Please could someone help with speccing an upgrade

[danielphillips1]

Hi all,

 

I would like to improve my home pc, and turn it into a machine that would be dedicated towards single frame still renders (not animation fly throughs), for architectural visualisation, 3DS max and Vray.

 

I have up to £2000 to spend in upgrading my rig.

I assume I focus on new MOBO, CPU and Ram?

 

Would I see much of difference between a 10 core and 12 core cpu?

Should I bump the RAM to 64gig?

 

How much of a speed impact would going from my old generation quad core, to a 10 or 12 core cpu be? Would we only be talking fractions of a second, or large boosts?

 

My current spec is the following;

Gigabyte GA-EX58-UD5 Intel X58

Intel® Core™ i7 965 3.20GHz Extreme

Corsair XMS3 12GB (6x2GB) DDR3 1600MHz

1x Samsung Evo SSD - 250GB

1x Samsung Evo SSD - 500GB

2x 500GB 7200RPM SATA II HDD

Nvidia Geforce GTX 780 ti

Creative Sound Blaster X-Fi Titanium.

 

Regards,

Daniel.

[Tommy L]

Add another 12G RAM and your workstation is fine. Buy a dedicated node for rendering, you could also buy a second graphics card for your workstation to better make use of GPU rendering for previews.

[danielphillips1]

I really appreciate your reply,

 

Could you please link me to a dedicated node?

In terms of Ram, I currently have 6 sticks of 2 gig, should I change these to 6 sticks of 8 gig?

[danielphillips1]

In terms of what I was initially looking it, here is the list and rough pricing (using overlockers as an example)

 

CPU - Intel Xeon E5-2670v2 2.50GHz 10-Core with Hyperthreading & Turbo (Socket 2011) - Retail

 

MOBO - Asus Rampage IV Extreme Black Edition Intel X79 (Socket 2011) DDR3 Motherboard

 

RAM - Kingston HyperX Beast 64GB (8x8GB) PC3-17100C11 2133MHz Dual Channel Kit (KHX21C11T3FK8/64X)

 

Bring it to a total of - £2,089 (I could probably shop around to bring this down to £2k)

 

What are ones thoughts on this? How will that improve rendering times? Is it worth it?

 

Thanks

Dan.

[jasonstewart]

Should be a good upgrade, CPU is almost 3x faster in Passmark vs. your current CPU. For 2k I would redo your workstation with 32gb of ram (unless you actually need the extra ram there is no point) and a 4930k (almost the same performance as the E5-2670v2 but for 1/3 the price). Then I would take the extra money I saved and build the "Almost $500 render Node" that Dimitris came up with:http://pcfoo.com/2013/02/8350-rendering-node/. Passmark wise that should get you rendering about 4x faster, may not be directly scalable in your rendering engine but would for sure be a big upgrade.

 

In the future you could add another node to increase speed even more.

 

Unless you want to extreme overclock (there is no point with Xeon since they are locked) that Mobo is IMO a waste of money. Check out the x79 deluxe, pro, or sabertooth all great 2011 boards.

 

That kingston ram is really expensive, check out Gskill great quality at very affordable prices.

Edited August 15, 2014 by jasonstewart

[danielphillips1]

Thanks Jason

 

So to sum up, is this correct;

Intel 4930K 3.40GHz

Something like - G.Skill RipJawsZ 32GB (8x4GB) DDR3 RAM

Asus X79 DELUXE Intel X79 (Socket 2011) DDR3 Motherboard

 

I was looking on various hardware sites, and I pieced these upcoming hardware together;

8Core, 16Thread i7-5960X when its released - Im assuming £1000 is the estimated price

DDR4 ram when its available - Dependant on pricing - 24 gig

X99 mobo when they are released

Corsair H100i to cool the cpu.

 

Is it still best to go with what you mentioned Jason?

Edited August 15, 2014 by danielphillips1

[Mateusz Gawad]

Hello, haven't written on the forums before but this topic just comes back a lot, so i guess it's a good time to drop my two cents as well.

 

First of all hi to everyone, and really you guys put a lot of stuff and effort around here, so i'll be glad to contribute as much knowledge as i have on the subject.

