HP Updates Z8 Workstations: Up to 56 Cores, 3 TB RAM, 9 PCIe Slots, 1700W

Only 8x8GB since I'm working on CPU limited simulation tasks. One of the CPUs and 4 of the DIMMs is on a daughter board in the Z640, which makes the whole thing even more impressive. It's a fairly compact machine. Storage is just a single SSD and single HDD. Graphics is a low end Quadro just to get the 4x mini-DisplayPort. The HP Z8xx series would probably be a better choice if a person is going to go crazy on the rest of the hardware. But that's just conjecture on my part based on the fact that my current processors are running near TDP (but aren't being throttled down based on the monitoring I have done.)

You see workstations like this in what I'd call "interactive simulation." Stuff where you're designing something complex via simulation and want to be able to rapidly evaluate changes. For example ANSYS (hybrid finite element simulation and CAD) works best on a single workstation and you want it to be fast so you can make quick changes. Or when your geometry is so complex it doesn't fit in any reasonable amount of memory and your code doesn't support distributing it (which is very hard for a lot of simulation types). The (Sauber? I think HP sponsors them) F1 car from the image in the article is a good example of someone who might one one of these.

A lot of battle-tested engineering software doesn't do distributed-memory parallelism well. I've personally used MCNP and GEANT4 (particle physics software) on a 48 core, 1 TB RAM workstation. It's interesting browsing the web while the computer is running gigantic calculations in the background on two [physical] CPUs, hooray for the process scheduler doing a good job.

Usually, engineering simulations (finite element analysis, etc)

May I suggest compiling software? For people without access to distributed build and test a box like this is really nice. I do have access to a huge distributed build farm but even then I use a beefy HP workstation because the software that drives the build farm is itself a bit of a pig, and massively multithreaded.

I'm an OpenShift Consultant for Red Hat. Super awesome to be able to run or build from scratch a 9 node HA OpenShift cluster without incurring an AWS or other infrastructure costs.

My usage is for building IaaS software. With a machine like this, I can have VMs with VTX passthrough turned on, and simulate a cloud, without a rack of physical servers.

As a remote employee it allows me to most of my development locally, and then use larger environments for a shorter period of time later on.

Combine 3 or 4 of them, and I can have an actual cloud running under my desk, for testing things like kubernetes deployments, with enough capacity for a few concurrent test environments.

As I said above they run basically silent, so I can use them in a shared office space without annoying my neighbors.

It lets me defer the inevitable Chrome Tab Bankruptcy. But seriously, having a heavy build system with lots of dependencies will easily max out a desktop like this.

dreamworks gives their artists machines like these so they can see the results of physics simulations on their animations in real-time while eduting instead of waiting for a bake step including things like hair and cloth

Security reviews. Powerful enough to run multiple VMs to simulate caching server, web server, app server, and database tier while simultaneously recording and as necessary tampering with all network communications while running static and dynamic code inspection tools across multiple VMs. Load it up with some high-end GPUs as well for hash cracking. Memory is a big deal as well. I work with some programs that if you are running anything under 64 GB of RAM you could be waiting all day for results and 256 GB is preferred but can still be sketchy as that may not leave enough for VMs. If you are a researcher doing fuzzing, again, this sort of power is key to reduce a fuzzing run from maybe a month or two on a a high-end laptop to a few minutes.

Running inadequate machines waste time of otherwise expensive engineers. The other day I was giving a live demo of an interception proxy and accidentally clicked on a very large HTTP response. The machine I was using only had 32 GB of RAM so I had to open task manager and kill the app so I could re-start it because the meeting would have been over by the time it loaded due to disk caching.

I used one a couple of years ago for image processing. TIFF stitching workloads were still a "go get a cup of coffee" timescale but it was amazingly fast for everything else.

I use a similiar machine as a workstation for FEI's AMIRA. Actually have about 400 GB of RAM.

Basically 3D rendering.

To do FPGA synthesis. It takes ages to finish a proper place and route.

