New Computer Build for Max

[PBGV103 729]

Hey guys some of you may remember me mentioning starting a computer build before. Now it's actually going to happen, but I'd like to get a little advice from someone with experience first.

Currently I have a core 2 duo 2.8GHz (T9600) MacBook Pro that's almost 3 years old. I'm running SC4 and Max 2011 on Win7 and haven't had too many issues until the last couple months. With my current project in Max I'm using almost 7 of 8GB of memory during renders and have been having frequent bsod and freezing issues.

My plan is to replace it with a Xeon system, but as they are incredibly expensive, I'd like to have some input from you guys. Below are the parts I'm looking at. As for a budget I'd like to stay around $2000 USD. I realize I could go with something cheap like an i5-2500k build and cut the cost down, but I'm worried that I won't see the multithreaded performance increase of a Xeon and I wouldn't have many options for future upgrades other than replacing the entire system. With a dual socket Xeon board I can start out with one midrange CPU and replace it or add a second one later. Still, my concern is that at a maximum of 2.8GHz the Xeon I can afford right now will be the same speed as my current CPU for SC4, and I will then continue to have scrolling and other performance issues in the game.

So I guess my question is: Would I see any increase in game performance with a Xeon (from the 45nm to 32nm die shrink and new architecture) or would i need to go with either and i5/i7 or $2000 Xeon?

Keep in mind that exports on my failing mac usually take 24-48hrs or more so the more cores the CPU has the more I can cut down on that time.

[Aaron Graham 4,417]

I know I'm not an expert but have you took a look at the High End CPU and the GPU Benchmarks on hardware performance, it can help you out if you do not know about this.

Okay I think it a good price but you could try to find a faster CPU that is cheaper, I was looking at the (Intel Xeon E5-1650 @ 3.20GHz) and it is about $100.00 cheaper but I do not see it in newegg yet. Here's is the link to your CPU on your laptop (Click here) as you can see it's a lot slower than the (Intel Xeon E5-2630 Sandy Bridge-EP 2.3GHz) and the Intel Xeon E5-1650. The Intel Xeon E5-2630 Sandy Bridge-EP 2.3GHz from what i'm seeing is about 4 faster than your laptop and the Intel Xeon E5-1650 @ 3.20GHz from what i'm seeing is about 6 times faster that your laptop CPU so that should be like having six of your laptops rendering your building that should be really fast and having two CPU's your computer is something I would only dream to have. Here's a list of High End CPU's (Click Here).

Links agian to compare the CPU and price.

Intel Core2 Duo T9600 @ 2.80GHz (Your laptop CPU) Ranked 442

Intel Xeon E5-1650 @ 3.20GHz

Intel Xeon E5-2630 Sandy Bridge-EP 2.3GHz

Here's the link of High End GPU

http://www.videocard...h_end_gpus.html

The GTX480 seems to be a bit faster than the Radeon HD 7850 and is cheaper than the Radeon HD 7850. One thing I heard is that the Radeon GPU tend to get really loud when are hot, you can fin some youtube videos on that.

No matter what I said I still think you have picked a great set of hardware for your rendering monster.

  Edited June 8, 2012 by Aaron Graham  

[PBGV103 729]

Thanks Aaron. Unfortunately the E5-1650 only has one QPI link so it won't work with a second CPU if I go that route later on. I'm also leaning toward an AMD GPU for eyefinity. There's no Nvidia equivalent unless you go with a Quadro card with Mosaic, and I'd like to play around with multiple monitors in SC4.

I'm just wondering if clock to clock (2.8GHz to 2.8GHz) the Xeon will be noticeably better than the Core 2 Duo. Generally there's a 10-25% improvement in speed at the same clock rate with new CPU architectures. Since that Xeon will only overclock to about 3GHz and my Core 2 Duo will do 3.3GHz I'm wondering if I would see a decrease in game performance.

[A Nonny Moose 8,261]

Seems to me you want to take a sledgehammer to a gnat. Your greatest need seems to be RAM. Can you not jack up your current machine? (probably not if its a laptop). Unless you are actually building a server, I would look for something cheaper. Have you looked at AMD seriously?

Why do you want six cores? What software do you currently have or plan to acquire will run that much multi-tasking?

I am not a hardware type, but I've worked all my life with big mainframes and I don't see why you want to have one. You mention running only two programs, and you complain about BSODs. How about getting rid of the BSODs altogether and get out of the Microsoft camp?