 

Recently there is a lot of hype on who is working on what and how to get the most juice of the specs. We live in such a wonderful time that there is just an option for every project and workflow that you can imagine. Personally i see a trend of getting away from how to work and going straight to the question what is the best hardware setup.

 

You have already determined on what kind of programs and subjects you are gone use your PC, but i want to ask you a question at this point: How many times a month your current platform is failing to deliver an expected performance in this area of work ? And by expected performance i mean how often your scenes are not fitting in your ram, do you find yourself rendering too long and missing your deadlines, your PC is slow in moving your files ?

 

I'm asking just to make you think about is it really necessary to update right now, the thing you posted, the upcoming hardware, would be better in a long run, and personally i would wait until then. The octa core will be a really nice thing for these kind of tasks, and a lot cheaper than xeon. Also you would go into ddr4 probably so its a nice thing as well and it will last longer.

At the moment i do wait actually for almost exact setup that you mentioned, and i'm currently on almost the same platform as you are. My work is a heavy loaded freelance, and i must say that i don't have any problem in delivering stuff, exteriors, interiors, they all fit in around 10gb ram, and they are heavy in scatters and high poly (unoptimised scene i had to work on was 24 mil and it went well after some tweaks). Also if you can afford rendering overnight, there is no problem, in my case i can deliver something around 6k and it will render through.

 

So it's quite a long post just to point that out. sorry for that First time poster.

And don't think i'm saying all those things just to discourage you or something, because there are a lot of ways to speed up your work that are not necessarily connected to the box under your desk. I'm telling you this because i made my living out of this for a while and it was like somebody just opened my eyes to all the possibilities of speeding your work. I saw a lot of companies that were upgrading like crazy, and would not listen to any argument, because they thought that was the golden rule for super speed and awesome work. And it's mind blowing how much of an actual speed you gain, when you rethink your approach and focus on actual work ethics and your approach, and setup of your work place.

 

So if you are interested in just upgrading to get the best out of hardware, i would wait for the octa core and new ram, to get the most for the money. Your setup will handle it, maybe just update your RAM, the advice on it was pretty good, and don't buy the most expensive stuff.

 

But if you would like to work faster and better i would be more than happy to share some tips and advice on some other stuff you can buy and setup. Or if any one is actually interested in this. I would be glad to share what i learned, and maybe some one would find it useful.

 

 

sorrrry for the loooong post, be well!

[RyderSK]

5960X will definitely make you happy when rendering :- ).

DDR4 might not so much, but if the price difference will be small, why not ? But go for 32GB (4x8 modules, quadchannel, possible to upgrade to 64).

 

Trash the H100i cooler :- ) Just because it's water doesn't mean it's good. It's popular, because it's relatively cheap for what it is. It also has cheap, loud fans. Either have small custom loop done by builder, or just go for high-end air-cooler like Noctua NH-D15. It will be just as powerful, with less hassle, and will end up being more quiet all around.

[Dimitris Tolios]

Also, don't overspend on overclocking motherboards to use with Xeons that don't overclock or i7 that you won't o/c altogether, or even drop all this $$$/€€€/£££ for "deluxe" or "WS" versions and the whole mambo jumbo with massive power supplies etc etc.

 

 

Single CPU non overclocked with powerful GTX/Radeon = good 500W supply does it. Doesn't matter if it is 10C or 12C or 6C. All are speced below 150W, will actually be working @ 120W or lower even @ 100% load. GPUs are speced below 250W 99% of the cases, and will VERY rarely hit that.

 

 

Dual CPU / Multi GPU, overclocking etc etc changes the picture, but be real: you are probably thinking that a $200-250 motherboard won't do good a $600-1,500 CPU, but you are wrong. Yes, $100-200 and there do add up, most of the times for no reason.

All X79 boards are within 1-2% of the stock performance - with the more expensive ones not necessarily being the fastest as all those extra controllers for USB/SATA etc that you won't use, just add to the overhead of the system.