I use a 2x12 core Xeon v2, 128GB RAM. I work full time on OS kernels, mainly FreeBSD. It helps a ton for building kernels, world, packages, and running lots of VMs. Thinking of doing the Talos II as the next step.

Imagine doing web dev in a language where you refresh the webpage to see if it worked vs having to run a long build. The former is better, even though it's not really as good as careful reading of your changes, but there are many times throughout the day when you just want the compiler to bark at you to clear any stupid type or syntax errors. The faster your machine and more parallel the build, the closer you get to that web dev experience.

I've got a workstation - not quite that spec but not far off it. I work in game development, and we work on a multiplayer game. Being able to run multiple instances of the game is crucial, (game server, plus 2,3,4 clients, sometimes with rendering, sometimes without). On top of that, our codebase is gigantic, and takes about 40 minutes to build on our workstations (slowly growing). On my i7 at home, it takes well over 2 hours.

I don't have one, but I nearly every day wish I had. I work on database, currently a lot of performance stuff. It's really annoying to do performance work on consumer HW - not enough memory (32GB is the biggest I have seen in portable laptops, and that's what I have) for interesting loads, regular thermal throttling screwing up benchmarks, single socket systems having drastically different performance characteristics than bigger NUMA ones...

Having access to machines remotely can replace some of that, but it definitely costs time.

My desktop is a 2690v4 with 128GB RAM. As a computational physicist I have access to a number of supercomputers to do larger simulations on, but queue time on them can be a hassle for 'smaller' jobs that a regular desktop can't handle. So having a decent middle ground machine is very useful for low resolution testing of jobs you'll throw at the supercomputer later.

Compiling C++ codebases. I have a 170 KLOC codebase that compiles in 20 minutes on a single core. Ideally you'd have at least 40 cores to make changing headers bearable at all.

I do 'small data' analysis and having a machine powerful enough to not have to deal with clusters is a godsend and saves me so much time. With 64 gigs and a dozen cores most of my needs are handled. And if I had 128 gb and 24 cores I'd be able to do even more.

Not a user of such behemoths myself, but in my organisation we've got a lot of GIS people and a few design/architect folks that work a lot with huge datasets, complex models etc that require a fairly large amount of horsepower.

It's funny the things that need a lot of beans to run; I was sitting with one of the GIS guys and he just clicked a button which then caused all 24 cores on his rig to flatline for about 4 minutes. The end result? A map of our city suddenly had a few boxes drawn over the top of it. As it turns out he was pulling in data from half a dozen sources to map out areas where there was government run housing that experienced higher than average levels of emergency services call outs. Apparently, in the past doing that manually would probably have taken several people about a week to do and wouldn't have been as accurate.

> How's the buying experience for computer hardware on taobao?

pretty happy with my experience. vendors posted the stuff on the same day, delivered within 24 hours as I live in Shanghai. it might be different story if you live in other countries as the refund policy says you can return the stuff within 7 days with no questions asked, you probably can't do that within the 7 days time frame if the package need to be posted overseas.

> I bought everything for a dual Xeon 2670v1 system (starting with Dell T5600 as base) from ebay and it came out to $1000 last year including $150 for 64GB of ECC RAM.

I also have a dual E5-2670v1 system, actually I am posting from it now. I found a single E5-2696v4 is going to be 40-50% faster than the dual E5-2670v1, dual 2696v4 almost triples the CPU performance for my workload. this has been confirmed by both cinebench score and benchmark results of an in-house application I developed. RAM bandwidth is not going to be a big jump.

> Dual E5-2696v4 seems like the current sweet spot for DIY.

Dual E5-2696v3 with some firmware/bios hack to push all cores running at higher speed is probably more $ efficient. You can get a faster system that is $500 cheaper. ;)

And last I checked there isn't a single Ryzen based workstation on the market, something I wouldn't expect to change anytime soon.

Yes if you want something that needs the fastest CPU you probably don't want this machine, but in that case you are probably willing to spend a lot more. The laptop I upgraded from had a score of 250, so for me it's a big difference.