Have you ever tried Linux? SC4 runs fine on Ubuntu 12.04 with wine 1.5.5. Your other program will probably run either native, if it is available, or with either wine or mono. Linux is much more graphics oriented than Windows. Linux has no problem running as many cores as you like, has multiple inter-process communication methods, and is generally the Swiss army knife of operating systems. And it is free. You can test it on your present machine without risk. Just download and burn the live-cd boot, and boot it. It can run entirely in memory and not affect your other system, or you can install it dual-boot.

I find it hard to believe that you would spend all that money then purchase bloatware from Microsoft. Hardware is not always the answer.

  Edited June 9, 2012 by A Nonny Moose  

[PBGV103 729]

What bloat wear are you referring to? Also the BSODs are occurring because the CPU in my laptop is beginning to degrade, despite having kept the heat lower than stock, the long renders and OCing seem to be taking their toll. I now can only get a stable OC at 3.25GHz when I used to reach 3.3GHz. The reason I'm going for so many cores is because vray rendering is very efficiently multi-threaded in 3ds max. Clock for clock a six core processor should be able to render 3 times faster than a dual core processor, not taking into account other slight improvements in speed due to hyper-threading and architecture. From the benchmarks I've seen, the e5-2630 should render about five times as fast as my t9600 core 2 duo.

Linux sounds interesting and I might have to check it out, however creating a dual boot system seems to be confusing the issue when I already have a free copy of windows ultimate from my school. I've also heard about the issues you've had with graphics settings for SC4 in ubuntu. A single 4 core i7-3770K overclocked to 4.4GHz should provide the same performance in rendering as the e5-2630; however, the LGA1155 platform doesn't offer any upgrades from there, whereas with the xeon system I could upgrade the single cpu to a very fast 8 core, and/or add a second cpu. So basically if the i7 isn't fast enough I have to replace the entire system, but if the xeon isn't fast enough I have several options for upgrades.

You might ask why I need so much power for 3ds max when some other designers don't have that need. The answer is a combination of things, for one I usually have many more objects in my scene just because I hate taking the time to tediously attach countless vertices of polys, but more that that, I try to create reproductions of building as close to the real thing as possible. That take tons of tiny details and heaps of complicated materials. I should also mention that I once tried making a scene with only a single fountain in it. It consisted of 2 simple materials and a particle system. I cancelled the preview render after it ran for THREE DAYS with out results. So yes, I think I need a faster computer.

[A Nonny Moose 8,261]

In bloatware, I am referring to Windows in general and any "free ware" you get with it. Windows if full of code for special deals with Microsoft's Application wares, and this puts stuff in the operating system that should not be there. Generally speaking, Microsoft writes code as if it was the only user on the machine.

A six core processor will only render faster if it can multi-thread the render. Is 3DS Max coded for that? If so, how many cores can it use? Multiple core usage is only there is the appropriate co-routines and forks are present to create separately dispatchable tasks. More processor speed is better than more cores unless the code is specifically written to take advantage of it. It is not magic. However, you can dedicate one processor to a task, extend its time slice, and get more CPU cycles in real-time processing if you don't mind the O/S doing its other business on the remaining cores.

I think you'll find that RAM is the key. At least 2GB per processor core (in your case at least 12GB, so 16 should be a great plenty). However if you add another 6-core processor chip you'll probably want some more RAM if you can multi-thread that deeply.

Linux issues are something you can play with when you get a free moment. I do suggest you give it a try since it is easy to do. As for graphics issues, they are with my GPU, which, if I was doing the kind of work you are doing, I would change. Being retired on a tight budget makes me tolerate stuff I would not when I was working. I've been on Linux now since around 2007 or whenever Vista appeared. I seem to have an el-cheapo AMD/ATI chip set, and getting it to perform is a problem. I run software mode because I just can't be bothered debugging it. There is definitely a problem between the game, what the driver can do and what the on-the-fly api layer can hack. Lately, wine has improved quite a bit, but the problems with fglrx remain.

I guess my basic concern is that you get what you need and not something you can afford but maybe is overkill. Some of the left over loot might well go to better software. You can't count on free software from Microsoft forever. Sooner or later they will want $400 bucks or more from you. Be careful not to run up a student loan for this. You'll regret mortgaging your future. I think you should try and find some way to benchmark your applications on this machine before you actually commit. Which is better, paying your fees at school or paying interest on a loan?

