Lightnews — Scholar-powered news

Hamza M @hpcwrangler.com · 1h

The Nvidia DGX Spark is NOT as supercomputer. No more than your iPhone is.

I wish reviewers would stop calling it that. I can't even take you seriously if start with that premise.

Yea I said it.

#HPC #Supercomputing

Hamza M @hpcwrangler.com · 1h

Not even from the AWS console.

Other regions work fine (same code). Not changed anything on our end.

Anyone seen anything like this before? Any ideas?

Hamza M @hpcwrangler.com · 1h

I'm have a weird problem on #AWS but only in eu-west-1 and eu-west-2.

We have some code that automates VM instance creation and then connects to the instance via SSH. Was working fine globally last week.

Now, any VM created in eu-west-1 or eu-west-2 is unresponsive. Can't SSH in to it.

1

Hamza M @hpcwrangler.com · 16h

If we're going to redo DC power cabling all over again why not do it right.

Wouldn't it be cool to have a PoE style protocol but with wiring and connectors gauged for 1MW? Plug in a single cable into a rack. If the protocol negotiates everything is fine you get power, else it stays off.

James “👻🎃 Emacs 🎃👻” Cuff @forbiddenunix.com · 17h

800V racks baby!

“Up to 5% improvement in end-to-end power efficiency

Maintenance costs reduced by up to 70% due to fewer PSU (power supply unit) failures and lower labor costs for component upkeep

Lower cooling expenses from eliminating AC/DC PSUs inside IT racks”

#hpc #cloud #ai

Rich Miller @techjournalism.bsky.social · 17h

NVIDIA and 20 of its largest customers are preparing to r change how they power GPU clusters, enabling more efficient & powerful AI factories.

At OCP, NVIDIA says CoreWeave, Lambda, Nebius and Oracle are now designing for 800 VDC data centers.

datacenterrichness.substack.com/p/ai-factori...

Hamza M @hpcwrangler.com · 16h

Haha... I'm not sure if you're joking but you're probably right either way in that's what they'll aim for! 😂

1

Hamza M @hpcwrangler.com · 16h

So that's a total of 26GW of new capacity for just OpenAi announced in just the last 3 weeks or so....

right....

1

Hamza M @hpcwrangler.com · 16h

Cheaper (thinner) cables too.... but i reckon no one will touch a rack anymore unless the whole thing is powered down...so maintenance and down times might not improve by as much as projected.

will need DC-DC conversion at the rack level though surely and that's not particularly efficient.

1

Hamza M @hpcwrangler.com · 21h

I should try and remember to post this on here more regularly. Your weekly #HPC news with a dose of KPop Demon Hunters

cloudhpc.news/quantum-81/

Quantum #81

Issue #81 of the weekly HPC newsletter by HMx Labs. KPop Demon Hunters styled HPC this week. Quite silicon focused with news from AMD in the AI space, Intel getting 18A working, updates on Cloud my ta...

cloudhpc.news

1

Hamza M @hpcwrangler.com · 4d

Are you trying to combine #AI with other #HPC workloads? Attempting business automation? Be curious to hear your take

cloudhpc.news/ai-adoption-...

AI Adoption (in FSI)

I don’t often talk about AI but I thought I’d share a few insights and join a few dots on what I’m seeing on AI adoption within financial services.

cloudhpc.news

Hamza M @hpcwrangler.com · 5d

no 😞 I'm out here slumming it

😂

Hamza M @hpcwrangler.com · 5d

Can anyone tell me where I can find some HBv5?

It allegedly exists according to this:
learn.microsoft.com/en-us/azure/...

but is rather absent from here:
azure.microsoft.com/en-gb/explor...

not sure if there's an az CLI command that could find it without listing all VM types in every region...

Hamza M @hpcwrangler.com · 5d

Ah well we can't have that! Take my jet instead. No jacuzzi i'm afraid but i'm sure you'll make do!

