What wattage PSU do I need for this workstation configuration?

[classic35mm]

I'm not an IT person or computer expert. I am considering a Lenovo ThinkStation P300 Tower with the following configuration (Windows 7):

Xeon E3-1241 v3 (8MB Cache, 3.50GHz)
8GB PC3 1600MHz uDIMM
NVIDIA Quadro K420 1GB (DVI+DP)
16x DVD +/- RW DL Win 7 only
2.5" 256GB SATA SolidState Drive
3.5" 2TB SATA HDD 7200 rpm

The P300 is available with these two different power supplies (under the "Form Factor" section on the configuration page):

Tower with Intel C226, 280W 85% [included in price]
Tower with Intel C226, 450W 92% [add $50 US]

Should I get the 280 W or the 450 W power supply?

I asked that question to two Lenovo reps, separately, and I got these answers.

You will want the 450W power supply if you ever plan to make any upgrades such as a higher end graphics card. The 280W is pretty small and cannot handle upgrades much past the base specs. If your configuration on lenovo.com requires more power, it will default to the 450W.

The short answer is that the configurator (online) shouldn't allow you to configure the system with a PSU that is not sufficient to support the configuration. Generally, the 280W is sufficient for most configs. The 450W comes with a 6 pin aux graphics power cable that is capable of supporting up to a K4200 card in the tower. The K4200 card is rated at 108W, so it requires additional 12V power beyond what the PCIe slot provides in order to function.

I'm not sure if those two answers are completely in agreement or not. On the one hand, I would like the option to upgrade down the road (e.g., additional memory, additional hard drives, perhaps an additional optical drive). On the other hand, I do not expect to upgrade the graphics card, at least not to something as expensive as the Quadro K4200.

When I use Lenovo's online configurator and load up a P300 with some top specs (Xeon E3-1281 v3, four 8 GB sticks of ECC RAM, two 4 TB HDDs, two Bluray drives, and even the Quadro K2200), the 280 W PSU is still an option. It's only when I select the Quadro K4200 that the configurator forces me to choose the 450 W. So I'm not entirely confident.

So I decided to try Cooler Master's wattage calculator for the system I mentioned at the top of this post (Xeon E3-1241 v3, 8 GB non-ECC RAM, Quadro K420, DVD RW drive, 256 GB SSD, 2 TB HDD). I think that the Xeon E3-1241 v3 is FCLGA1150, so I selected "LGA1150". Cooler Master doesn't have the Quadro K420 (45 W max) listed as an option, so I chose a card from their list with a similar wattage (Quadro K600: 41 W max). And, "the Oscar goes to ..."

... Cooler Master tells me that I will need a minimum PSU of 272 W. I realize there may be some "fudge factor" included in the calculation, but still, that's awfully close to 280 W, it seems. Another thing is that my existing computer, a desktop from 2007 by another manufacturer, has a 375 W PSU. I don't want to "downgrade," if that's the right word.

So, what are my options? Do you have any advice?
I guess one option is to just go with the 280 W, but wouldn't I be risking unexpected shutdowns and general wear and tear?
Another option is to go with the 450 W, which is probably a lot more than I need. Since it is oversized for my purpose, it might run inefficiently and cost more power from the wall. Are there other risks associated with a too big PSU?

One other question I have is, which option would lead to a hotter (or at least noisier, due to fans running more) machine: a too small PSU or a too big PSU? I'd like to keep the temperature and noise down in the room as much as possible.

Thanks so much for your time and patience.

 

[peche]

you can easily pair that hardware with a thermaltake TR2 430W 80+gold psu, or any 430/450W power supply,

Regards,

 

[Frick]

I'd say go for the bigger one. You'll have something to play with if you decide to expand on something, and PSU's tends to be at their highest efficency at around 50% load, and it'll run cooler and quieter.

And there's a lot more to PSU's than just wattage, but the big OEM's (HP, Lenovo etc) tends to use tried supplies, so it'll last a very long while.

 

[peche]

take this as a reference!

 

[Bill_Bright]

Another option is to go with the 450 W, which is probably a lot more than I need. Since it is oversized for my purpose, it might run inefficiently and cost more power from the wall. Are there other risks associated with a too big PSU?

If you were considering a monster 1000W PSU, then your concerns about running inefficiently with such a low load "might" be valid (depending on the quality of the PSU). But a 450W PSU is not very big so any low load this system would present will not matter in that regard with any significance.