By the way, if you are looking to add custom content to things like the new SimCity (2013) you'd better be sure what the poly limit is.

[PBGV103 729]

Thank Nonny. 3DS Max itself generally only uses 1-4 threads depending on what you're doing with it, and the viewports are very gpu dependent; however, the rending is done by vray which divides the scene in small square sections and each thread handles a section. I saw one video that demonstrated it utilizing 128 threads on an 8U server (8, 8 core xeons).

As far as the Windows thing goes, I got the full version of Win7 Ultimate (actually Enterprise but they're the same) and a product key, so I should be fine until I have to upgrade to Win8 (which doesn't look so promising btw).

[A Nonny Moose 8,261]

Sounds like that really wants a processor hypercube. You'd need Niagara Falls to cool it, and could run only during the day because the falls are shut off at night. Beware of Windows dispatch (time-slicing) algorithm. I sometimes feel that it is responsible for some BSODs. I am not sure they really have that sorted out, even though how to do it was well known in the 1950s. (The Atlas-8 had 8 separate CPUs).

[PBGV103 729]

You may be interested in this video. It shows 3ds max (and an older version at that) rendering with 16 core (32 threads) each individual thread is shown as one of those tiny little boxes moving around. The one draw back to this demonstration is the guy only used 16GB of memory. Skip to around 16 minutes to see what I'm talking about.

  Edited June 11, 2012 by PBGV103  

[A Nonny Moose 8,261]

Interesting piece of monitor software. I notice it says it is running on a "server". I hope you are prepared to do likewise. I don't think the Server software is any different from what you have.

On Linux if you wanted a real server, you run Apache. The attitude is somewhat different.

[PBGV103 729]

You mean where he's using Windows Server 2008 R2 for an OS? I'm not sure why he picked a server OS, and from what I've heard, a very expensive one, but the guys over at the EVGA SR-X forum are all running Windows Ultimate because it supposedly handles up to something like 128 cores. I've been keeping up with the talk over there about the SR-X board and so far of the hundred or so that own one now, it seems the majority of them are having a problem of some kind. The list so far is: LAN drivers disappearing at any overclock even 1%, loading DIMMs 7 & 8 causing no boot, no audio or SAS drivers, USB staying powered with the computer off (this one seems extremely strange), being able to boot with CPU0 but not CPU1, not recognizing dual GPU cards, freezing at boot with PCIe storage/boot drive, and the list goes on. Apparently the board is fairly new to the market and several BIOS updates are on the way, but I'm beginning to wonder if I'll have to wait too long for those updates to get here. I was planning on purchasing this month... I also looked at just getting a i7-3930K and an Asus Rampage Extreme IV, but I've heard they've had their share of issues too like bad BIOS and bad power management and voltage controllers leading to at least one $1000 CPU being burnt up, and apparently ASUS won't replace your CPU if they destroy it. So I'm not sure what to do.

From what I've heard and read 3DS Max and Vray generally get better performance for Intel CPUs than AMD, but the lower cost, overclocking, and keeping the same sockets over several generations is starting to sound attractive. I would just be afraid that Bulldozer and all that other BS would hinder performance when there's not optimization for it in Max. Plus, I'm an Intel fan. Hahaha

So aside from AMD, what is a good Intel based option for a system that's upgrade-able within 1-3 years because I do foresee me throwing more and more demanding stuff at this computer. I'd also like to keep it in the $2000-2500 range since computers lose at least %20 of their value in the first year alone.

[A Nonny Moose 8,261]

Well, you have been doing your homework. Microsoft wants an inordinate amount of money for what they call a server but which is really a relabelled copy of Ultimate I suspect. If Ultimate can really handle 128 cores, it can handle a full-scale mini-hypercube. I wonder what else it does besides manage the processor queue in one of its cores? I'll bet the watchdog timer code is just about as complex as it gets. (The watchdog monitors for a fault on any processor indicating an instruction not complete failure).

For ordinary desktop computing use, my last two machines have been AMD. I am very happy with them, but have no feel for what is going on in their skunk works (they do have one) that resulted from the acquisition of ATI.

I think you have got the idea of the value of a machine a little understated. It is worth only scrap value the instant you turn it on. A computer is a hot plug in the wall into which you pour money. Depreciation is a fiction dreamed up by the bean counters.

[Zades 7]

Currently I have a core 2 duo 2.8GHz (T9600) MacBook Pro that's almost 3 years old.

