Qualcomm Cloud AI 100 Ultra – 150W AI Accelerator Card with 128GB at 548GB/s

Comments

Comment by tokepi 4 hours ago

If the specs seem outdated, it’s because this chip launched in 2024. It’s available at cloud providers including Cirrascale

Comment by jauntywundrkind 3 hours ago

Thats relieving!

I'm starting to believe in Qualcomm! The Snapdragon X2 Elite Extreme is alas only available in a single system, with it's tri-channel on-package (!) memory good for 228 GB/s. But it's a beast. The core is just good, even in the mainstream form in other systems: some of the best single core performance you can get, shy of Apple.

And they are upstreaming their stuff much better! They hired a bunch of really good known FOSS developers and it seems to be working!

Now they are talking up High Bandwidth Compute as their next big thing. And I want to believe, even if it makes me nervous that they too intend to leave consumer computing completely in the dust. https://www.qualcomm.com/news/onq/2026/11/qualcomm-hbc-valid...

This is, as commented, at M4 Max levels of GB/s. It is not impressive. It has some uses! But if this was their modern opening salvo I would not be impressed.

Comment by redanddead 3 hours ago

> I'm starting to believe in Qualcomm

Someone drop the YouTube link of their robot presentation

Comment by thway15269037 4 hours ago

What's the price?

Comment by simlevesque 4 hours ago

According to their own website, they cost 1 cent.

Comment by wewewedxfgdf 3 hours ago

Less than I expected for this level of GPU power.

It's a trend I like.

Comment by bigwheels 3 hours ago

8,750 in stock at 1 cent, too. The more expensive 47 and 63 cent variants are all sold out.

Comment by thway15269037 4 hours ago

Yeah, I clicked that too. But knowing qualcomm they obviously cost an arm and a leg. Still interesting how much.

Comment by bigwheels 3 hours ago

My research indicates the price is $10-17K USD.

Comment by InsidiousWombat 21 minutes ago

that tracks, $0.01 for chip, and $9999.9 for the memory

Comment by IgorPartola 3 hours ago

I mean at ten bucks it’s not a bad deal but that’s quite a range.

Comment by 2 hours ago

Comment by 3 hours ago

Comment by 4 hours ago

Comment by dingdingdang 4 hours ago

Honestly, let me buy one, it's many times faster than the minipc I run local ai on so if the price is right I'm in.

Comment by martin1975 3 hours ago

hopefully less than 5090?

Comment by wmf 2 hours ago

Not a chance.

Comment by pearl_panel5 5 hours ago

Curious how the memory bandwidth actually holds up under sustained inference loads compared to the spec sheet numbers.

Comment by swiftcoder 4 hours ago

So it's an M4 Max Mac Studio, RAM-wise, with a GPU's worth of compute?

Comment by trvz 3 hours ago

(I know it's you comparing it to the M4 Max, not them, but:) it's Qualcomm, so it'll be compared to the M4 Max, while being slower than a M2 Max.

Comment by 0917236141 2 hours ago

Seems like a reasonable general-purpose card.

Comment by zitterbewegung 4 hours ago

Sort of interesting design It has half a terabyte of SRAM on the accelerator with a 128 GB of RAM replacing HBM with SRAM but keeping regular RAM.

Comment by 2001zhaozhao 4 hours ago

half a gigabyte not terabyte lol

Comment by zrm 4 hours ago

Yeah, this thing is unusual. It resembles a consumer GPU but using normal DRAM with more channels, so then you get more RAM but less memory bandwidth than GDDR/HBM, which is the same thing Apple does. And the TDP is pretty attractive for what it is.

But what is that amount of SRAM for? It simultaneously seems like a lot, because SRAM is so expensive, but still not very much, because what can you fit in 0.5GB? Is there a specific workload that needs something that looks like this?

Comment by zamadatix 3 hours ago

The Intel D270, Nvidia DGX Spark, and (somewhat) AMD Strix Halo also use the same approach, so it seems to be getting pretty common.

SRAM wise it doesn't seem too insane. About the same ratio of memory:SRAM as the 5090 (IIRC). Still more than I would have expected. I assume they found it helps process the current layer faster or something.

Comment by aetherspawn 5 hours ago

Ok so it’s significantly worse than Apple M6 and we don’t know if it’s cheaper or not

Edit: Ah sorry you’re right, the bandwidth is lower but the TOPS is a lot higher. I’m not sure how much effect you get from higher TOPS with such low bandwidth though.