That said, I definitely would get a "Gold" 80 PLUS certified PSU because even at a 20% load, a Gold certified PSU must maintain as a minimum, a very respectable 87% efficiency.

One other question I have is, which option would lead to a hotter (or at least noisier, due to fans running more) machine: a too small PSU or a too big PSU? I'd like to keep the temperature and noise down in the room as much as possible.

Heat is not a factor of PSU size, only the PSU's efficiency. For example, if you put a 200W load on a 280W 80% efficient PSU, the PSU will draw from the wall 240W with the extra 40W wasted in the form of heat.

Put the same 200W load on a 450W 80% efficient PSU, the draw from the wall is the same 240W with the same amount of wasted energy - 40W - again in the form of heat.

Remember, regardless the wattage capability of the PSU, it will only draw from the wall what the connected parts demand, plus what is lost due to inefficiencies.

Noise may be a factor though. 200W on a 280W supply is pushing the PSU to 71.4% capacity and that may cause the PSU fan to spin up sooner and to full speed (and noise level). While 200 on a 450W supply is just 44.4% of capacity so the fan may be able stay off longer and spin slower once signaled to spin up.

Do note too that Gold rated supplies are typically the Cadillacs of the PSU line while Basic 80 PLUS certification, Bronze certified and un-certified generare entry-level Chevys and Yugos. That means the Gold rated supply is likely to have a better quality fan too so even when spinning full speed, it is likely to be quieter. Better quality fans in Golds is certainly not a steadfast rule, but it is a common practice.

Note the TR2 430W is not 80 PLUS certified and is rated at a lousy 70% efficiency at full load only. I recommend avoiding any PSU that is not 80 PLUS certified. This TR2 450W Gold is rated at a minimum of 87-92% efficiency from 20-100% load. There is a TR2 450W Bronze too with 82-85% efficiency rating for $10 less. Bronze is good, Gold is better.

 

[peche]

Note the TR2 430W is not 80 PLUS certified and is rated at a lousy 70% efficiency at full load only. I recommend avoiding any PSU that is not 80 PLUS certified. This TR2 450W Gold is rated at a minimum of 87-92% efficiency from 20-100% load. There is a TR2 450W Bronze too with 82-85% efficiency rating for $10 less. Bronze is good, Gold is better.

my mistake ...
http://www.thermaltakeusa.com/Power_Supply/TR2_Series_/TR2/C_00002575/TR2_450W_Gold/design.htm
you are completely right!

 

[classic35mm]

Thanks for your time, everyone! From your advice it seems that I should go with the 450W 92% rather than the 280W 85%, and that everything should be fine.

 

[Bill_Bright]

my mistake ...

Easy mistake to make when companies like TT use very similar or even the exact same model number for totally different PSUs. I note on Newegg, TT has 4 different 500W supplies, all in the "TR2" series model line. Two un-certified, one Bronze and one Gold.

 

[Bill_Bright]

From your advice it seems that I should go with the 450W 92% rather than the 280W 85%, and that everything should be fine.

I don't think you will be disappointed.

 

[newtekie1]

Go with the 450W power supply, it has a 6-pin that allows for graphics cards upgrades. That is why the K4000 requires the 450W power supply, the 450W has a 6-Pin power connector the 280W doesn't.

But like I said in your other thread, get the i7-4790 version without a graphics cards with the 450w power supply. Then add a GTX960 yourself.

you can easily pair that hardware with a thermaltake TR2 430W 80+gold psu, or any 430/450W power supply,

Note that you can't put a normal PSU in this computer. The motherboard use a proprietary connector.

 

[Bill_Bright]

Note that you can't put a normal PSU in this computer. The motherboard use a proprietary connector.

Where do you see that?

I do see on this page that it uses a "fixed power supply". I realize this is a "workstation" and they do often have some proprietary aspects to them, but "fixed" power supply does give me some concern. That can certainly mean more expensive (if possible) repairs or replacements (and limited upgrade options) a few years down the road.

Maybe we need to back up.

@classic35mm - why are you looking at a "workstation" versus a standard PC? What will you be doing with this computer?

 

[peche]

Easy mistake to make when companies like TT use very similar or even the exact same model number for totally different PSUs. I note on Newegg, TT has 4 different 500W supplies, all in the "TR2" series model line. Two un-certified, one Bronze and one Gold.

