Anyone wanna help me with a 3,055.00$ semi-budget build?

[Spork Schivago]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

Why the hell didn't he RMA the defective components?!

What you are posting really sounds like something is broken.
I hate it when you don't do have problems with the hardware and don't try to fix it...
What would you have done when it was an Intel?


I have a Ryzen myself (1700x on BIostar X370GT7) and no problems whatsoever, though only DDR4-2400 Memory...

My friends have had horrible luck with Ryzen builds, but every time, it turned out to be someone thinking they knew how to hook up components, just to find out they plugged a header in wrong and was sending 5V to GND or something stupid like that.

Finally, after the second RMA, I insisted (instead of telling) that if they wanted me to work on any more of these, when they get the parts back in the mail, instead of having X or Y assemble it for them, just bring it over and have me assemble it. If it's assembled and not working, I'm not going to look at it. It's a waste of my time (I wasn't charging because they were friends) and a waste of their time.

First time I pieced together the PCs, they had no issues. Except one guy, bought a 300$ brand new "gaming" rig and has some serious bottleneck issues. Don't know what he was thinking he'd get for 300$. That was a pre-built one. He just asked me to max out the CPU, so we put the fastest Ryzen it could handle, which was an APU, but I warned him, it wouldn't make much of a difference because of how the system's built, which it didn't.

[diif]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

Yes, for SolidWorks alone, we might not notice much difference, but I'd still like to know, benchmark wise, which would be faster?

Eventually, the designs won't be basic at all, but very complex. Now, it doesn't matter so much, but it will determine which way our soon to be company goes in many ways.

The PC we either build or the server we upgrade will essentially be the heart of our company. There's other software programs that will be running on it, some that will require multiple cores, some that have the potential to be very write / read intensive (in regards to the hard drives), hence the question.

If you can only tell with benchmarks then it makes no difference either way IMO.
Me, I'd make do, see if your idea works first before expanding.
Never forget KISS.

[Snayperskaya]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

I thought I had answered these earlier, but maybe I didn't. The program mainly benefits from higher core clock speed, although multi-threading will come in handy during rendering.

32GB of RAM is what the company recommends for optimal performance. More would be okay, but isn't needed for this program alone. Fastest RAM is most desirable.

We will be using 1 expansion slot, for the GPU. Unless, of course, eventually, we purchase another CPU, more RAM, another secondary riser, and another P4000. Then each riser card would be using 1 slot each.

Yes, E5-269* is expensive. I was hoping though, that with something like the E5-2696v4, used, we could run the program in a virtual machine. Although the 22-cores would only help during rendering for the program and the base core clock speed is what the program would mostly be using, the virtual machine itself would benefit from the multiple cores.


I understand that lower ranks have less density per module. With our board, the RAM is guaranteed to run at the speeds specified, regardless of the ranking. But it seems there's more to registered RAM than clock rates, voltages, and speed. Their load-reduced RAM is supposed to be much faster than their non-load reduced RAM. Both are registered. Load reduced is more expensive.

If you check the chart here:

https://support.hpe.com/hpsc/doc/pub...r_na-c05240460

And scroll down to the v4 processors, and look at the load reduced, you'll noticed that the dual and quad ranked 32GB and 64GB modules match in clock rate, speed, and voltage. With a certain Dell server, I was reading for optimal performance, purchase dual rank, not single.

But with HPE, I have the choice of Dual Rank Load Reduced, Quad Rank, and Octo Rank, even though the Octo Rank is not listed on that page. So would the best option be purchasing one 32GB dual rank load reduced stick? Or, would I have faster speeds by purchasing four 8GB Single Rank sticks? All of my choices that I'm considering are the 2400MHz ones for the v4 processors. With this server, I don't think there's any options for overclocking, and I do not want to overclock the server. What happens if I purchase four 32GB dual rank Load Reduced sticks? Would they outperform the single rank four 8GB non-Load Reduced sticks?

The RAM has it's only cooling system, the CPUs have their own cooling systems, etc.

Thanks.

I am quite familiar with Dell and HP rackserver solutions (as I work with them on a regular basis). If you can get them with your budget, by all means go for it. They are pretty reliable, the only thing you'll need to get them going for quite some time is updating firmwares (even that can be done fairly easy).

