If the specs seem outdated, it’s because this chip launched in 2024. It’s available at cloud providers including Cirrascale
rpcope1 8 hours ago [-]
Given the rush to dig MI50s and V100s out of the trash and press them into service, 2024 seems like future space alien technology.
eyegor 6 hours ago [-]
Holy cow the v100s go for like $1k on ebay now... They were selling for $100 or less a couple years ago.
jauntywundrkind 12 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.
redanddead 12 hours ago [-]
> I'm starting to believe in Qualcomm
Someone drop the YouTube link of their robot presentation
Those guys were fast on the draw to cover the body with those sheets.
Rohansi 6 hours ago [-]
> tri-channel on-package (!) memory good for 228 GB/s
Why is on-package memory impressive/better in this case? It's one channel more than your typical consumer x86 system (192-bit bus vs 128-bit) which is something but you could technically have quad channel with two LPCAMM2 sockets?
WarOnPrivacy 6 hours ago [-]
Buy for the low, low price of - nope. Not for you.
Cirrascale (8-card Ultra server, 128 GB each, 512 GB system RAM):
$4,699/mo (monthly term) or $3,759/mo (annual term)
Cloud GPU rental (per-GPU): from $5.15/GPU/hr
At $5 per hour these are absolutely terrible value. You can rent an H200 for that price and an H200 is much better in every way. Weird
teruakohatu 17 minutes ago [-]
Not if the customers cannot rent a H200 (because they have been told not to rely on foreign gpus)
txdv 1 hours ago [-]
I guess they get their customers on scale, can you get hundreds of h200?
thway15269037 14 hours ago [-]
What's the price?
simlevesque 13 hours ago [-]
According to their own website, they cost 1 cent.
wewewedxfgdf 12 hours ago [-]
Less than I expected for this level of GPU power.
It's a trend I like.
bigwheels 12 hours ago [-]
8,750 in stock at 1 cent, too. The more expensive 47 and 63 cent variants are all sold out.
thway15269037 13 hours ago [-]
Yeah, I clicked that too. But knowing qualcomm they obviously cost an arm and a leg. Still interesting how much.
bigwheels 12 hours ago [-]
My research indicates the price is $10-17K USD.
InsidiousWombat 9 hours ago [-]
that tracks, $0.01 for chip, and $9999.9 for the memory
IgorPartola 12 hours ago [-]
I mean at ten bucks it’s not a bad deal but that’s quite a range.
12 hours ago [-]
13 hours ago [-]
5 hours ago [-]
13 hours ago [-]
dingdingdang 13 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.
martin1975 12 hours ago [-]
hopefully less than 5090?
wmf 11 hours ago [-]
Not a chance.
pearl_panel5 14 hours ago [-]
Curious how the memory bandwidth actually holds up under sustained inference loads compared to the spec sheet numbers.
zitterbewegung 14 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.
2001zhaozhao 14 hours ago [-]
half a gigabyte not terabyte lol
zrm 13 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?
tromp 2 hours ago [-]
0.5 GB of SRAM is exactly what the Cuckatoo32 [1] Proof-of-Work scheme needs for optimal solving efficiency. Solutions are 42-cycles in random bipartite graphs of 2^32 edges on 2^32 + 2^32 nodes, which can be found by repeatedly removing edges that have only one adjacent edge. Tracking which nodes have incident edges takes 2^32 (completely randomly accessed) bits which is exactly 0.5GB. Another 0.5 GB is needed to track alive edges but those can be accessed serially for which DRAM suffices.
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.
swiftcoder 13 hours ago [-]
So it's an M4 Max Mac Studio, RAM-wise, with a GPU's worth of compute?
trvz 12 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.
0917236141 11 hours ago [-]
Seems like a reasonable general-purpose card.
skavi 7 hours ago [-]
is this hexagon?
rsendv 6 hours ago [-]
Yeah, bunch of HTPs
aetherspawn 14 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.
zozbot234 12 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.
14 hours ago [-]
dist-epoch 14 hours ago [-]
Apple M6 has 80-100 TOPS.
Qualcomm Cloud AI 100 Ultra has 800 TOPS, almost 10 times more.
grosswait 13 hours ago [-]
This could be very fast at running embedding models (if it supports the right modes)
ipsum2 13 hours ago [-]
Considering that memory bandwidth is the most important for inference for most neural networks, that matters more.
12 hours ago [-]
dist-epoch 13 hours ago [-]
Modern inference stacks use different GPUs for decode versus prefill. This is a GPU suitable for prefill.
varispeed 14 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.
simoncion 4 hours ago [-]
> I would think players would like to bring something new to the table instead of bringing what was okay two years ago or so.
Speaking of... from [0]:
Introducing Qualcomm Cloud AI 100 Ultra
High performance and cost-optimized cloud AI inferencing for generative AI
Written by
Rashid Attar
Nov 13, 2023
Today we announce the Qualcomm Cloud AI 100 Ultra, a new member of our portfolio of cloud artificial intelligence (AI) Inference cards, purpose-built for generative AI and large language models (LLMs). ...
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.
throwawayffffas 11 hours 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.
zozbot234 10 hours 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.
entrope 10 hours 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.
throwawayffffas 8 hours ago [-]
MI350P, H200 NVL and Nvidia RTX PRO 6000 are all 600W but compute wise all of them should be 3 to 4 times more performant than A100 Ultra according to their specifications.
600w is way off the vf curve though... Tiny bit of tuning and you'll get ~80-90% of the performance at ~150w nominal. At least that's what I got out of some pro 6000s when I spent 15 mins to underclock it.
varispeed 12 hours ago [-]
But there are already plenty of products for most purposes.
tempest_ 14 hours ago [-]
5 year old GPUs are still at crazy prices.
Plenty of businesses want in local inference and are priced out.
varispeed 12 hours ago [-]
You can't do quality inference on such a toy. It's a myth.
redanddead 12 hours ago [-]
A great opportunity for huawei
Nobody helps China like American corps helps China. Domestic anticompetitive behavior spurs demand for international competition
creato 6 hours ago [-]
What exactly do you propose they do? Do you think they are holding back supply? Should they sell at less than market price? That's just handing the profits to scalpers/middle men.
I'm sure NVIDIA, AMD, Intel, Qualcomm, etc. would sell more hardware if they could, but they can't.
throwawayffffas 11 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.
mciancia 14 hours ago [-]
8 of those cards give you 1TB of ram and 4.4TB/s
PhunkyPhil 13 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
msandford 13 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.
throwawayffffas 11 hours 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.
Rendered at 10:32:58 GMT+0000 (Coordinated Universal Time) with Vercel.
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.
Someone drop the YouTube link of their robot presentation
https://youtube.com/shorts/aZmt5E151Bk
Why is on-package memory impressive/better in this case? It's one channel more than your typical consumer x86 system (192-bit bus vs 128-bit) which is something but you could technically have quad channel with two LPCAMM2 sockets?
It's a trend I like.
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?
[1] https://github.com/tromp/cuckoo
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.
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.
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.
Qualcomm Cloud AI 100 Ultra has 800 TOPS, almost 10 times more.
Give me 1TB RAM and 5TB/s or something that exciting.
Speaking of... from [0]:
[0] <https://www.qualcomm.com/news/onq/2023/11/introducing-qualco...>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.
https://www.servethehome.com/the-amd-instinct-mi350p-is-a-hb...
Plenty of businesses want in local inference and are priced out.
Nobody helps China like American corps helps China. Domestic anticompetitive behavior spurs demand for international competition
I'm sure NVIDIA, AMD, Intel, Qualcomm, etc. would sell more hardware if they could, but they can't.
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.
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.