One more thing. Is this a memory intensive app ? We just had an app in the DC community that was very memory intensive. While I can't prove that the Genoa's 12 channel ddr5 4800 was the reason, it absolutely kicked butt in this app. Without benchmarks on this app, I can't say what difference it will make, but just saying 12 channels of DDR5 4800 is a LOT of bandwidth.Funny you say that, I have an intern this summer who will be doing some benchmarking for me on these servers just to understand where we currently stand. One of the parameters we wanted to study was, as you mentioned, the impact of storage speed on the analysis runtime. The IT Department doesn't believe that the VMs introduce a slow-down so this is one way to verify that...
I don't suspect it is. The software itself doesn't use a lot of RAM. If I were to spec out a Genoa SKU, I'd probably do a pair of sticks for every CCD on the SKU, e.g. if it's 8 CCDs then I'd have them install 16 sticks. The goal would be a full dual channels per 8 cores, like how it would be on desktop. RAM sticks are also ungodly expensive, especially when I know I don't need the capacity. I'd like to avoid paying for more RAM than I need but at the same time not skimp on the number of memory channels.One more thing. Is this a memory intensive app ? We just had an app in the DC community that was very memory intensive. While I can't prove that the Genoa's 12 channel ddr5 4800 was the reason, it absolutely kicked butt in this app. Without benchmarks on this app, I can't say what difference it will make, but just saying 12 channels of DDR5 4800 is a LOT of bandwidth.
Coz they don't know what a context switch is. Writing to physical storage requires a system call which involves moving from user mode into the privileged kernel mode. This can only happen with a context switch which is expensive. I measured a performance hit of more than 40% in 4K random writing on the 980 Pro inside a VM. The more your application depends on writing, the more it will suffer. If it buffers the writes into faster RAM and writes at once, it might fare better. But if it is continuously doing 4K writes as it runs, it will fare miserably.The IT Department doesn't believe that the VMs introduce a slow-down so this is one way to verify that...
I pay between $72 and $76 per 16 gig dim. I pay about $924 for 192 gig DDR5 4800 registered. Not bad if you ask me. And this is supermicro certified for Genoa, micron usually.I don't suspect it is. The software itself doesn't use a lot of RAM. If I were to spec out a Genoa SKU, I'd probably do a pair of sticks for every CCD on the SKU, e.g. if it's 8 CCDs then I'd have them install 16 sticks. The goal would be a full dual channels per 8 cores, like how it would be on desktop. RAM sticks are also ungodly expensive, especially when I know I don't need the capacity. I'd like to avoid paying for more RAM than I need but at the same time not skimp on the number of memory channels.
Gah dang. Dell is overcharging us out the wazoo then. I could've sworn I saw something like a few hundred for a 32 GB stick on their system configurator.I pay between $72 and $76 per 16 gig dim. I pay about $924 for 192 gig DDR5 4800 registered. Not bad if you ask me. And this is supermicro certified for Genoa, micron usually.
Proof:Gah dang. Dell is overcharging us out the wazoo then. I could've sworn I saw something like a few hundred for a 32 GB stick on their system configurator.
This is what I see on Dell's website... Gotta love those prices...Proof:
Supermicro (Micron) 16GB 288-Pin DDR5 4800 (PC5-38400) Server Memory (MEM-DR516L-CL01-EU48)
16GB DDR5 4800 Server Memory 1.1V 1Rx8 ECC UDIMM MEM-DR516L-CL01-EU48 Supermicro Certifiedstore.supermicro.com
And this is DDR5, NOT DDR4 and that makes a big difference
32 gig dimms $122
Supermicro (Hynix) 32GB 288-Pin DDR5 4800 (PC5-38400) Server Memory (MEM-DR532MD-EU48)
32GB DDR5 4800 Server Memory 1.1V 2Rx8 ECC UDIMM MEM-DR532L-HL01-EU48 Supermicro Certifiedstore.supermicro.com
If an IT manager had a brain, he would buy from smaller companies. Supermicro sells blades fully configured, and they charge way less than Dell.
why? does our forum fella or gal sell themselves for leaks?You are cut-and-dry evidence of that.
They never do mark, they never do. Head of our IT dept is as daft as they come. The ah heck knows to buy Amd, but it stops there.If an IT manager had a brain
Dell and hpe are the apple of the enterprise world. Cisco and Ubiquiti are the apple of the networking world. good in name and rapport, abysmal documentation or support outside the confines of a manual or help document. all 4 are household names.This is what I see on Dell's website... Gotta love those prices...
View attachment 84283
Dell charges for their enclosure design, gotta recoup that bs r n d money, but also their multi tier support you get in touch with when sh** hits the fan. Ya got your email and phone tiers 1 and 2 depending on how they classify em now and then the highest tier when they'll have some weasel drive out to replace a part or if you're some big shot company they'll fly out the parts same day. I don't think at any company I've ever been with has our it director or head known who the personal managers were assigned to our accounts. Current place of work has a contracts total of about 90m split between dell and hpe and we're treated like loiterers at a walmart coffee shop in toledo.My point exactly. Now rant..... I have like 10 servers (all EPYC) with server grade hardware, all purchased from newegg, supermicro, Amazon, or ebay (cpus on ebay). I have like 1/10th the money invested in them that Dell and the like charge. I have like 100% uptime with a 24/7/365 load. My only downtimes are the local power company, and that is not very often. The only place I really save than companies would not do is SOME of my CPUs are not retails, but QS, and a couple of ES. That added price would only be a couple thousand more per CPU. It does not justify the prices that Dell and the like are charging.