I'd just like to point out that the cooling solution on those rackmount chassis requires datacenter-style air conditioning (cold air being blown directly into the front of the server), unless you can tolerate the fans going crazy. Do you already have a cabinet for them? If not, I'd suggest you take a look at the tower versions, since their cooling system is better designed for non-datacenter infrastructure operation, like the HPE ML350 Gen9 or Dell's T630. They have more efficient cooling solution than rackservers due to not being constricted to 1/2 rack units.

Also, servers like this don't overclock at all. Never delved into it, but you can't do it like in a desktop motherboard where you can input clocks and voltages directly - while a BIOS mod or something may allow you to do this, I'm pretty sure the overall system reliability would be severely compromised. Neither server/workstation boards like Supermicro, Tyan or Intel allow native overclocking. Asus or other brands may allow you to set a higher multiplier considering the specific CPU model isn't locked by Intel.

CPU: If single core performance is essential, higher clock models like the 2637, 2643 and 2667 are better suited. Even with lower number of cores, they pack quite a punch on mixed scenarios.

Passmark have a nice database on overall performance for single-threaded, all cores and multiple CPU performance:

https://www.cpubenchmark.net/singleThread.html
https://www.cpubenchmark.net/high_end_cpus.html
https://www.cpubenchmark.net/multi_cpu.html

RAM: I'd suggest you try to use all the channels the motherboard offers for higher bandwidth. There's an article on Anandtech on LRDIMM vs RDIMM: https://www.anandtech.com/show/6068/...-latest-twin/6 . Too bad those server pieces are few and far between.

If you want to dive more into it (server equipment, that is), ServeTheHome's have some nice articles. https://www.servethehome.com/

[jiroy]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

Why the hell didn't he RMA the defective components?!

What you are posting really sounds like something is broken.
I hate it when you don't do have problems with the hardware and don't try to fix it...
What would you have done when it was an Intel?


I have a Ryzen myself (1700x on BIostar X370GT7) and no problems whatsoever, though only DDR4-2400 Memory...

Some taboos here , someone hiding technologies to his adversaries . No way to prove it but longtime experience and self conclusions . Two companies against one , you fill the blanks Stephan , i can't start an inquiery here and i'm not willing to . All i can say , i follow the strongest companies . About this friend , he's an expert , he doesn't visit me but when in something exceeding his knowledge , but I learnt the hard way those kind of friendship , and frankly , I took the conversations where suits me , not his curiosity .
According to my experience , and his briefing , for a company to release three consecutive Bios updates , then they know there's something wrong .. That's all i know .

[Spork Schivago]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

If you can only tell with benchmarks then it makes no difference either way IMO.
Me, I'd make do, see if your idea works first before expanding.
Never forget KISS.

Right, and now, we are making do before expanding, but I still need to know which option would be faster.

It does make sense. Right now, we are not ready to run the other software, we're no where near ready to run that other software. But we will be. We have a few other things to do first. When the time comes, then we have to spend money to upgrade the server to handle that other software.

The whole point is can we upgrade the server now to run SolidWorks, and handle the other software? Or do we need a dedicated workstation for SolidWorks. It's too slow on the current system I have, running Windows 7 Pro, so an upgrade is needed now or a new workstation.

For what it's worth, FPS isn't what really matters with these Quadro's. They're not meant for gaming, they actually are kinda the opposite of what you'd want for a gaming system. With a gaming system, FPS would count a lot, and faster FPS, the better the card. With Quadro's, the cuda cores is what matters. The more the cores, the faster it can "work", if that makes since.

Here's an article, which shows a consumer grade ("gaming" card compared to a Quadro):

http://www.tomshardware.com/reviews/...0,2258-10.html

I can't find the real good article I'm looking for that explains the difference between a desktop card and a Quadro, but here's what nVidia has to say:

https://www.makeuseof.com/tag/best-s...ard-computers/

And this one comes closest:
https://graitec.co.uk/hardware/cad-w...aming-graphics

So, with the Quadro's, you are right, the P2000, for example, would be comparable to the P4000, if we were only planning on using SolidWorks with it, and it wouldn't justify spending the extra cash for the P4000. We would never notice a difference in SolidWorks.

[diif]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

When you're rotating a 3D object FPS is important and how else would you compare one card to another. It's a visual representation of it's performance. I'm well aware of the differences between Quadras and gaming graphics cards.
Have a look at this site https://www.pugetsystems.com/all_articles.php
Plenty of info with benchmarks showing the slight differences between Quadras.

