CRYSTALCODER TPS calculator

Open weights Mohamed bin Zayed University of Artificial Intelligence (MBZUAI),Petuum,University of Southern California,Carnegie Mellon University (CMU),University of Illinois Urbana-Champaign (UIUC),University of California San Diego,LLM360 6.7B parameters December 2023

Each card below is assessed against this model at the context length and minimum quality you choose. Speed is an estimate for a single request, calculated from the card's memory bandwidth and the size of the model once compressed.

Calculated for this model

589 cards that can run it

818 cards we hold specifications for

Smallest card that fits

Tesla K20c

5 GB · IQ4_XS · 27.4 tok/s

Fastest card

B200

506 tok/s · 180 GB

Which GPUs can run CRYSTALCODER?

Set the inputs, read the answer

Context length tokens

A longer conversation needs more memory, which can push this model off smaller cards.

Minimum quality

Hides cards that would only fit the model by compressing it below this point.

Search cards

Card memory

Also show cards that cannot run it

589 cards match

Calculating

					Needs	Quantisation	Fit
506 tok/s 303–809 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	7.9 GB	Q8_0	Comfortable
506 tok/s 303–809 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	7.9 GB	Q8_0	Comfortable
404 tok/s 242–646 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	7.9 GB	Q8_0	Comfortable
404 tok/s 242–646 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	7.9 GB	Q8_0	Comfortable
323 tok/s 194–517 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	7.9 GB	Q8_0	Comfortable
309 tok/s 185–495 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	7.9 GB	Q8_0	Comfortable
309 tok/s 185–495 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	7.9 GB	Q8_0	Comfortable
296 tok/s 178–473 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	7.9 GB	Q8_0	Comfortable
263 tok/s 158–420 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	7.9 GB	Q8_0	Comfortable
263 tok/s 158–420 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	7.9 GB	Q8_0	Comfortable
263 tok/s 158–420 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	7.9 GB	Q8_0	Comfortable
249 tok/s 149–399 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	7.9 GB	Q8_0	Comfortable
212 tok/s 127–340 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	7.9 GB	Q8_0	Comfortable
212 tok/s 127–340 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	7.9 GB	Q8_0	Comfortable
212 tok/s 127–340 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	7.9 GB	Q8_0	Comfortable
212 tok/s 127–340 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	7.9 GB	Q8_0	Comfortable
212 tok/s 127–340 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	7.9 GB	Q8_0	Comfortable
162 tok/s 97–259 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	7.9 GB	Q8_0	Comfortable
162 tok/s 97–259 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	7.9 GB	Q8_0	Comfortable
137 tok/s 82–219 · low confidence	CMP 170HX 8 GB NVIDIA	8 GB	1,490 GB/s	Sep 2021	6.3 GB	Q6_K	Tight
135 tok/s 81–216 · low confidence	Data Center GPU Max 1550 Intel	128 GB	3,280 GB/s	Jan 2023	7.9 GB	Q8_0	Comfortable
132 tok/s 79–211 · low confidence	Data Center GPU Max Subsystem Intel	128 GB	3,210 GB/s	Jan 2023	7.9 GB	Q8_0	Comfortable
129 tok/s 77–206 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	7.9 GB	Q8_0	Comfortable
129 tok/s 77–206 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	7.9 GB	Q8_0	Comfortable
129 tok/s 77–206 · low confidence	A800 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Aug 2022	7.9 GB	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

Full specification

Everything on record for this model. Most of it describes how it was trained rather than how it runs — useful context for judging how much work went into it, and how it compares with models built at a different scale.

Origin

Who built this model, where, and when it was published.

Organisation: Mohamed bin Zayed University of Artificial Intelligence (MBZUAI),Petuum,University of Southern California,Carnegie Mellon University (CMU),University of Illinois Urbana-Champaign (UIUC),University of California San Diego,LLM360
Organisation type: Academia,Industry,Academia,Academia,Academia,Academia,Research collective
Country: United Arab Emirates, United States of America
Published: 11 December 2023
Authors: Zhengzhong Liu, Aurick Qiao, Willie Neiswanger, Hongyi Wang, Bowen Tan, Tianhua Tao, Junbo Li, Yuqi Wang, Suqi Sun, Omkar Pangarkar, Richard Fan, Yi Gu, Victor Miller, Yonghao Zhuang, Guowei He, Haonan Li, Fajri Koto, Liping Tang, Nikhil Ranjan, Zhiqiang Shen, Xuguang Ren, Roberto Iriondo, Cun Mu, Zhiting Hu, Mark Schulze, Preslav Nakov, Tim Baldwin, Eric P. Xing

What it does

The problem areas the model was built for. A model can carry several of each.

