New PC Build, ~$600 USD

[Varine]

http://pcpartpicker.com/p/L8Wyqs

PCPartPicker part list / Price breakdown by merchant

CPU: *Intel Core i5-4430 3.0GHz Quad-Core Processor ($189.99 @ Newegg)
CPU Cooler: Gelid Solutions CC-Siberian-01 51.9 CFM CPU Cooler ($9.99 @ Newegg)
Motherboard: MSI Z87-G41 PC Mate ATX LGA1150 Motherboard ($115.66 @ Newegg)
Memory: Crucial Ballistix Sport 8GB (1 x 8GB) DDR3-1333 Memory
Storage: Seagate Barracuda 320GB 3.5" 7200RPM Internal Hard Drive
Video Card: Gigabyte GeForce GTX 650 2GB Video Card ($139.99 @ Newegg)
Case: Thermaltake VM30001W2Z ATX Mid Tower Case ($45.98 @ Newegg)
Power Supply: Corsair CX 430W 80+ Bronze Certified Semi-Modular ATX Power Supply ($53.98 @ Newegg)
Total: $555.59
Prices include shipping, taxes, and discounts when available
*Lowest price parts chosen from parametric criteria
Generated by PCPartPicker 2014-06-17 03:36 EDT-0400

As my old desktop is getting rather old and sluggish, I want to get a new one. The target is cost is 500 to 700 USD, and the most taxing uses will be somewhat intensive modelling applications ( high resolution 3DS Max/ZBrush scenes, nothing too crazy) and ideally new games (largely single player RPG's and some multiplayer FPS), highest settings not necessarily mandatory.

If anyone has suggestions for different parts that would increase the cost/performance ratio, that would be awesome.

 

[Slapshot136]

this PSU instead

not sure why you are going for a single stick of ram..

 

[Varine]

Because I only need 8 gigs right now, and that's a fairly simple one to upgrade later.

 

[Accname]

Having several smaller sticks is better though.
Energy consumption and fetch time grow exponentially with size.

 

[jonas]

Having several smaller sticks is better though.
Energy consumption and fetch time grow exponentially with size.

Based on what? What is in there except an address decoder and an or-tree? Sounds more like logarithmic fetch time (in the number of blocks). The power consumption of a decoder is scary, though - but I think it's only exponential in address length and thus linear in the number of blocks. Putting two of them with a multiplexor in front and an or gate behind should not improve anything.

 

[Accname]

Bigger ram needs more transistors and they all need to be clocked together.
The clocking signal has to travel farther around the chip to reach all those transistors in about the same time.
Since the signal is limited in speed by the speed of light, it will inevitably be slower when you have more transistors.
Its the actual size of the chip that matters here. When you deal with nanoseconds every millimeter is important.

 

[Varine]

Easy enough change, cheap enough I will probably buy more anyway.

 

[jonas]

Bigger ram needs more transistors and they all need to be clocked together.
The clocking signal has to travel farther around the chip to reach all those transistors in about the same time.
Since the signal is limited in speed by the speed of light, it will inevitably be slower when you have more transistors.
Its the actual size of the chip that matters here. When you deal with nanoseconds every millimeter is important.

Thanks for the explanation of wire delay. Now please explain your claim that the increase in delay is exponential and that power consumption is exponential in the size of the stick.

I just read your post more carefully. The clocking signal is certainly not the limiting factor, and the transistors are not clocked - maybe you meant registers?

Note also that the registers do not have to be clocked at the same time, but in such a way that setup and hold times are satisfied. That is, the input signals need to have stabilized and stay stable around the time at which you clock.

 

[Accname]

Thats what I was thought in my lectures about hardware components.
Lots of studies and graphs and shit which was graphically showing how latency and energy consumption grow in comparison to memory size.

I dont know the sources by heart but you can probably google them and tell us the results.

 

[jonas]

I just went to Giorgi who works directly with the hardware (using VHDL and FPGA-boards). He supported my view.
Petro who was also there said that the wire energy loss is at most quadratic because of the matrix form of the decoder (which is not a planar graph). However, I'm not even sure of that.