 

 

With PSUs, we have to understand that the goal for most companies & R&D departments is for everything to be more and more efficient, despite being faster - not the other way around. When you are not rendering, all but one core and 1/2 threads are idling, your GPU is mostly idling, and when either CPU and GPU are asked to do something while you model, it happens for a very short burst and then you go back to idling. The average consumption of your system - while you are working - is 20% or less than its "peak load"

 

- Disclaimer: guesstimating here, it can be 10%, it can be more - but most s1150 systems for example will be pulling

 

Same for air-cooling: Noctua and Thermalright dual towers are like 100% more powerful in removing heat vs. what is used in industrial / server applications for systems that will be running 24/7 in 1/3 or less the space of you average Midi tower. You are safe. CLC is just for looks, or very tight situations (ITX cases maybe) where dual tower air-coolers don't fit, while at the same time you can route tubes & rads a tad further and will fit.

 

 

I also don't care for 780Ti and/or Titan etc cards for viewport acceleration - only if you game on the side.

A 780 is already maxing out what most viewport engines in popular modelers can provide - perhaps limited to VRAM only in certain very specific applications that again are not ArchVIZ related. If you need serious work in OpenGL apps like solidworks, than a $700ish Firepro or Quadro (usually the Firepro is better value/performance in that) will beat hands down any $700-1000-3000 GTX or Radeon.

 

 

We often go for blink and not just essence/performance - I know I do - but we should do so in an informed way.

Edited August 15, 2014 by dtolios

[danielphillips1]

How would you go about speccing a high end rig for speeding up rendering Dimitris?

[Dimitris Tolios]

How would you go about speccing a high end rig for speeding up rendering Dimitris?

 

 

Cost is the bigger factor for "high-end". You can always get more performance dropping more $$$, the thing is where the $/returns starts making little sense.

 

 

Pretty much anything s2011 and with something more than a GTX 760 / Quadro K2000 can be thought of as "towards-the-high-end" - performance wise.

But you can get a decent s2011 i7-hex starting around $1,500 for the tower, and go all the way to $5K and more, depending on GPU choices or going 2P Xeons.

 

 

Soon the 5xxx Haswell i7s will be in the picture too, with the "5960X" or whatever the $1,000 range 8-core flagship will be called adding decent gains for $400 over the current 4960X which is probably not a wise investment over the $600ish 4930K that is the vanilla s2011 workstation CPU atm.

 

 

So depending on if you will drop all your monies on one basket - i.e. an all-in-one workstation + rendering machine or a workstation + nodes, the answer varies.

 

 

4930K + Asus X79 (doesn't need to be, but think those had the best / more stable bios when overclocking, so I would assume all-around) + 32/64GB RAM would be my suggestion for a budget "high end" workstation. I would not go 10C/12C Xeon in a 1P system, as I would prefer more grunt per core (higher clocks) that is only available in 8C or lower CPUs.

 

 

If you really want to squeeze as much as you can in a 1P system - without factoring O/C and with what's available today - the fastest CPU atm is the E5-2687W v2: 3.4GHz 8C, but $1,500+ just on the CPU or the slightly cheaper E5-2667 v2 . Yes, you can get all the performance out of it with a $200 MoBo. Basically that buy's you close to the performance you will be getting with the i7-5960X @ stock speeds by the end of Fall~Xmas for $1,000. DDR4 might ruin the "value" of this proposition, but depending on the money you want to spend, it might worth the wait.

 

 

If I really wanted the best all-in-one solution available today, I would probably go for a 2P with either a pair of E5-2687W v2 or the slightly cheaper E5-2667 v2. Ofc then we are talking C602 chipset dual s2011 boards.

 

 

If I wanted a great WS and 1-2 powerful nodes that I would extract the most rendering power out of without spending additional money for extra Vray licenses (basic plan allows WS+2 nodes), would probably be 1-2 2P systems, based on E5-2680 v2 or similarly a good mix of high-speed, 10C CPUs. The 12C option is far more expensive for the added aggregate of cores*Ghz added.

 

 

RAM: if you really need 16GB on your 1P system, you will need 32GB in a 2P. If 32GB are needed (cause you break 16-24GB, not that you fill up 32) on your 1P system, a 2P system might need 64GB, as there is no seamless RAM sharing due to each CPU having its own memory controller. It's not exactly 2x separate systems with 32GB each - like dual GPUs are that falsly advertise a 2+2GB card = 4GB card or 6+6GB card = 12GB, but to be safe, you could think of it as such.