I checked prices for sourcing my own hardware on the same platform, but it would have been a lot more expensive. The problem here in Europe is there is less second hand enterprise hardware available (if I remember correctly, a compatible motherboard alone would have been ~€200, and the RAM was about the same) and importing something from outside Europe means you need to pay customs fees, which in my country adds ~30% to the price.

ebay has bazillions of them.

FYI your link doesn't work.

I have it under my desk, it's quiet compared to desktops from a couple of generations ago, but loud compared to modern laptops (i.e. that are usually silent). I like the ability to run dual 4K displays which my laptop can't :)

I hope you don't need USB 3 for anything then.

I see. So essentially it would make sense to get one with smaller memory and buy more when price goes down. Thank you for explaining.

According to the SSG website, Autodesk Maya will fully support the SSG, including the 2TB memory.

"ProRender plug-in for Maya has full support for the out-of-core rendering feature found on the Radeon™ Pro SSG which allows the application to tap into the full 2TB of memory found locally on the card. This allows users with very large models to render fully accelerated by the GPU. "

If you are talking about srlf-build, there is reason to avoid that for people who can afford. It is called warranty. Hardware rig like this goes through testing and the support package usually is pretty good. I don't have first-hand experience, but I know Oracle would send engineers onsite to troubleshoot if required. Unless you are ready to tackle the build, the testing, and the support of the hardware rig, you want to avoid. You don't have to be Dropbox or Netflix (I don't believe they build custom servers like facebook and Google does, perhaps they buy from other vendors or from opencompute project).

Indeed, we recently asked invoices for a new machine for deep learning with 768GB memory, modern Xeon CPUs and the latest generation nVidia Tesla. The HP offer costed nearly twice as much as Dell's.

Because there is huge demand due to smartphone production. Up to 2015 prices were OK, you could buy a 8GB DDR3 non-ECC dimm for $20. Now you need like $50.

1.8 implemented multithreading.

I've purchased whole racks full of servers that didn't have as much compute power and cost a heck of a lot more than this. I use a nice mobile Titan but it is nothing compared to this.

What crazy times we live in. I can't justify buying one of these. I keep tying to come up with a good reason. I suppose YouTube would finally run smoothly.

I use my z820 and z620 for OpenShift, VMs & running my 6 monitors.

I thought people used OSX/Win for video editing?

Do you generally run openstack on host OS itself or within VM? Or to clarify - I have noticed some problems when Openstack runs inside virtualized environment, but if it directly runs on host OS itself, it seems to be fine.

Interesting - what differentiates workstation-class machines like these from dedicated server hardware? They're marketed differently, right?

You got that last sentence right: I’m not aware of any apps which can take advantage of even half this many cores. Intel is deficient in terms of memory bandwidth, so most things begin to flatten out at 8 physical cores or so, and take a nosedive when you attempt to use all of them.

While you are right that there are very few apps able to take advantage of quad GPUs, for anyone who does serious number crunching it’s not a problem at all to use all of them using popular deep learning frameworks. They aren’t just for deep learning, you can do a lot more with them. And e.g. PyTorch will automatically parallelize certain things across multiple GPUs for you, with near linear speed up.

CST will use them for EM simulation, with near linear performance increase for added GPU. I have 4 Teslas in my Microway workstation. Fortunately it also allows for the addition of a decent graphics card, so I can run 3x 4k monitors.

I somewhat disagree. Maybe there are few compiled applications that can use those resource, but it's easy to write code that uses all of them.

I have a 40 core CPU machine with a GTX 1080 Ti GPU. I run deep learning models with 90% GPU utilization and those 40 cores are barely used.

I would love to have 3 more GPUs to run in parallel to test different neural network architectures. Sometimes I'll run a CPU script at the same that processes machine learning data that uses all 40 cores.

I would use 4000 CPU cores and 10 GPUs in a machine if I could get them and I don't even do machine learning full time. I'm personally happy to see this trend of more core counts.