Tt uses to recycle names on several items, there are like 3 commander case models that are basically the same one...

 

[Bill_Bright]

Well, TT is not alone in this but I still find it irritating, and often confusing.

 

[classic35mm]

@classic35mm - why are you looking at a "workstation" versus a standard PC? What will you be doing with this computer?

Yes, please, let's back up. I will be using the system for normal business use, HD video streaming, and some scientific simulations. My scientific simulations will mostly use Matlab and Mathematica, as well as my own home-grown Fortran and C/C++ codes. I want a system with 4 cores / 8 threads, but don't need ECC memory.

Actually, that is one of my other questions: what is the difference between a desktop and a workstation? I've asked around here and there (although not on this forum, at least not yet), but I've gotten mixed messages at best.

From what I understand, workstations are generally more expandable. They support Xeon processors in addition to the Core i series. Because they support Xeon processors, workstations can handle ECC memory, more than 32 GB of memory, and dual processor configurations.

Lenovo's desktop line is ThinkCentre. Lenovo's workstation line is ThinkStation. Since both lines support Core i series processors, it's possible to configure a ThinkCentre (M83 or M93p) with virtually the same specs an entry-level workstation (P300). The only difference is the video card; the ThinkCentre ships with a NVIDIA GeForce GT 620 1GB card, whereas the ThinkStation ships with a Quadro card. Other than that, the specs are pretty much the same (each has Core i7-4790 processor, 8 GB of non-ECC memory, and two hard drives: one 256 GB SSD and one 2 TB HDD). Interestingly, their prices are nearly identical, with the desktop being actually slightly more expensive than the workstation (by less than ~$100 US).

In the situation of very similar specs and very similar prices, what is the difference between a desktop and a workstation? How do I choose one over the other?

The ThinkCentre's GeForce GT 620 seems to be rather outdated, and not very advanced for a video card. That's one reason that I'm considering the ThinkStation. Another reason is that, since I'm considering a discrete video card, my thinking is: why pay extra for a Core i7, which has integrated graphics that I wouldn't be using? Why not go with a Xeon that doesn't have integrated graphics so is a little cheaper? But do you think this is faulty reasoning?

If a workstation has a different type of motherboard than a desktop, will I be restricting myself in terms of parts down the road if I go with a workstation?

 

[GorbazTheDragon]

Actually, that is one of my other questions: what is the difference between a desktop and a workstation? I've asked around here and there (although not on this forum, at least not yet), but I've gotten mixed messages at best.

Workstation typically refers to more rigorous hardware verification. For graphics workstations it can also mean 10 bit per pixel for better colour depth. Often workstation parts are closer to server parts as they go through more validation and are built for reliability rather than typical desktop features such as overclocking. ECC memory is also part of the increased reliability.

From what I understand, workstations are generally more expandable. They support Xeon processors in addition to the Core i series. Because they support Xeon processors, workstations can handle ECC memory, more than 32 GB of memory, and dual processor configurations.

This is where they start to overlap with servers. IF you need the CPU performance this is what allows the workstation to work under many different loads. Some programs require GPU power while others CPU, some need both.

Lenovo's desktop line is ThinkCentre. Lenovo's workstation line is ThinkStation. Since both lines support Core i series processors, it's possible to configure a ThinkCentre (M83 or M93p) with virtually the same specs an entry-level workstation (P300). The only difference is the video card; the ThinkCentre ships with a NVIDIA GeForce GT 620 1GB card, whereas the ThinkStation ships with a Quadro card. Other than that, the specs are pretty much the same (each has Core i7-4790 processor, 8 GB of non-ECC memory, and two hard drives: one 256 GB SSD and one 2 TB HDD). Interestingly, their prices are nearly identical, with the desktop being actually slightly more expensive than the workstation (by less than ~$100 US).

In the situation of very similar specs and very similar prices, what is the difference between a desktop and a workstation? How do I choose one over the other?

Depends on if you need features that the Quadro card provides. Otherwise they are trying to rip you off in typical pre-built fashion.

The ThinkCentre's GeForce GT 620 seems to be rather outdated, and not very advanced for a video card. That's one reason that I'm considering the ThinkStation. Another reason is that, since I'm considering a discrete video card, my thinking is: why pay extra for a Core i7, which has integrated graphics that I wouldn't be using? Why not go with a Xeon that doesn't have integrated graphics so is a little cheaper? But do you think this is faulty reasoning?