Domain: Language
Task: Language modeling/generation, Question answering, Code generation

Size

How large the model is and how much data it was trained on. Parameters are the figure that decides whether it fits on a given graphics card.

Parameters: 6.7B
Training data: 1,382,000,000,000 tokens

Training compute

The arithmetic performed to train the model, measured in floating-point operations. It is a measure of what the training run cost, not of how fast the finished model answers you.

Training compute: 5.6 × 10²² FLOP
How it was established: Operation counting

The training run

What it physically took to train: which chips, how many, for how long, and what that drew from the wall.

Cloud vendor: Cerebras
Data centre: CRYSTALCODER is trained on the Cerebras Condor Galaxy 1 (CG-1), a 4 exaFLOPS, 54 million core, 64-node cloud AI supercomputer

Availability

Whether you can obtain the model and run it on your own hardware, which is what decides if any of the graphics-card figures on this page apply.

Weights: Open — downloadable
Model access: Open weights (unrestricted)
Training code: Open source
Hugging Face: LLM360

How it is classified

Labels the source dataset applies when tracking notable models, and how confident it is in the entry.

Record confidence: Confident

Sources

Where this record came from and when it was last checked.

Reference: LLM360: Towards Fully Transparent Open-Source LLMs
Last updated: 28 November 2025

The extremes

The ten fastest GPUs that run CRYSTALCODER

Ranked by estimated tokens per second, newest card first where speeds tie. Because generation is bound by memory bandwidth, this ordering follows bandwidth rather than any gaming benchmark.

The smallest GPUs that still run CRYSTALCODER

The cheapest route in, by memory capacity. A tight fit runs the model but leaves nothing spare for a longer conversation.

What the numbers mean

What it takes to run this model

Minimum card

Tesla K20c

Memory needed

4.4 GB

Fastest

506 tok/s

CRYSTALCODER reaches a parameter count of 6.7B. That is small enough that hardware is rarely the obstacle, including on cards several years old. The number of cards we track that can run it: 589.

The smallest card that holds it is Tesla K20c, with a memory capacity of 5 GB, running it at a compression of IQ4_XS and producing around 27.4 tokens per second.

The fastest we calculate for it is B200, generating roughly 506 tokens per second on the strength of a memory bandwidth of 8,000 GB/s.

What this model is

CRYSTALCODER was published by Mohamed bin Zayed University of Artificial Intelligence (MBZUAI),Petuum,University of Southern California,Carnegie Mellon University (CMU),University of Illinois Urbana-Champaign (UIUC),University of California San Diego,LLM360, in the country recorded as United Arab Emirates, during December 2023. It comes out of an organisation categorised as academia,Industry,Academia,Academia,Academia,Academia,Research collective.

It works in the domain of Language, and is recorded as performing the task of language modeling/generation, Question answering, Code generation.

Because its weights were released, nothing about running it depends on a provider staying available — it is yours once downloaded. On Hugging Face it is published under the organisation LLM360.

What decides the speed

Half the cards that hold it manage more than 27.3 tokens per second. Clearing the ten tokens per second that roughly matches reading speed: 562 of them.

It is a dense model, so every parameter is read for every token produced. That makes speed track memory bandwidth almost exactly — a card with twice the bandwidth generates roughly twice as fast.

Memory here is estimated from size rather than computed from the architecture, which is not recorded for this model — the numbers are indicative rather than exact.

Training and provenance

Producing it required arithmetic totalling around 5.6 × 10²² FLOP. That figure measures what producing the model cost, and has no bearing on how fast it answers.

Training consumed a corpus of around 1,382,000,000,000 tokens of text.

Step by step

How to choose a GPU for CRYSTALCODER

The table above has already assessed every card we hold specifications for against this model. Getting to your answer takes six steps.

01

Start from the memory column

Start from what it actually needs, which is the requirement of CRYSTALCODER, needing around 4.4 GB at a compression of IQ4_XS. Capacity is the gate — a card either holds it or it does not.
02

Match the context to your actual use

Set the context to what you will actually use. The cache grows with the conversation, and it is the usual reason a card that seemed fine stops fitting CRYSTALCODER.
03

Choose how far you will compress it

Each card runs the least-compressed copy it can hold, reaching a compression of IQ4_XS on the smallest card that fits. Setting a minimum quality drops the cards that only manage it by squeezing further than you would want, and holds the comparison at one level.
04

Rank by throughput rather than spec sheet

The speed ordering is effectively an ordering by memory bandwidth, for CRYSTALCODER. It will not match a gaming ordering, because generation is bound by memory bandwidth. The card topping the list is B200, at 506 tok/s.
05

Read the fit column last

A tight fit runs, but leaves nothing spare for a longer conversation, in the case of CRYSTALCODER. Comfortable means you can grow the context later. That difference matters more than a few tokens per second, so buy for comfortable if you expect to.
06

Check the card from the other side

Each card page repeats this sweep for every model we hold, answering what else the hardware is good for beyond CRYSTALCODER.