[Spork Schivago]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

So just to clarify a bit more, the questions about the Quadro and PCI-E lanes are because we really want to run SolidWorks in ESXi / vSphere, which can take advantage of the higher cuda cores and SolidWorks would, I'd think, run smoother, in a VM with a P4000 over a P2000. SolidWorks recommends the P Series.

With a workstation with 16 PCI-E lanes, I just don't see how a Quadro, P2000 or P4000, could reach maximum potential, running at 16x, while the M.2 is running at 4x, and a VM would be out of the question. We'd still need to upgrade the server for the other softwares we plan on running, that should be ran on a server, not a workstation....

Generally, software like SolidWorks is ran on Workstations, not servers, so that's why I need to know which would be faster for SolidWorks, the server with 40 lanes, load reduced registered dual rank RAM, 12Gbps SSD SAS drives in RAID0, or a workstation with 16 PCI-E lanes, an m.2, 32GB (4 x 8GB) of 2666MHz DDR4....sorry if I sound like I'm being frustrating. That's not my intentions, and I do appreciate all the suggestions and answers....

But I feel the answer to my questions still haven't fully been answered.

[Spork Schivago]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

When you're rotating a 3D object FPS is important and how else would you compare one card to another. It's a visual representation of it's performance. I'm well aware of the differences between Quadras and gaming graphics cards.
Have a look at this site https://www.pugetsystems.com/all_articles.php
Plenty of info with benchmarks showing the slight differences between Quadras.

Sorry, I didn't know you understood the difference and assumed maybe you hadn't, I misunderstood your point.

Here, let me try to explain differently. Benchmarks for video cards are going to differ depending on the system, right? If we benchmark the P4000 using something like 3DMark benchmark to benchmark a P2000 vs a P4000, on the server, running a ESXi / vSphere, we'll receive different scores, versus running on the i7-8700K, wouldn't we? Especially because there physically isn't enough lanes on that i7-8600K. The video card uses all 16 of the available lanes, thus creating a bottleneck.

Do you see what I mean? Also, it's more than just about the card and SolidWorks. Although SolidWorks generally uses just on core and that core clock speed matters the most, ESXi / vSphere uses multiple cores, the client OS uses multiple cores. The hard drive controller will definitely affect the score of the Quadro, I would think, because with 16 lanes available on the workstation, and the card using all 16, now the CPU has to "pause" and talk to the m.2 which is using 4 of those lanes, whereas the server doesn't.

Although the server won't have as high of a core clock speed, it'll have a heck of a lot more cache, and many more threads, many more PCI-E lanes.

I'd like to see a benchmark of the P2000 and P4000 on a i7-8700K with an M.2 4x, and then of one on a DL380 Gen 9 with a E5-2699v4 or E5-2696v4, same card, in a VM, with the SAS SSD drives. But I can't find anything like that and I doubt we will.

Most benchmarks are about using the same hardware, to give the actual unit under test the optimal conditions, to try and make it as fair as possible.

It's like benchmarking spark plugs and the miles per gallon, but using a straight 4 and then a V8. It's not a viable test. The V8 has a different configuration and even if the different spark plugs are used, there's many other factors affecting the miles per gallon. The straight 4 should always get better miles per gallon because of the setup.

Do you see where I'm coming from?

[diif]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

Because in the real world I doubt you'll tell the difference and without all the hardware to do real testing, nobody can really tell and there's always faster hardware.

It's like trying to decide between two cars, one has a top speed of 170mph the other 172mph. Technically one is faster but who is going to see the difference.

I wouldn't use RAID0 though, I don't see a need.

[Spork Schivago]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

Because in the real world I doubt you'll tell the difference and without all the hardware to do real testing, nobody can really tell and there's always faster hardware.

It's like trying to decide between two cars, one has a top speed of 170mph the other 172mph. Technically one is faster but who is going to see the difference.

I wouldn't use RAID0 though, I don't see a need.

So you wouldn't see an increase in performance using hardware RAID0 with the 12Gbps SAS drives, with read / write intensive applications running along side SolidWorks?


As for ESXi, there'd be no benefits for choosing the P4000 over the P2000, even with multiple VMs? Those extra cuda cores wouldn't matter at all? We know some programs can use them do the hard work.