If a workstation has a different type of motherboard than a desktop, will I be restricting myself in terms of parts down the road if I go with a workstation?

Most of the Xeons for LGA1150 have integrated graphics, keep in mind that this can be handy if for any reason you need to replace your GPU. Also, you can sometimes run more monitors by connecting some to the iGPU and others to the dGPU.

I don't really know what kind of workloads you will be experiencing in those programs. If you are going to be relying on a lot of CUDA you will probably want to throw some extra money in the GPU, the ones you mentioned are really measly in performance, to the point they are pretty much a waste of money when compared to an iGPU. Depending on if you are going to do FP64 or FP32 you will need to chose between a Quadro and Geforce card.

If you are comfortable with the work required to build and maintain a computer I would highly suggest buying parts individually.

 

[newtekie1]

Where do you see that?

The factory service manual.

 

[classic35mm]

Workstation typically refers to more rigorous hardware verification. For graphics workstations it can also mean 10 bit per pixel for better colour depth. Often workstation parts are closer to server parts as they go through more validation and are built for reliability rather than typical desktop features such as overclocking. ECC memory is also part of the increased reliability.

I just find it odd that a workstation with the same specs as a desktop (except for the video card) is actually slightly less expensive than the desktop. If the workstation is likely to have more rigorous hardware verification, then I'll probably head in that direction. But I just don't want to be somehow constrained down the road as far as repairs/replacements/upgrades if I choose the workstation over the desktop.

Depends on if you need features that the Quadro card provides. Otherwise they are trying to rip you off in typical pre-built fashion.

In terms of specs, the Quadro appears to be marginally more capable than the GeForce that they are offering. But what do you mean by "Otherwise they are trying to rip you off in typical pre-built fashion."? Being ripped off is really no surprise -- I'm sure it would be cheaper for me to build my own, but I don't have the time or know-how to do that. But since the desktop and the workstation are nearly the same price for nearly the same specs, aren't I being ripped off equally with the desktop and the workstation? Or what do you mean?

Most of the Xeons for LGA1150 have integrated graphics, keep in mind that this can be handy if for any reason you need to replace your GPU. Also, you can sometimes run more monitors by connecting some to the iGPU and others to the dGPU.

That's a very good point; I never thought of that.

If you are going to be relying on a lot of CUDA you will probably want to throw some extra money in the GPU, the ones you mentioned are really measly in performance, to the point they are pretty much a waste of money when compared to an iGPU.

But, to do CUDA, I need a discrete GPU, right? My understanding is that Intel's iGPUs do not support CUDA, at least not right now. Is this correct?

Depending on if you are going to do FP64 or FP32 you will need to chose between a Quadro and Geforce card.

I guess GeForce cards generally support FP32 whereas some Quadro cards support FP64?

 

[dorsetknob]

What are you expecting to outlay money wise for your intended purchace

If the members here know an approx budget then they can sugest parts for a build

you could then take the list to some local vendors to see if they could build a system for a competitive price
Local built ( and sourced ) and maintained may be advantageous to you

also you have not mentioned your visual requirements as in display monitor size/ resolution

 

[newtekie1]

Most of the Xeons for LGA1150 have integrated graphics, keep in mind that this can be handy if for any reason you need to replace your GPU. Also, you can sometimes run more monitors by connecting some to the iGPU and others to the dGPU.

Actually, most 1150 Xeons don't have integrated graphics. Some do, but the majority don't.

But, to do CUDA, I need a discrete GPU, right? My understanding is that Intel's iGPUs do not support CUDA, at least not right now. Is this correct?

Yes, you'll need an nVidia GPU to do hardware accelerated CUDA. But, like I said in your other thread, a custom build from NCIX would be a much better option. Mainly because you can get a GTX960 for the same price.

why pay extra for a Core i7, which has integrated graphics that I wouldn't be using?

The integrated graphics are not why the i7 is more expensive. The i7-4790 is more expensive than the E3-1241 because the i7-4790 is faster.

 

[Caring1]

The simple answer is, you don't need a 450W PSU if you don't plan on adding a more powerful Graphics card.
The original configuration is fine, and there are Graphics cards that do not need additional power if you do want to change that.

 

[Frick]

Yes, you'll need an nVidia GPU to do hardware accelerated CUDA. But, like I said in your other thread, a custom build from NCIX would be a much better option. Mainly because you can get a GTX960 for the same price.