Comment by zozbot234 2 hours ago

> I’m not sure how much effect you get from higher TOPS with such low bandwidth though.

Raised arithmetic intensity. It would be quite good for LLM prefill, and decode would want to adapt by pushing parallelism higher (saving on memory bandwidth by reusing the same layer weights for multiple streams) or, if unable to do so (e.g. due to KV cache constraints) by resorting to speculative decoding (where the expensive 'validate' step handles tokens in parallel within a single stream). It does not make the platform inherently unviable. It would also do well at other tasks such as image/video generation, which is already quite compute bound.

Comment by 4 hours ago

Comment by dist-epoch 4 hours ago

Apple M6 has 80-100 TOPS.

Qualcomm Cloud AI 100 Ultra has 800 TOPS, almost 10 times more.

Comment by grosswait 3 hours ago

This could be very fast at running embedding models (if it supports the right modes)

Comment by ipsum2 4 hours ago

Considering that memory bandwidth is the most important for inference for most neural networks, that matters more.

Comment by 3 hours ago

Comment by dist-epoch 3 hours ago

Modern inference stacks use different GPUs for decode versus prefill. This is a GPU suitable for prefill.

Comment by varispeed 5 hours ago

Aren't these specs already outdated? Seems meh. I would think players would like to bring something new to the table instead of bringing what was okay two years ago or so.

Give me 1TB RAM and 5TB/s or something that exciting.

Comment by entrope 4 hours ago

For most purposes, this seems reasonable? This looks like a single-height card, so fitting four of these in a workstation or server would give 512 GB RAM and 2 TB/s total bandwidth. That's pretty attractive for a lot of current LLMs, even if one wishes for 1 TB capacity at 5 TB/s bandwidth.

Comment by throwawayffffas 1 hour ago

If you are going down this path, MI350P and RTX PRO 6000 Blackwell is what you would want. A lot more compute a lot more vram bandwidth.

Comment by entrope 1 hour ago

Maybe. I don't know power for the MI350, but the RTX PRO 6000 is normally a 200-300W card, so you pay 2x the power for 3/4th the RAM and 3x the bandwidth. The performance frontier looks very different in terms of power to hold a mid-size model and throughout per user or across users. Price and inference stack would be big factors in what I would choose.

Comment by zozbot234 1 hour ago

> A lot more compute a lot more vram bandwidth.

But this gives you significantly more memory capacity for a compute/vram bandwidth tradeoff (i.e. arithmetic intensity) that's ultimately comparable to Blackwell, perhaps even more skewed towards compute. This seems good if you care about being able to store larger sparse models and more KV caches because that ultimately bottlenecks your ability to do inference.

Comment by varispeed 3 hours ago

But there are already plenty of products for most purposes.

Comment by throwawayffffas 2 hours ago

This is all the CPU companies playing catch up with Nvidia and AMD.

The thing you are describing is much closer to AMD and Nvidia latest generation server offerings.

See MI455X for the absolute cutting edge and MI350P for something you can use outside of a datacenter.

Comment by tempest_ 4 hours ago

5 year old GPUs are still at crazy prices.

Plenty of businesses want in local inference and are priced out.

Comment by varispeed 3 hours ago

You can't do quality inference on such a toy. It's a myth.

Comment by redanddead 3 hours ago

A great opportunity for huawei

Nobody helps China like American corps helps China. Domestic anticompetitive behavior spurs demand for international competition

Comment by mciancia 4 hours ago

8 of those cards give you 1TB of ram and 4.4TB/s

Comment by PhunkyPhil 4 hours ago

It may be a semantic thing but I don't think bandwidth pooling works like that? Unless your blocks are distributed perfectly between cards there's going to be data passed between each card and from VRAM to compute. Correct me if I'm wrong though

Comment by msandford 3 hours ago

I think the way it works is that the weights have to get streamed through the compute and the weights can be split very easily. With multiple cards you do have a coordination/communication phase after each layer of weights but that's usually much less significant than the compute phase.

Comment by throwawayffffas 1 hour ago

It depends on how you slice the weights. You can split along layers which means each GPU waits for the previous one to finish or you can split the tensors across gpus.

Splitting the tensors is closer to n times the effective vram bandwidth but there are still overhead, sync waits, and reduce steps that eat away at the overall performance.