I guess the last question would be the persistent RAM available from HPE. Max I can buy per chip is 8GB. It's supposed to be fast, but not as fast as the persistent RAM for the G10 servers (the 16GB ones). Use it like a hard drive, but have the speeds of RAM.

Thank you!

[diif]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

What intensive read write applications ?

[Spork Schivago]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

What intensive read write applications ?

It'd be a database that would have the possibility of being queried quite a bit, along side an Apache server, running in another VM. There'd be two identical setups there, one that's production, one that's development, with the virtual machine manager isolating them.

So one VM with 7 Pro running SolidWorks, one two VMs running something like CentOS 7 with Apache, cPanel / WHM, MariaDB, with the website tied into the database, for user account settings, forum messages, support tickets, etc.

There might be more VMs, depending on what the server can handle.

But if the workstation with the i7-8700k with a P2000 is the way to get the most out of SolidWorks, the server wouldn't be running 7 Pro with SolidWorks, that'd be on the workstation, with the M.2 and a P2000.

We still will be purchasing RAM (load reduced) and a CPU with many more cores for the server. Was trying to save some money buy just upgrading the server, instead of having the faster workstation, seeing how what I have now doesn't seem to be cutting it.

[Snayperskaya]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

the server with 40 lanes, load reduced registered dual rank RAM, 12Gbps SSD SAS drives in RAID0, or a workstation with 16 PCI-E lanes, an m.2, 32GB (4 x 8GB) of 2666MHz DDR4

I think you're be going a bit too far on it.

* SAS disks have a pretty high cost, let alone SAS SSDs. Grab a couple high-endurance SATA (or NVMe if you need the extra performance) for the ongoing projects and mechanical HDDs for archiving;

* RAID 0 isn't recommended for any production scenario. If you need throughput, go for RAID10.

* I am pretty sure there are no 8GB LRDIMMs. Those are used for higher densities (32GB and up only IIRC).

GPU comparison:
https://www.pugetsystems.com/labs/ar...rformance-741/

CPU comparison:
https://www.pugetsystems.com/labs/ar...rformance-741/


-----

VMware (or any other hypervisor that matter) don't lock the cores into VMs. You can have multiple VMs sharing the same resources.

https://blog.heroix.com/blog/vmware-...ver-allocation

[Snayperskaya]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

Double post.

[Spork Schivago]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

I think you're be going a bit too far on it.

* SAS disks have a pretty high cost, let alone SAS SSDs. Grab a couple high-endurance SATA (or NVMe if you need the extra performance) for the ongoing projects and mechanical HDDs for archiving;

* RAID 0 isn't recommended for any production scenario. If you need throughput, go for RAID10.

* I am pretty sure there are no 8GB LRDIMMs. Those are used for higher densities (32GB and up only IIRC).

GPU comparison:
https://www.pugetsystems.com/labs/ar...rformance-741/

CPU comparison:
https://www.pugetsystems.com/labs/ar...rformance-741/


-----

VMware (or any other hypervisor that matter) don't lock the cores into VMs. You can have multiple VMs sharing the same resources.

https://blog.heroix.com/blog/vmware-...ver-allocation

You are correct about the RAM, for the server, I meant would one 32GB load reduced dual rank be faster than 4 x 8GB non-load reduced single rank RAM.

Yes, RAID0 isn't recommended for production environments, that is why I was saying RAID10. But because RAID10 is simple RAID0 mirrored, for simplicity sakes, I might have used the term RAID0.

Again, with the GPU comparisons, you guys are solely going on GPU for SolidWorks, where the 12Gbps SSD SAS RAID10, lots of load reduced RAM, 22 core processor, P4000 series GPU would be a bit overkill, to say the least, but the idea was to use the server for multiple virtualized servers, hence the need for very fast performance.

I did not state any of that originally, because I thought this was a simple question and wasn't expecting answers saying you don't need the P4000 or this is overkill for SolidWorks, etc. I was just looking for a simple, could we get SolidWorks to run faster on the server with the Xeon E5-2696v4 or E5-2699v4, rather than running it on a workstation with an i7-8700k, bothing having the P4000 (or P2000 video card).

Here, you can see, depending on the environment and what you're doing, the P4000 can outperform the P2000: http://www.develop3d.com/hardware/qu...ed-iray-pascal

Not that it really matters much. We still plan on running multiple virtual servers on the server, and need to upgrade it. But do we try to upgrade it to run SolidWorks in a VM or do we purchase the parts to build the workstation with the 8700k? Without SolidWorks on a VM, I don't think I'd need a GPU accelerator for the server, but I'd still need one for the workstation, and the P series are much cheaper if they're not for the server.

