You don't need to be an 'investor' to invest in Singletrack: 6 days left: 95% of target - Find out more
Someone here might be able to answer my question. I run a small geology consultancy and use an older HP Z820 with Xeon dual processor work station with 128GB DDR3 1866MHz Ram and a 6GB video card for working on large datasets. This computer is getting on, so I have been looking at replacing it with a HP Z8 running dual Xeon Gold processor which has a slight bump up in processor speed, but runs faster DDR4 ram.
The great thing with the HPs is you can hot swap HDDs into them and they seem pretty damn bomb proof
I'd also run the OS and the data drives on PCIe slots and my current unit can't have the boot drive on PCIe slot.
I have also been looking at the AMD Ryzen Threadripper 24 core computers which run much faster DDR5 ram and seem to have blazing CPU speeds. I don't have a lot of experience with the Ryzens and wanted to know if they would be suitable for working with big datasets and graphic displays (with a decent GPU included).
Thanks in advance
What do you mean by large datasets, and what are you doing? Interpreting big 3D seismic datasets, looking at well logs, xyz gridding/mapping?
@wbo. Yep lots of modelling work designing 3D surveys and gridding and mapping xyz grids.
Not directly an answer to your question but most of the big O&G firms have moved to using Virtual Machines hosted on the cloud provider of your choice.... I don't know how feasible that would be for a small consultancy and it'd also depend how your data is being provided - not easy to load a SEGY straight onto the cloud for example.
How old is your current workstation if it's DDR3? Google says the Z820 was initially released back in 2012... The reason I ask is your statement:
which has a slight bump up in processor speed
...in reality it will likely be significantly faster, you can't just compare clock speed when comparing CPUs, especially across generations.
Are hot-swappable drives much of a benefit? Not sure how much storage you need but if you've got lots of drives you're probably better off with external storage anyway. There's days I'd be looking at M2 NVMe drives for anything internal.
Is your app generally CPU, memory or GPU bound (where's the current bottleneck and what are the vendor recommendations for it)? Ryzen vs Intel can be tricky to compare on paper as they both have areas they perform better than each other in so you'd usually want to find benchmarks for your specific application, or ask the vendor if they've done any comparative testing themselves.
Yes the Z8 is a good workstation and this generation is a decent incremental jump from the last one and yours is probably 3 'HP generations' back.
One thing to check with a Ryzen CPU is that they support EEC memory, this is very much worth having. They may well do, i just can't remember.
you can’t just compare clock speed when comparing CPUs, especially across generations.
Also depends on whether your data processing is multi-threaded or not, modern processors have more cores, but these only benefit you if your application can utilise them.
Not directly an answer to your question but most of the big O&G firms have moved to using Virtual Machines hosted on the cloud provider of your choice…. I don’t know how feasible that would be for a small consultancy and it’d also depend how your data is being provided – not easy to load a SEGY straight onto the cloud for example.
A lot of faff for not much benefit unless your application has been specificly designed to use micro-services and can really take advantage of cloud computing. I look after a dozen or so VM Workspaces on AWS and they run much slower than my laptop - IO is impressive though. We only really use them as it allows us to manage customer's machines which we can't do if they use their own laptop etc as their own company IT normally lock them down so it's an endless PITA to install our applications and maintain them.
SO designing 3D rather than interpreting? Ray tracing and such rather than loading multiple big surveys?
Gridding is almost always single thread. As above, we're remote displaying off a cloud, with actual work being mostly done on GPU's. There are a lot of pro's for us to use remote machines, mostly that we can get the CPU's closer to the data servers. There's a lot of IO going on, FWIW I'm using 2 times 8 core Xeon Gold
What program? Commercially available?
Thanks for the comments and the Z820 is getting on a bit now and has 1 NVMe drive in it for a working data drive, and C drive is run from an SSD.
The programmes I use use multi threading and I'm familiar with Xeon processors, but noticed that the AMD RRyzen CPUs have some pretty fast speeds, but that might not translate into a speed bump for some of the apps I use.
More than likely to go with a Z8 as I can pick up a 2nd hand unit for a decent price. It will also be able to run the C drive from a PCIe NVMe which will help things. I'm guessing that the DDR4 2666MHz ram is a bit quicker than the DDR3 I'm currently running.
I have been pretty impressed with these HP workstations and they are pretty robust and the only problem I've had is a power supply failing. Added an upgraded GPU, more ram and swapped the processors out for some 12 core units about 3 years ago.
Cloud is good, but even with 500mbit up and down internet big file transfers take a while and its good to have everything at hand.
We used those Z820's for recording video footage on location, they really are tough as old boots. But recently upgraded to EPYC/Threadripper based workstations. Can't actually remember which but COVID and cryptocurrency driven supply issues meant we couldn't get HP's again so had them custom built and it still took a long time to get Quadro cards.
They really are substantially faster, the Z820's could manage 4 channels of HD Video and 64 channels of 24bit audio (and was noisy whilst doing it). The new one does double that and barely ticks over. It needs a number of cores per channel for encoding, but doesn't need flat out speed from them, so your software's mileage may vary. The only downside is the case/build isn't as robust, so we've had to build it all into a half height server rack in a shock resistant flight case. But that's a limitation of aftermarket tower coolers.
Ours are built with 8x 2.5" drive bays and they're all hot-swappable. With the added bonus that the newest 2.5" SSD's can actually do all 8 channels at the same time if needed (it just fills up very quickly, better to do 4x in parallel than have to keep swapping and rebuilding file structures several times a day).
I got to borrow a Threadripper workstation as a trial at work, Lenovo P620 based (32-core threadripper, lots of RAM and NVMe SSDs, retail price was almost £5K). It was disgustingly fast in my tests, faster than the outgoing dual Xeon servers.
It gets very messy working out what you want from your hardware - a reservoir simulation will generally have a single, but very large 'lump' of data you're manipulating. Seismic processing is usually on a smaller piece of data (a single trace or a cluster) but will repeat this MAY times (terrabytes) so IO and ability to multithread is what matters.
I ended up buying a 2nd hand HP Z8 with 2 x Xeon Gold 6152 processors for USD1900. This should run along a bit quicker and the CPU benchmark test it estimates twice as fast.
+faster DDR4 ram
+PCIe SSDs on the mother board