 

 

The nodes need no GPU to speak of.

For the WS, the GPU choice depends on the programs you are using.

$250+ GTX cards (e.g. GTX760 or better) cut it pretty well with current Autodesk products - Revit / Maya / 3DS etc. Might even be faster than any quadro sub K5000 in most cases.

Issue with viewports is, that for the most part there is quite a bit CPU overhead in viewport performance = the CPU is pre-defining what it is for the GPU to draw, and this is a single threaded task = having a 10C or 4C CPU won't make a difference. In reality the faster clocked, newer architecture 4C like the 4770/4790 line will be a tad faster in pure viewport performance than any s2011 CPU mentioned above - regardless of cost. Once you hit GTX 760 performance levels, the GPU is pretty much putting out frames faster than the CPU can prepare it for it. So going for a $500 780 or more expensive 780Ti or Titan GPUs has very diminishing to no returns.

 

 

Radeon R9 GPUs are also very very good. I just fixate with NVidia due to mediocre Vray RT compatibility using AMD cards that appears to be amended with the latest 3.0 versions - yet I cannot guarantee it.

 

 

Again, there will be new players with the 8xx GTX line being out next couple of months, probably giving you better than 780 performance, with less watts, for $400-450 (depending on availability).

 

Sorry for constantly being long-winded, I am just trying to save you some disappointments in the CPU/GPU department, typing as my stupid non-high-end-HP Z420 4C Xeon rig renders that lobby in VRfS. =)

Edited August 15, 2014 by dtolios

[numerobis]

I think i will build my new workstation in october... if the new parts are available then - Haswell-E has been announced for August 29 now.

 

- 5960x watercooled (by my aquacomputer Cuplex Kryos Delrin/ MoRa 3) - i hope that 4.2-4.3 GHz will be possible.

- Maybe ASRock X99 WS with Ultra-M.2 (PCIe 4x) ...but i'm waiting for the ASUS announcements, maybe they will also feature a 4x PCIe M.2 slot

- A good 256GB M.2 PCIe SSD - hopefully there will be some more available until october besides the Samsung XP941 - The Intel PCIe Solid State Cards are a bit expensive..

- 32GB or 64GB RAM. Looking at the prices i think i will start with 32GB.

- Maybe one or two GTX 880 8GB when it comes out (Maxwell Render seems to support CUDA in the near future! presented on Siggraph ) - or a 870 when it comes with 8GB too, but i don't think so.

[danielphillips1]

Can I have some feedback on parts that I was thinking of;

 

CPU - 8Core, 16Thread i7-5960X

MOBO - MSI X99S - http://www.guru3d.com/news-story/msi-x99s-sli-motherboard-photo.html

RAM - Crucial Ballistix Elite 32GB (4x8GB) DDR4 3000MHz - http://www.overclockers.co.uk/showproduct.php?prodid=MY-183-CR&groupid=701&catid=8&subcat=2557

Is 3000mhz overkill? How important is the rams speed?

 

My current graphics card is - Gigabyte GeForce GTX 780Ti WindForce 3x 3072MB GDDR5.

Should I get another one of these, or look at something different? What is recommended for GPU rendering?

 

Thanks

Dan.

[Dimitris Tolios]

Don't over-spent to get fast(er) RAM. DDR4 is already a tad faster than DDR3 in the same clocks, and the Haswell Architecture has little to no notable benefit going DDR3-2133 in most applications - definitely no notable benefit in ArchVIZ applications, Adobe etc.

 

 

DDR3-2400 is clear overkill, DDR4-3000 I would think is even MOAR overkill!

 

 

Mobos: I would just pick the company I like/trust the most if I wasn't overclocking. In my case that would be Asus & Gigabyte, with Asrock 3rd and MSI following. If I would plan on overclocking the 59xx CPU, I would "wait and see" which company did the best job in the BIOS + power delivery design to achieve the most stable clocks with the less additional voltage - e.g: For s2011 that was Asus, for s1150 was Gigabyte (afaik).

 

 

Also note that first impressions state that with current BIOSs, you cannot overclock the CPU without the onboard ram controller having issues in anything but 2133 or so speeds, so again, wait and see before you invest lost of money in RAM - having it running @ 3000 an not giving you real benefits, is better (at least you keep your bragging rights = epeen) than not even running it @ 3000!