1

Hamza M @hpcwrangler.com · 5d

Shouldn't Microsoft be flying you there on a private helicopter? 😁

1 1

Hamza M @hpcwrangler.com · 6d

Title to this one is straight from @forbiddenunix.com 😁

Let’s talk more across #HPC there’s a surprising amount in common even where you wouldn’t expect it.

cloudhpc.news/hpc-in-finan...

HPC in Finance and Academia: Chocolate and Peanut Butter

HPC in academia and financial services has a surprising amount in common and I don’t just mean the obvious challenges of operating at scale.

cloudhpc.news

2

Hamza M @hpcwrangler.com · 7d

Its the deal structure that gets me.... looks like another infinite money glitch type affair.

Also I think it ends up giving Nvidia some ownership of AMD... assuming OpenAI actually hold the AMD shares that whole thing isn't just a funding trick (which it probably is).

1

Hamza M @hpcwrangler.com · 9d

Thanks James. I really enjoyed working with you too!

This paper was a lot of fun and I think we’re not quite done with this topic… more to come with other CPUs

Oh and we tangle with the beast that is numerical stability too 😁

1

Reposted by Hamza M

James “👻🎃 Emacs 🎃👻” Cuff @forbiddenunix.com · 9d

with @hpcwrangler.com

a legit #hpc legend, it was a total blast my friend!

www.linkedin.com/feed/update/...

#hpc #cloud

#hpc | James Cuff

Working with Hamza was one of the best things I've ever done. I've told him and others a number of times that his moral compass is pointed so inordinately and accurately due north, that it isn't even...

www.linkedin.com

1 1 4

Hamza M @hpcwrangler.com · 20d

Also, given hyper scalers all run CPUs you and I can’t buy anyway… does it matter?

If Big Cloud Co gets CPUs that are 5% than what you can actually buy doesn’t that offset any hypervisor tax anyway?

1

Hamza M @hpcwrangler.com · 20d

You think the idea for this would be to replace hypervisors and allow multi tenant workloads on bare metal? Sounds iffy.

Whats the real overhead of a hyper visor these days?

1

Hamza M @hpcwrangler.com · 21d

Was a bit of an exciting week for large scale compute, #HPC and AI in the UK last week and more big news from Intel, Nvidia and Microsoft. Eek.

cloudhpc.news/quantum-78/

Quantum #78

Issue #78 of the weekly HPC newsletter by HMx Labs. Well that was a bit of an ai-coaster. Nvidia taking a stake in Intel and investing heavily in the UK. Microsoft and others get in on the action to s...

cloudhpc.news

1

Hamza M @hpcwrangler.com · 23d

That's brilliant. So well articulated!

Hamza M @hpcwrangler.com · 23d

This. In Spades.

Depends on how you're measuring cost effectiveness but you'll get students to understand what's going on a heck of a lot faster when they can see it and not have to imagine a bunch of VMs and what's in them.

1

Reposted by Hamza M

Suhaib Khan @suhaibkhan.bsky.social · 23d

Beyond the water that cools the servers, data centers indirectly contribute to water use through the electricity generation needed to power their operations.

That indirect use often makes up 80%+ of the overall water use.

spectrum.ieee.org/ai-water-usage

#HPC #AI @spectrum.ieee.org

How Much Water Do AI Data Centers Really Consume?

There's a lot of confusion about how much water is used by AI. Get the real story on data centers' water consumption, and read about tech for doing better.

spectrum.ieee.org

1 5

Hamza M @hpcwrangler.com · 23d

We didn't actually order and build the cluster, but some basic maths before purchase showed the same thing when we put together our test rig.

cloudhpc.news/new-beowulf-...

That was without even spending 1k on a blade kickstarter bundle!

New Beowulf Cluster Time

We need a new test HPC cluster at HMx Labs. Time for some fun with hardware.

cloudhpc.news

1

Hamza M @hpcwrangler.com · 24d

Come on man. We need somewhere safe for all those poor destitute bits and FLOPs. Won't someone think of the bits! 😁