My plan is to replace it with a Xeon system, but as they are incredibly expensive

In my honest opinion that is way overkill. Xeons preform around the same level as i7s but can take in more ram and support some other functions (reason why its often used for servers). Unless you are doing something extremely memory and cpu intensive (SC4 is not one of them), and i7 is really all you need. But if you have the money to spare and just want to buy a super system, sure go ahead with a Xeon. While buy a Xeon, you might as well dump in 32GB of ram (8GB x 4). Oh didn't see the export time at the end. Yea get 2 CPUs for that if you can afford. Personally though, if you are going for an all out high end build, buy an extremely powerful computer, don't settle for a tier under.

With a dual socket Xeon board I can start out with one midrange CPU and replace it or add a second one later. Still, my concern is that at a maximum of 2.8GHz the Xeon I can afford right now will be the same speed as my current CPU for SC4

First, you cannot just compare CPUs by its clock speed. Sure it's a general factor but you can't just say oh CPU A has 0.5ghz more than CPU B, therefor it is faster. CPUs have come a long way since Core 2 Duos. First, most of the time the CPU is bottlenecked by the memory due to clock speed not being fast enough on the memory (and they are still slower even today compared to CPU speeds). However, that is all solved with multi-channeling, where multiple sticks of RAM will process at the same time. Compound that onto the fact that i7s have 4 and xeons have 6 cores (for the more standard versions) hyperthreaded, givinging 8 and 12 cores each.

Getting an i7 or an Xeon as an upgrade to a Core 2 Duo will GREATLY increase your preformance oon a game like SC4. I run Sim City 4 on a laptop with an underclocked Sandy Bridge i7 at 2.0Ghz with 8GB of ram (graphics card is GTX 560M if that matters to you), game runs fine until you get into 700,000 population (then it takes a while to calculate every time you plot down a building) and even then the bottleneck is Sim City itself as my CPU is nowhere near maxed out. And adding another CPU will not help SC4 run faster (actually not too sure, I need to confirm this but I'm pretty sure this is the case).

First off, as a 32bit game, SC4 will not use up more than 4GB of ram. From looking at my graphs while running the game, it also will not use more than 2 cores at once (I'm still unsure about this, so don't completely trust me on it). Needless to say that simply will mean SC4 at max will use around 20% of your CPU and still suffer from stalling time due to calculations at high populations.

Regarding eyefinity, nVidia has Surround View, but I never use any of those so I can't really comment on them.

[PBGV103 729]

SimCity 4 is not multi-threaded at all, so it will never use more than one thread at a time. In fact I've seen reports that Intel's hyperthreading can actually decrease performance on the game. Luckily hyperthreading can be turned off in most BIOS. The only increase I may see clock for clock between the 2.8GHz Core 2 Duo and the Xeon is from architecture improvements, but many of those improvements are software dependent from what I've read, so I may not see any difference whatsoever.

My hope for this computer is to create a system that not only plays SC4 very well, but also significantly cuts down on my 48hr renders. Of course rendering is where the Xeon would excel.

[A Nonny Moose 8,261]

You are completely correct about the game. At best it is a P4 game and maybe not as good as that. Remember the limits on 32-bit addressing. The game can never use as much as 2 GB if that, notwithstanding that systems may allocate huge space for it. I get a page table that could hold up to 16GB, mostly virtual. But that is just the O/S being generous. These pages never exist.

[Zades 7]

SimCity 4 is not multi-threaded at all, so it will never use more than one thread at a time. In fact I've seen reports that Intel's hyperthreading can actually decrease performance on the game. Luckily hyperthreading can be turned off in most BIOS. The only increase I may see clock for clock between the 2.8GHz Core 2 Duo and the Xeon is from architecture improvements, but many of those improvements are software dependent from what I've read, so I may not see any difference whatsoever.

My hope for this computer is to create a system that not only plays SC4 very well, but also significantly cuts down on my 48hr renders. Of course rendering is where the Xeon would excel.

May I ask, around what population do you start to notice that the in-game time in SC4 starts stalling/stopping?

[PBGV103 729]

The biggest city is 485000 in population, but there's a lot more jobs (from all the plop-able CO$$$ I think). When I say it lags, the time doesn't move at all when it's supposed to be going its fastest, and on every time mode, scrolling is jittery. I think the scrolling may be a graphics driver issue, but it's a very fast, dizzying stutter when scrolling. Most of the time after changing the zoom or angle I have to wait a minute or so for the automata to settle down from stuttering around the screen. Load time for the game is pretty short after packing plugins, but loading this particular city takes a few minutes. I'm hoping the SSD will speed that up.