Maybe what you remember is in comparison to address width?

However, he also said that motherboards are optimized to work with multiple RAM sticks, especially if they have the same delay. Then some memory accesses can be performed in parallel.
So even though your argumentation for it was incorrect, your suggestion was valid.

 

[Accname]

You can not say whether it was incorrect since I did not give a real argument. Only if I would link you the sources, which I wont because I dont remember them, you could really say much about it.

But, on the other hand, the sources were a little bit dated. It might very well be that they were still talking DDR2 or even DDR1. I cant remember.


Edit: By the way, this is the same reason why caches in the CPU are multi-layered. You have a level 1 cache which is very small but fast and efficient, and then at least a level 2 cache if not a level 3 and 4.
Each level becomes bigger by about 2 - 4 times but also slower.
Its the same reasoning.

 

[jonas]

Don't you think we're hijacking this thread? Anyway, since we have such an engaged discussion going on and varine is not always nice to us:

But, on the other hand, the sources were a little bit dated. It might very well be that they were still talking DDR2 or even DDR1. I cant remember.

It makes no difference, the underlying architectures are very similar.

Edit: By the way, this is the same reason why caches in the CPU are multi-layered. You have a level 1 cache which is very small but fast and efficient, and then at least a level 2 cache if not a level 3 and 4.
Each level becomes bigger by about 2 - 4 times but also slower.
Its the same reasoning.

The reasons are different, and what you said is not completely true.
Your reasoning for buying two (parallel) sticks instead of a large one was that latency is exponential in memory size, this is false (latency is linear). There is a simple argument for this: If buying two sticks and putting them together would have the square root of cost and delay of a large stick, people would use that technology to build larger sticks rather than the technology they are using now.

Caches, however, are not put in parallel, but in sequence. This means that storage capacity is not increased by more cache levels.
The different levels are to compromise between:
1) cost (more & larger caches are more expensive)
2) performance (smaller (/SRAM) caches have a lower latency*, larger caches can keep more data)
3) synchronization overhead (caches that are shared with more processors become more sequential, cache misses make the access more sequential -> bad for parallel computing)

Thus the different levels are shared to varying degree, use different RAM-architectures, and have different sizes.

*: while smaller caches are indeed faster (latency is linear), another important point is that there are different RAM-architectures.
For example, SRAM is much more performant and expensive than DRAM (with the same address size, of course).

 

[Varine]

This is far more entertaining than what I was talking about anyway.

 

[thewrongvine]

This is far more entertaining than what I was talking about anyway.

Much more interesting.

I didn't know / I did know but I just forgot that you do 3d modelling.
How well does 3DS Max and ZBrush run for you now? I started learning Maya a few weeks ago and I run it on my laptop, and it handles scenes decently for the most part, but rendering is quite slow, which is expected I suppose, being a laptop that overheats and slows down so quickly.

 

[Varine]

Well as I threw most of my old PC away it doesn't run at all. I only currently have a chromebook, which I have installed Ubuntu on next to Google's OS, so I would have to use wine and know it will be slow. I can use blender okay, but I don't know the interface as well as I do with 3dsm so I haven't really used it at all, and have focused on my drawing.

 

[thewrongvine]

Well when ya get your new desktop, if you ever want to do some requests for 3d models for vfx usage in videos, let me know because that would be awesome. I basically stopped trying to learn the modelling part as it's so time consuming, and even though I'd love to learn more and be at least decent at it, I can't devote the time to that while learning other things. So I'm focusing more on learning how to better render and export/integrate already-made/commissioned models into live action scenes for vfx.

 

[Varine]

If I find time I'll let you know. My work schedule is like 60 hours a week right now and I have to put most free time to my projects - fucking so scattered right now as to what I'm doing. I'm not the best modeller by any means anyway, I understand the program well but I'm not great with organic stuff. Most of my experience was architrctural visualizations, not so much characters and animation.