My last point..... INSURANCE, I bet you can buy that way cheaper than Dell support. How else is it that they make so much money >???? And my point on reliability.... When the hardware is server grade, and like never fails, You don't need much money to recover, as it VIRTUALLY NEVER happens ! And I am speaking about enterprise uptime !Dell charges for their enclosure design, gotta recoup that bs r n d money, but also their multi tier support you get in touch with when sh** hits the fan. Ya got your email and phone tiers 1 and 2 depending on how they classify em now and then the highest tier when they'll have some weasel drive out to replace a part or if you're some big shot company they'll fly out the parts same day. I don't think at any company I've ever been with has our it director or head known who the personal managers were assigned to our accounts. Current place of work has a contracts total of about 90m split between dell and hpe and we're treated like loiterers at a walmart coffee shop in toledo.
yep, not my company and isn't my responsibility. as you said and I've said to saylick or someone else here, epyc is going to be tested more thoroughly by amd than threadripper or ryzen would be, even their pro lines. server grade stuff is way different than consumer or prosumer or "low end" business sumer. it's expensive for a reason. odm pricing is their pricing on top of oem part pricing which is heavily discounted for the odms by amd or intel. dc's that need to qualify hardware do it for their own reasoning whatever that reasoning may be. that's all I'll say on the matter for now.My last point..... INSURANCE, I bet you can buy that way cheaper than Dell support. How else is it that they make so much money >???? And my point on reliability.... When the hardware is server grade, and like never fails, You don't need much money to recover, as it VIRTUALLY NEVER happens ! And I am speaking about enterprise uptime !
I'd warrant that celestial mechanics of possible Earth impact objects is plenty important 😅I would say earthquake monitoring and analysis is of greater importance in the immediate time frame over things millions of light years away.
one moment you're here and the next you're nothing or a few hours or days later depending on the impact. live each day as if it were you last, especially once you hit 50.I'd warrant that celestial mechanics of possible Earth impact objects is plenty important 😅
NASA's remit covers plenty of material that is of immediate and long term interest to the rest of us, including EV aircraft among other things.
I think the main problem with IT Managers is Liability if custom build hardware fails. Most of them can suggest and build custom servers easily. But in the event of a failure (the reasons really doesn't matter whether it is due to the custom build or not), they would get the blame for custom build. Therefore, why should they risk that type of scenarios when the money spent on the hardware is not their own? Same reasons for not switching Intel Systems to AMD Systems.A lot of these IT people are wannabes. They think they know about IT but what they really excel at is spouting a bunch of bull in the most convincing manner to someone who has no clue, which is what most higher-ups are like. I had this cloud "consultant" warn me that changing the VM CPU type from Cascade Lake to Ice Lake in Azure isn't recommended coz we hadn't tested it. In what world does a simple CPU change lead to issues? I went against his advice and pushed for the change and of course, nothing happened. That's the whole point of being in the cloud. Flexibility. I don't know if being abundantly cautious is an act of his or not but it sure makes non-techies dealing with him think that servers are very complicated things where any misstep can lead to a disaster and of course, he's there to help get them through anything.
They seem to be fine with that (RMB, PHX) - the cadence of announcements is the key. Slides at CES and availability starting in H2.CES might be too optimistic, unless AMD is now going to accept the recent practice of announce at CES, ship after Computex.
Exactly!I think the main problem with IT Managers is Liability if custom build hardware fails. Most of them can suggest and build custom servers easily. But in the event of a failure (the reasons really doesn't matter whether it is due to the custom build or not), they would get the blame for custom build. Therefore, why should they risk that type of scenarios when the money spent on the hardware is not their own? Same reasons for not switching Intel Systems to AMD Systems.
Joke aside, do anyone realizes how often we still hear "AMD isn't really compatible" etc. ?Exactly!
"No one ever got fired for buying IBM" they used to say 😄
We use Windows Server. I don't know anyone in my circle who uses Linux. Pity, I know. You need to be highly proficient in debugging stuff if you intend to use Linux for production. I'm not even going to ever try using it coz I'm not paid enough to let my company enjoy saving the cost of Windows licensestbh igor makes a point but even I'd recommend any company to qualify a system before making a cpu change. Usually you do it in the run up to deciding to change not change it out of the blue. ice lake did have problems on some linux distros from what i remember.
I'm not obsessed with bananas though I AM drinking banana flavored milk on some days. I'm scared of eating a banana thanks to the organ damage I sustained from my bout with diabetes. Anything that will raise my blood sugar quickly is a complete no no for me. I usually get obsessed with stuff for a while and it usually ends badly. In my twenties, there was one week where I ate only homemade fries and drank banana milkshake (liters of it!) and tea. It stopped when I felt a really sharp pain in my back like someone stabbed me with an electric bolt. Ultrasound revealed my kidneys were full of gravel. Drank 3 liters of water every day for almost a month to flush them out. No more banana milkshake for me!dangle a bunch or two of bananas and he'll be a happy man. altho i wish igor would upgrade his personal taste to banana walnut bread like civilised people. nothing beats a warm load of banana walnut bread. was making 2-3 loaves a week during covid. thank the heavens for stretchy trousers.
Interesting. I can't imagine running anything on Windows. Automation/server orchestation (odd configs, odd installation packages, ), almost no support for service management (systemd-like dependencies), horrible scripting experience, network management, antivirus, updates applied randomly, etc. Besides the container solutions are far from smooth last time I checked.We use Windows Server. I don't know anyone in my circle who uses Linux. Pity, I know. You need to be highly proficient in debugging stuff if you intend to use Linux for production.