[A Nonny Moose 8,261]

This whole business is dependent on volume and your processor speed. In my larger city of around 300K Sims and many supporting objects, I've found that the processor for this task is running at 100% a lot of the time. This tends to delay frame updates and/or cause stuttering of some objects like message balloons in software rendering mode. I have an unsolved issue with my graphics driver (fglrx) that prevents running in hardware mode. Things have slowly improved over the last couple of releases, and I am hoping the October release will help a lot.

Meanwhile, it really depends on the number of objects including Sims being visited on a regular basis and the complexity of the network as the pathfinder runs about every 100 Sim-days. The pathfinder will always cause a processor spike and make cycles unavailable to other tasks. The Xeon processors may help this somewhat, being that they are faster than a scared cat, but only if the game uses the proper instruction set.

[PBGV103 729]

Nonny, so clock for clock (both run at 2.8GHz for SimCity) you think there's a chance the Xeon will handle that pathfinding issue better than the Core 2 Duo?

[A Nonny Moose 8,261]

Nonny, so clock for clock (both run at 2.8GHz for SimCity) you think there's a chance the Xeon will handle that pathfinding issue better than the Core 2 Duo?

There is only one way to find out. In situations like this I always ask the machine.

You might need a hardware monitor to find out what is going on. I am not sure that with a microprocessor this is possible. In the bad old days with big circuit boards for processors you could probe the board to the Operation Complete signal or even list the instructions that were being executed.

Try and devise a benchmark that can be executed by both machines with identical input. Set up a city with lots of network paths, then just call it up on each machine and see what happens if the only thing you do is turn on drawpaths. In both cases, it may be too fast for a visual inspection.

[PBGV103 729]

That seems like a fun experiment. The only problem is I don't have access to the hardware and like I said before, everyone's policy seems to be "return for replacement only" so if the Xeon isn't good enough I'm out of luck once I've bought it.

  Edited June 16, 2012 by PBGV103  

[A Nonny Moose 8,261]

Well, all you have is advertising and hearsay. It would be good if you could get a "try before you buy". It is not all the money in the world, but it is your money and it should be spent wisely. Good luck with this.

Heck, what would it cost them to load up SC4 and take it for a test drive with your test city? Instead of dickering directly with them, it is too bad you couldn't make it look like a bid competition.

[PBGV103 729]

Haha the bidding scenario would be pretty entertaining, but unfortunately this stuff is coming from Newegg so I don't really have anyone to test it out. I'm 90% sure I'm going with the Xeon option now, basically just because it's upgrade-able and has a good reputation for handling 100% load 24/7 and still lasting years. I figure it'll hold up no matter what I throw at it. The issues with the SR-X board seem like they're slowly getting worked out and most of what's left is problems the "extreme users" are having, such as running dual GPU cards and/or 4-way SLI with ECC memory. I doubt I'll ever get server memory or buy $2000+ in GPUs. There's also issues with the RAID controller driver, but I don't see myself setting up raid 10 anytime soon (although mirroring sounds pretty nice).

I may even put the whole thing off till sometime in July when EVGA releases the 1500W PSU designed specifically for this board, or maybe I'll just use the 750W and 300W PSUs I have lying around here.

[A Nonny Moose 8,261]

Sounds like great fun. Get an external hard disk connected by a USB port, and do back ups once a week. This will save all the work with mirroring and RAIDs.

I am a poor pensioner with almost no income, but I have two external HDDs connected over channels. One on-line for automatic backups, and one off-line for backups of backups. If it is all on-line at the same time, I have 1.8 TB of space, but I only use about 2% of that.

Disks are so huge these days. I remember when a 96K (36-bit) words was considered a lot of storage. We ran Expo '67 on two of them in the operations control centre. We had two Univac FastRand 17K drums but they were damaged in shipment and the heads fell on the drums when we powered them up. Budget couldn't stand new drums, so we reconfigured the operating system to run off the big disks which had an average access time of 192 ms. (That's milliseconds, and you only got that if you had separate seeks running on the four separate heads). Base access time was 500 ms. When these drives were discontinued, the nickel/cobalt disks made great coffee tables. They were three feet across. We had 48 of them. Montreal was a great place at the time, and I suppose it is still the most cosmopolitan city on the continent.

  Edited June 17, 2012 by A Nonny Moose