generating and idles between requests — so sustained draw over a working day is usually well below the rated figure.
The board occupies a dual-slot, measuring 285 mm long, and needs 2x 8-pin. Worth checking against the case and power supply already in the machine, since the largest cards need considerably more of both than a typical desktop provides.
It connects over PCIe 1.0 x4. The interface governs how quickly a model is loaded from disk into the card, not how fast it runs once there, so a narrower link costs a few seconds at startup and nothing thereafter.
The extremes
The largest AI models a CMP 90HX can run
The biggest open-weight models that fit on this card, newest first. Each is shown at the best compression the card can hold.
The fastest AI models on a CMP 90HX
Where this card produces tokens quickest. Smaller models dominate here, because generating each token means reading the whole model out of memory once.
Step by step
How to work out the tokens per second of a CMP 90HX
You do not have to calculate anything by hand — the gputps.com calculator on this page has already worked it out for every model this card can hold. Reading off the answer takes six steps.
-
01
Find the model in the table
The table lists 381 models this CMP 90HX can run. Search by name, or by size — typing 27b matches on the parameter count even when the name never states it.
-
02
Match the context to your work
Set the context to your real working length. Short questions cost almost nothing; a long document can consume a large share of the card's 10 GB.
-
03
Choose how far you will compress
By default the table picks the least-compressed copy that fits. Setting a floor removes models that only qualify through heavy compression.
-
04
Read the speed and the range
Each speed is an estimate for a single conversation, with a range beneath it — 322 tok/s on Gemma 3 QAT 1B at the top end here. The same card and model vary by thirty to fifty per cent between inference runtimes.
-
05
Read the fit verdict last
The fit column separates models that just fit from those with room to spare — worth checking against the card's 10 GB before settling on one.
-
06
Check the same model from the other side
Following a model through to its own page lists all the hardware that can run it, so you can see where the CMP 90HX sits against the alternatives.
Answers
CMP 90HX — common questions
How much power does a CMP 90HX use?
The CMP 90HX has a rated board power of 320 W, and a 700 W system power supply is suggested. Generating text draws hard in bursts and idles between requests, so average consumption over a working session is normally well below the rated figure.
How much cache does a CMP 90HX have?
The CMP 90HX has 128 KB of L1 cache, and 5 MB of L2 cache. Cache absorbs part of the memory traffic that would otherwise reach the main bus, so a larger L2 gives a modest lift to generation speed beyond what bandwidth alone predicts.
What are the TFLOPS of a CMP 90HX?
The CMP 90HX is rated at 21.9 TFLOPS at half precision and 21.9 TFLOPS at single precision. These are peak arithmetic ceilings rather than achievable rates, and text generation reaches only a small fraction of them because it is limited by memory bandwidth instead.
How many tensor cores does a CMP 90HX have?
The CMP 90HX has 200 tensor cores across 50 streaming multiprocessors. They accelerate the matrix arithmetic a transformer is built from, which mainly speeds up processing a long prompt rather than producing the reply.
Does the CMP 90HX support CUDA?
Yes. The CMP 90HX reports CUDA compute capability 8.6. Capability 7.0 and above has tensor cores, which modern inference software uses; below that it falls back to slower code paths for quantised models.
What bus interface does the CMP 90HX use?
It uses PCIe 1.0 x4. This governs how fast a model is loaded onto the card rather than how fast it runs once loaded, so it costs a few seconds at startup and nothing during generation.
Is the CMP 90HX good for running local AI models?
Its memory limits it to smaller models and its bandwidth gives usable, if unspectacular, generation speeds. In total it runs 381 of the models we track. Whether that is enough depends entirely on which model you want — the table above answers that directly.
Can a CMP 90HX run a model that does not fit in its memory?
It can be split, with the overflow held in system memory — but that part drags the whole thing down, and none of the 10 GB figures on this page assume it.
Would two CMP 90HX cards be twice as fast?
No. A second CMP 90HX doubles the memory to 20 GB, which lets you hold models neither could hold alone, but generation does not split that way. These figures describe one card.
What AI models can a CMP 90HX run?
381 of the 679 open-weight language models we track fit on a CMP 90HX and can be run locally on it. The table on this page lists every one, with the memory it needs, the quantisation it runs at and an estimated generation speed.
What is the largest AI model a CMP 90HX can run?
The largest model in our catalogue that fits on a CMP 90HX is Ling-lite-1.5 ("Bailing") at 16.8B parameters, compressed to Q3_K_M. It generates roughly 51.7 tokens per second and needs about 8.9 GB of the card's memory.
How many tokens per second does a CMP 90HX produce?
It depends on the model. On a CMP 90HX the fastest model we track is Gemma 3 QAT 1B at about 322 tokens per second, while larger models run proportionally slower because each token requires reading the whole model out of memory once. Speeds are estimates for a single conversation at a time.
Can a CMP 90HX run a 7B model?
Yes. For example a CMP 90HX runs DeepSeek Coder 6.7B at Q4_K_M, using about 8.3 GB of memory and generating around 111 tokens per second.
Can a CMP 90HX run a 13B model?
Yes. For example a CMP 90HX runs DeepSeekMoE-16B at Q3_K_M, using about 8.1 GB of memory and generating around 302 tokens per second.
How much memory does a CMP 90HX have?
A CMP 90HX has 10 GB of GDDR6X memory. Around a tenth of that is reserved by the inference runtime and the driver, leaving roughly 9 GB available for a model and its conversation.
What is the memory bandwidth of a CMP 90HX?
The CMP 90HX has 760 GB/s of memory bandwidth, across a 320-bit memory bus. This is the single best predictor of how fast it generates text, because producing each token means reading the entire model out of memory once.
What type of memory does a CMP 90HX use?
It uses GDDR6X clocked at 1.19 GHz. HBM types are found on datacentre accelerators and carry far more bandwidth than the GDDR used on desktop cards, which is why they generate tokens considerably faster at the same capacity.
Who makes the CMP 90HX?
The CMP 90HX is a NVIDIA product, with the chip manufactured by Samsung, on a 8 nm process.
When was the CMP 90HX released?
The CMP 90HX was released in July 2021.
The other direction
Looking at it from the other side?
This page starts from the hardware. If you already know which model you want and need to know what it takes to run it, start from the model instead.