Interesting that in the article, they do not say what the chip actually does. Not even once.

the real news is: "and teases an Nvidia-friendly roadmap"

The sole reason amazon is throwing any money at this is because they think they can do to AI what they did with logistics and shipping in an effort to slash costs leading into a recession (we cant fire anyone else.) The hubris is magnanimous to say the least.

but the total confidence is very low...so "Nvidia friendly" is face saving to ensure no bridges they currently cross for AWS profit get burned.

I've had to repeatedly tell our AWS account reps that we're not even a little interested in the Trainium or Inferentia instances unless they have a provably reliable track record of working with the standard libraries we have to use like Transformers and PyTorch.

I know they claim they work, but that's only on their happy path with their very specific AMI's and the nightmare that is the neuron SDK. You try to do any real work with them and use your own dependencies and things tend to fall apart immediately.

It was just in the past couple years that it really became worthwhile to use TPU's if you're on GCP and that's only with the huge investment on Google's part into software support. I'm not going to sink hours and hours into beta testing AWS's software just to use their chips.

Amazon post: https://www.aboutamazon.com/news/aws/trainium-3-ultraserver-...

AWS keeps making grand statements about Trainium but not a single customer comes on stage to say how amazing it is. Everyone I talked to that tries it says there were too many headaches and they moved on. AWS pushes it hard but “more price performant” isn’t a benefit if it’s a major PITA to deploy and run relative to other options. Chips without a quality developer experience isn’t gonna work.

Seems AWS is using this heavily internally, which makes sense, but not observing it getting traction outside that. Glad to see Amazon investing there though.

Amazon aside, interesting future here with NVLink getting more and more folks using it. Intel is also onboard with NVlink. This is like an PCI -> AGP moment, but Nvidia's AGP.

AMD felt like they were so close to nabbing the accelerator future back in HyperTransport days. But the recent version Infinity Fabric is all internal.

There's Ultra Accelerator Link (UALink) getting some steam. Hypothetically CXL should be good for uses like this, using PCIe PHY but lower latency lighter weight; close to ram latency, not bad! But still a mere PCIe speed, not nearly enough, with PCIe 6.0 just barely emerging now. Ideally IMO we'd also see more chips come with integrated networking too: it was so amazing when Intel Xeon's had 100Gb Omni-Path for barely any price bump. UltraEthernet feels like it should be on core, gratis.

Not a single mention of any benchmarks or performance.

Anyone considering using trainium should view this Completely Factual Infomercial: https://x.com/typedfemale/status/1945912359027114310

Pretty accurate in my experience, especially re: the neuron sdk. Do not use.

Heavens to Betsy, I don’t know if you can hear me, But try supporting these things if you actually want them to be successful. About the 3rd day into trying to roll your own LMI container in sagemaker because they haven’t updated the vLLM version in 6 months and you can’t run a regular sagemaker endpoint because of a ridiculous 60s timeout that was determined to be adequate 8 years ago. I can only imagine the hell that awaits the developer that decides to try their custom silicon.

The Block floating point (MXFP8/4 stuff) is interesting; the AI stuff is really pushing basic data types that haven't moved for decades.

https://en.wikipedia.org/wiki/Block_floating_point

at some point the cost of transferring will dwarf the cost you pay to NVIDA. I bet that is their bet

Single-chip specs according to:

https://awsdocs-neuron.readthedocs-hosted.com/en/latest/abou...

https://awsdocs-neuron.readthedocs-hosted.com/en/latest/nki/...

Eight NeuronCore-v4 cores that collectively deliver:

    2,517 MXFP8/MXFP4 TFLOPS
    671 BF16/FP16/TF32 TFLOPS
    2,517 FP16/BF16/TF32 sparse TFLOPS
    183 FP32 TFLOPS

HBM: 144 GiB HBM @ 4.9 TB/sec (4 stacks)

SRAM: 32 MiB * 8 = 256 MiB (ignoring 2 MiB * 8 = 16 MiB of PSUM which is not really general-purpose nor DMA-able)

Interconnect: 2560 GB/s (I think bidirectional, i.e. Jensen Math™)

----

At 3nm process node the FLOP/s is _way_ lower than competition. Compare to B200 which does 2250 BF16, x2 FP8, x4 FP4. TPU7x does 2307 BF16, x2 FP8 (no native FP4). HBM also lags behind (vs ~192 GiB in 6 stacks for both TPU7x and B200).

The main redeeming qualities seem to be: software-managed SRAM size (double of TPU7x; GPUs have L2 so not directly comparable) and on-paper raw interconnect BW (double of TPU7x and more than B200).

When they're in short supply can we name them Unobtranium?

These product might be great, but seriously, who's choosing those names? Trainium, Inferentia? It's like let's just take the words from what they do, and put a little Latin twist on them? I know naming things is one of the great problems in computer science, but really they could come up with something a little better.