The server P2000 and P4000 are special in some way and not equivalent to the normal P2000 and P4000's (like the PNY ones). I don't know what the difference is, but of course, HPE recommends running the HPE version. If I recall, it's something with the firmwares.

[diif]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

If you checked the link I posted, you'll find benchmarks in Solidworks. There was very little difference between the different Quadras.
A single point of failure putting everything on one server isn't wise.

[Spork Schivago]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

Yes, I did check your link and seen that. I agree with the single point of failure and that was one of my main concerns, hence the reason for the redundancy for the hard drives at least.

I can have redundancy for the RAM as well (a spare chip), a redundant power supply, but that's about it.

If the CPU goes, everything goes. So there's pro's and con's. The biggest pro would be the money. I can afford, with the current parts in the server, to upgrade the RAM, the CPU, and the video card, getting something that would handle everything I needed, minus the price for vSphere.

Or, I could use all the money I have saved to build a semi-budget build workstation, and then save up all the money again to upgrade the server....

Optimally, it'd be best to have a duplicate server, but I cannot afford two seats for SolidWorks or another dedicated cPanel, which I do like very much (although I'll probably get flamed for saying that).

Tough decision I have to make. And obviously, redundancy doesn't mean no backups.

I'd rather upgrade the server, so I can get started with the rest (VMs, etc). But if that server won't run SolidWorks good in a VM and it'd run much better on it's own dedicated workstation with the 8700k, I'll just have to go that way and save up more cash.

For the webserver, the actual server is the redundant server, but I might setup some sort of load balancer or something, where half the traffic goes to my VPS, half goes to the DL380. I dunno yet. But at least the web server would have redundancy then.

I guess even though SolidWorks Pro runs extremely slow on my current workstation, I could always use that as a backup for SolidWorks if the new system (or upgraded server, which ever way I go) goes down....just move the seat over.

I'm more worried about the guest OSes sharing the card, as Snayperskaya pointed out. I was aware of that originally. I think I have to dedicate the Quadro (whichever one I go for) to just the VMs and use that pass-through or whatever it's called, and have the host OS use the built-in GPU (real cheap one). That's the only reason I would go for the P4000 over the P2000. Because 3 or more OSes would be using it at once...

If I go the workstation route, I can save 400$ or so and just go for the P2000. Then the OSes on the server wouldn't need any Quadro, but I'd still have to purchase another CPU, more RAM, two more SSD SAS drives (for the mirroring).

[Snayperskaya]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

If you passthrough the card to a single VM, you'll probably achieve close to 1:1 when compared to bare-metal (running natively). All servers come with a simple VGA controller that is enough for everything else.

https://www.youtube.com/watch?v=FvcxPufSRNo

[RJARRRPCGP]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

No Core i5, i7, i9 supports ECC RAM, you need a Xeon for that shit for whatever nonsensical reason.

And the strange thing was, Intel's web site reported that the Pentium G3258 supports ECC DDR SDRAM, LOL.

[Spork Schivago]

Re: Anyone wanna help me with a 3,055.00$ semi-budget build?

If you passthrough the card to a single VM, you'll probably achieve close to 1:1 when compared to bare-metal (running natively). All servers come with a simple VGA controller that is enough for everything else.

https://www.youtube.com/watch?v=FvcxPufSRNo

But if I passthrough the card to multiple VMs? Would only the VM using the cores use all those cores? Or would it be divided equally? Where each VM would get an equal share of cuda cores?

I was thinking of using the built-in VGA controller for the VMs that run CentOS 7, and only passthroughing the Quadro to the VM running 7 Pro with the SolidWorks.

Would that work or no?

My understanding, if I do use the passthrough stuff, the host OS won't have access to the hardware anymore. So, I couldn't run hashcat or something on the host to test security keys or something, for instance, using the Quadro, right?

Right now, I'm running KVM on CnetOS 7 with Windows installed, and I'm having an extremely hard time having it recognize the CPU. I try setting it up to just copy the host CPU, but then it doesn't start the VM saying something about missing instructions.

I think this is because of the recent microcode update for the Intel processors and KVM or Spice might not have caught up yet. So I have to pick a much slower processor.