Answers

CRYSTALCODER — common questions

CRYSTALCODER— can I run it on a GPU holding 12 GB?

Yes. The card GeForce RTX 3080 Ti, holding 12 GB, runs it at a compression of Q8_0, using about 7.9 GB and generating roughly 57.7 tokens per second. The fit is comfortable.

CRYSTALCODER— can I run it on a GPU holding 16 GB?

Yes. The card Tesla V100 SXM2 16 GB, holding 16 GB, runs it at a compression of Q8_0, using about 7.9 GB and generating roughly 71.4 tokens per second. The fit is comfortable.

CRYSTALCODER— can I run it on a GPU holding 24 GB?

Yes. The card GeForce RTX 5090 D V2, holding 24 GB, runs it at a compression of Q8_0, using about 7.9 GB and generating roughly 84.7 tokens per second. The fit is comfortable.

CRYSTALCODER— is it open source?

Its weights are published, so it can be downloaded and run on your own hardware. Note that open weights is not the same as open source in the full sense — it says nothing about the training data, the training code, or the commercial terms attached.

CRYSTALCODER— how many parameters does it have?

It has a parameter count of 6.7B. 6.7B We used the exact same model architecture as LLaMA 7B. That figure is the total, and it is what decides how much memory the model needs — roughly half a gigabyte per billion at the compression most people use.

CRYSTALCODER— who created it?

It was published by Mohamed bin Zayed University of Artificial Intelligence (MBZUAI),Petuum,University of Southern California,Carnegie Mellon University (CMU),University of Illinois Urbana-Champaign (UIUC),University of California San Diego,LLM360, based in United Arab Emirates, an organisation categorised as academia,Industry,Academia,Academia,Academia,Academia,Research collective.

CRYSTALCODER— when was it released?

It was published in December 2023. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.

CRYSTALCODER— what is it used for?

It works in the domain of Language, and is recorded as handling the task of language modeling/generation, Question answering, Code generation. A model can carry several of each, so these are the areas it was built for rather than a limit on what it will attempt.

CRYSTALCODER— where can I download it?

Its weights are published on Hugging Face, under the organisation LLM360. We do not host model files — this site calculates what hardware is needed to run them.

CRYSTALCODER— how much compute was used to train it?

Training consumed around 5.6 × 10²² FLOP. That measures what producing the model cost and says nothing about how quickly it answers once trained — inference speed comes from memory bandwidth, not from the training budget.

CRYSTALCODER— can I run it if it does not fit in my GPU?

Only by offloading, which is usually a false economy: the part held in system memory drags the whole thing down. The nearest miss we calculate falls short by 1.2 GB. Every figure here assumes the whole model is resident on the card.

CRYSTALCODER— would two GPUs run it faster?

Two cards buy memory rather than speed, which matters only if one card cannot hold it. The number that can: 589. So a second card is rarely the answer here.

CRYSTALCODER— why does the quantisation differ between cards?

Each card is shown running the least-compressed copy it can hold. The number of distinct compression levels across the cards that fit it: 4. Bigger cards get the more accurate version, and the quality floor above pins the comparison to one level.

CRYSTALCODER— how accurate are these speed estimates?

Every figure is derived from memory bandwidth and model size, not benchmarked, which is why each is published as a range rather than a single number. One example: 303–809 tok/s on B200. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

CRYSTALCODER— what GPU do I need to run it?

The smallest card in our catalogue that holds it is Tesla K20c, with a memory capacity of 5 GB. It runs the model at a compression of IQ4_XS using about 4.4 GB, and produces roughly 27.4 tokens per second. The number of cards able to run it in total: 589.

CRYSTALCODER— how fast is it on a GPU?

It depends on the card. The quickest we calculate is B200, at about 506 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and the number of cards clearing that: 562.

CRYSTALCODER— how much VRAM does it need?

It needs about 4.4 GB at a compression of IQ4_XS, which is what the smallest card that runs it uses. Less compression needs more: the figures in the memory column above are recalculated for each card, because each one holds the least-compressed version it can.

CRYSTALCODER— can I run it on a GPU holding 8 GB?

Yes. The card CMP 170HX 8 GB, holding 8 GB, runs it at a compression of Q6_K, using about 6.3 GB and generating roughly 137 tokens per second. The fit is tight.

Source

Original publication

Record last updated 28 November 2025

The other direction

Looking at it from the other side?

This page starts from the model. If you already own a card and want to know everything it will run, start from the hardware instead.

All AI models