 

 

GPU: nothing really faster out yet. I would hold on it, unless VRAM was a limitation.

I would in general go for 2x 780 6GB than a 780Ti (3GB only) or a single Titan Black if I was buying something now from scracth, or I would wait for the new Maxwell based 8xx cards that will probably out-perform the 780/Titan/GK110 cards, and probably launch with 4 & 8GB DDR5 256bit configs in the $500ish range (for the top of the line, mid range model - e.g. "880") and less for lower binned models ("870" etc). I would expect 2+ "870 8GB" (not sure of the actual name and launch date for the 8GB) will be the best value for a high end system, costing the same or less than a Titan Black today, with the 4GB versions already outperforming any GK110 card.

 

 

personally, I've sold my Titan and holding on my GTX 750Ti "spare" till 8xx cards drop in the market. Will take a couple more months for the market to settle & waterblocks to emerge, and then... =)

[danielphillips1]

Thanks for the heads up on the ram, i will not go over 2400 then.

Does the amount of sticks to make up a set, make any noticeable impact? For example, 32gb - is it better to get 1 set of 4x8gb (on an 8 slot mobo) or fill it with two sets of 4x4gb of ram.

 

So you would recommend holding out for the 870 gtx, and getting two?

[Dimitris Tolios]

You have the best GPU of this generation, no need to upgrade it after a few months.

If you need more speed, add a card. Doesn't have to be SLI with another identical to work for GPGPU rendering (if that's the goal).

 

I sold my Titan cause I could get out all the money I had invested in it - actually I made a good profit, otherwise I would stay with it.

 

I think the minimum possible configuration of sticks is better. Adding more sticks - in some occasions - causes compatibility issues.

At least that was the case with X79 boards that were a bit particular on which sticks they would allow to work @ high speeds in 8-stick configurations. No idea if that was a Mobo issue or more of a memory controller = CPU issue, but was a reality for many people who frustrated had to return mobos and/or memory trying to get it to work.

 

Again, more often that not there won't be an issue, but I would definitely go 4x8GB instead of 8x4GB, even if I had no aspirations for adding more than 32GB.

[danielphillips1]

What would be the trade off, for going M-ATX over ATX,

Evga revealed a M-ATX http://tech4gamers.com/evga-x99-micro-matx-motherboard-revealed/

 

So you dont need to run duplicate cards in SLI for gpu rendering, if one is to add another card?

Edited August 23, 2014 by danielphillips1

[Dimitris Tolios]

What would be the trade off, for going M-ATX over ATX,

Evga revealed a M-ATX http://tech4gamers.com/evga-x99-micro-matx-motherboard-revealed/

 

You won't have 8-dimm option, and your overclocking ability will probably be a tad lower than that of the top of the line ATX/E-ATX offerings - simply because the developers won't have the room to spare for better / larger power delivery circuits. That doesn't mean you cannot OC great: i am pretty sure the mATX ASUS X79 Gene ROG would OC just as good as my P9X79 Pro, and I do 4.9GHz on my 3930K, so plz...Its the RIVE that is better, that's all.

 

And ofc you will have less PCIe slots - again, I use only 1, for a single card on mine, unless I am toying with a FirePro or Quadro.

 

So you dont need to run duplicate cards in SLI for gpu rendering?

You don't "need" SLI for anything.

In GPGPU, each card needs to be compatible with CUDA/openCL (all GTX out there are), and if it is, it will be recognized and contribute.

Any combination of models, any number of cards.

The only limitation is that each card needs to fit the data set to be computed individually. Same is true for dual-gpu cards, which are practially two different cards on one PCB. VRAM is split exclusively, and not share between the two. Both in gaming/viewport acceleration and GPGPU. Thus it is a lie to advertize a R9-295x2 card as a 8GB card, or a GTX Titan Z as a 12GB card. Those are 2x4 and 2x6 cards, as each GPU can only have access to its own memory banks, even if the 2nd GPU onboard is idling (what is happening in 3DS for example).

[danielphillips1]

Thanks for the great and informative replies.

 

In terms of cooling, what would be recommended, if one had more than one card in the rig.