I'd say that depends on the OPs location, and the warranty/service deal he gets with the Lenovo. Part of why these machines cost more (as you undoubtedly know) is warranty and service plans. If he gets Lenovos pick up service or field repair the same day and does not have a local shop, go Lenovo. If he does have a local shop with decent pricing, go with them. If you have a high end machine and you need it RMAing is a true bitch.

 

[GorbazTheDragon]

I just find it odd that a workstation with the same specs as a desktop (except for the video card) is actually slightly less expensive than the desktop. If the workstation is likely to have more rigorous hardware verification, then I'll probably head in that direction. But I just don't want to be somehow constrained down the road as far as repairs/replacements/upgrades if I choose the workstation over the desktop.

If you chose something built by a company like dell/lenovo etc, you will likely be constrained with down the road upgrades as many don't adhere to standards (they want you to stick with their stuff) Take the power connector as an example.

The difference between the i7s and the Xeons is small on LGA1150, and as I was corrected, there are certain Xeons with iGPUs, many/most don't have. If you look at the MSRPs the ones with iGPUs will cost basically as much as the i7s.

In terms of specs, the Quadro appears to be marginally more capable than the GeForce that they are offering. But what do you mean by "Otherwise they are trying to rip you off in typical pre-built fashion."? Being ripped off is really no surprise -- I'm sure it would be cheaper for me to build my own, but I don't have the time or know-how to do that. But since the desktop and the workstation are nearly the same price for nearly the same specs, aren't I being ripped off equally with the desktop and the workstation? Or what do you mean?

Usually similar specced (by amount of cores) Quadros are significantly more expensive than Geforce cards. Compare K5000 vs GTX 680 for example. I guess this does taper off a bit at lower end cards, but the K420 should fall into the performance bracket somewhere around the GT630. The price of the K420 is slightly higher however.
You are kinda getting ripped of by nvidia on the Quadros, as they are pretty much selling the same chip, maybe with a few minor changes for a lot more, but that's beside the point.

But, to do CUDA, I need a discrete GPU, right? My understanding is that Intel's iGPUs do not support CUDA, at least not right now. Is this correct?

I guess GeForce cards generally support FP32 whereas some Quadro cards support FP64?

Intel and AMD GPUs don't support CUDA (at least not officially). For FP32 vs FP64, all Quadros should support FP64 (double precision), but Maxwell cards are much less powerful in FP64 compared to Kepler/Fermi cards due to the design of the architecture. There are other features on Quadro cards, which I don't know off the top of my head that are useful in certain scenarios, but the choice between Quadro and Geforce comes down to exactly what you will be doing, you will need to evaluate that yourself as I don't really know what you are planning to do in Matlab/etc.

In the case you are going to be doing detailed 3D rendering or physics simulations you will probably want something more powerful than the GT620 or K420, probably a GTX 750 Ti or 960 would be better from the GeForce side, and for Quadro cards you would be looking more in the direction of the K2000. (here is where you start to see the quadro prices take off...)

I would personally say build it yourself, it will give you more flexibility as far as the configuration and potential upgrades down the road. However this of course comes at the cost of your time

 

[Frick]

You are kinda getting ripped of by nvidia on the Quadros, as they are pretty much selling the same chip, maybe with a few minor changes for a lot more, but that's beside the point.

Well that sorta depends. If you it's not critical or something they are way to expensive, but if you're doing something more or less critical you'll want the support.

 

[GorbazTheDragon]

Yeah, but they lock out features from the Geforce that cost them hardly extra cost to implement. You of course could consider that there is extra support when you get a quadro card. Yes, you are paying for that, but the only reason NV can charge that much extra for basically the same card is because companies/researchers are willing to pay more than individual people. Most Quadro cards are sold to large orgs for their employers, while most Geforce cards are purchased by individual people as end users.

 

[dorsetknob]

Yeah, but they lock out features from the Geforce that cost them hardly extra cost to implement. You of course could consider that there is extra support when you get a quadro card. Yes, you are paying for that, but the only reason NV can charge that much extra for basically the same card is because companies/researchers are willing to pay more than individual people. Most Quadro cards are sold to large orgs for their employers, while most Geforce cards are purchased by individual people as end users.

your also paying for proper certified drivers of a higher Mission critical nature
AMD are no different with their Workstation Cards

Workstation Cards by the very nature of their Mission critical nature are always more expensive due not to the silicon involved but DRIVER SUPPORT