PC Upgrade Pending

[Megageth]

PS the problem with asking for advice is that everyone has an opinion and they hardly ever agree.

PPS I'm on the i5 > i7 bang for buck on a gaming rig train. i7 is the better CPU but does not add that much to games from reviews I have read.

PPPS There are new sockets coming out with support for DDR4 this year, I say OC the CPU you have.

PPPPS I have had a few.

Advice from http://www.tomshardware.com/reviews/...ck,3106-4.html

We have seen a small handful of games benefit from Hyper-Threaded Core i7 processors, though. Because we believe this is a trend that will continue as developers optimize their titles, we're including the Core i7-4770K as an honorable mention, now selling for $325. In a vast majority of games, the Core i7 won't demonstrate much advantage over the Core i5. But if you're a serious enthusiast who wants some future-proofing and values highly-threaded application performance, this processor may be worth the extra money.

PPPP...S If Mantle comes in to BF4 then you may find that an AMD cpu thrashes the i7 in BF4. Still to be seen.

What a post, I need a publisher.

[Skouperd]

HT virtually splits up cores to try and improve performance by letting a core switch contexts during blocking operations on a single thread.

hmm, that is not quite how multithreading works. CPU architecture is quite complex (obviously) but if you are keen to learn more about exactly how CPU's operate and work, may I recommend you look at something called NAND2Tetris as that will allow you to actually build your own CPU together with things like ALU (arithmic logic units) and logical processors. Just to make it clear, hyper threading is slightly more complex than your description and a CPU with hyperthreading available actually has two logical processors per core. There are certain units within the core that they do share (like Cache, memory addresses etc) but hyperthreading is not just software.

What happens with hyperthreading, and when you get poor performance is just bad coding be that form the OS or from the game developer. Given the crap quality of BF4 versus the pretty stable version of Windows, I would venture to guess the fault is not with Microsoft on this one. What happens from a programming perspective, and this is what you see when you start overclocking the kak out of a CPU is that not all the units (remember, I said certain units within a core is shared by the logic units) is not getting overclocked to the same extend as the logic cores etc. and this then causes a bottleneck.

Crappy programming does not realise that half the cores are HT cores and instead of spreading the workload out over the physical cores (core0, core2, core4 and core6) it will give it to core0,1,2, and 3 which effectively have 4 logic units processing the data and waiting for the shared units on the core. Whereas, in a perfect world, the workload should have been shared.

I seem to recall reading once that the die size increased roughly 5% (if memory serves me right) with the additional physical hardware needed on the CPU for the implementation of HT. So I say again, HT is not just software switches, there are actual hardware also doing the work.

Jarrod / Tiertjie, Let’s do a quick comparison between LGA2011 and LGA1150. For starters, Sandy Bridge, Ivy Bridge and Haswell parts all feature 16 PCIe lanes whereas the Extreme parts, i.e. Sandy Bridge-E and Ivy Bridge-E (i.e. Socket 2011) features 40 PCIe3.0 lanes that branch off of Ivy Bridge E. Remember the Ferrari versus the Mini-bus… more lanes is good, less lanes is bad.

I will agree that X79 does not have native USB3 ports, and that it generally only features 2 6Gbps SATA ports. However, saying that, motherboard manufactures have incorporated third party chips directly into their motherboards for either more SATA 6Gbs ports or additional USB3.0 support and with the amount of PCI-express lanes to play with, it makes absolutely no difference whatsoever. The difference between going native USB3.0 versus non native is that there will be a pico second increased in latency between the two. USB 3 is used to predominately use to transfer data, I assure you, you will NOT notice a difference between copy a 2GB file on a native USB3.0 controller versus an third party onboard controller. I could argue, that some motherboards have 3^rd party Ethernet adapters therefor surely that is worst for online gaming?

Personally, I would much rather prefer Intel to give me the PCI express lanes to do with them as I please. If I am in the mood for uber-fast hard drives (shall we say 8 SATA 3 drives) then I can plug in a proper high end raid controller, and not need to worry about the card impeding my graphics card’s bandwidth in the process. If I would rather prefer 4 Titans, then again, no problem I can plug them in and not have to worry about the natively supported USB3.0 features taking up all my PCI-express lanes which I will not use. Unless you are saying Paul will use USB3.0 to output his monitor, I don’t see a problem that it is not natively supported on X79.

Ok, so what else is the difference between 1150 and 2011 apart from the PCI-Express lanes discussed above? Shall we look at memory quickly? Obviously the better the CPU, generally the more L3 cache it supports, and as I’ve explained here: http://blog.skoups.com/?p=592 more cache is good. Quad channel memory versus dual channel memory, need I even go into this topic?

We can argue stable platform, mature platforms, blah blah blah.

I like your Steam graphs... if that is anything to go on, all games will be written for dual core CPU's running 2.4GHz.


I think the one thing we can all agree on is that graphics cards is the first thing that will show it's age no matter what built you go for. Going for the uber GPU now, and a crap CPU means that in 3 years time, Paul will need to spend another R8k to get his games playing. (with diminshing returns from day one after buying the system). My solution on the other hand provides Paul a solid platform within his budget still, playing games are fairly good FPS NOW, in 18 months, another say R2k or R3k for say a GTX770 which should give him another 18 months game time. 3 years from now, (i.e. 18 months after getting the GTX770) he could pick up another GTX770 for R1000, or he could go ape shit and just get himself the GTX970ti for R3500. Over a 3 year time frame, going my route, you have not only ensured that he is able to continously play the latest games at very acceptable frame rates, but you have also saved him a lot of money in the process. My estimated second hand prices is based on historical averages of how fast GPU's depreciate in value. I also have not taken into account inflation in any of my quotes but that should cut both ways so not too concerned about it.

[kmckelvin]

I'm hardly likely to explain all the theory and varying methods of doing HT in one sentence :P

However the operating system is responsible for scheduling threads to be executed on the CPU and more recent operating systems are a lot better at making sure that the primary cores are prioritised. From a programmer's point of view all of the scheduling is abstracted behind the concept of a Thread and handed off to the operating system to manage.

The place I've really seen HT kill performance is in running virtual machines.

Bottom line about my comment though is that I'm running a 2 year old CPU architecture with a top of the line gfx card, and the bottleneck is still the GPU, not the CPU or memory bus. And if there's any question about disk speed, running a mid-range Sandisk SSD I'm almost always the first person to spawn in on a map load.

Obviously there are places where a CPU makes a bigger difference. When it comes to database servers there's a noticeable improvement in performance with more CPU cores available (provided that the database is properly optimised in terms of indexes, data organisation with clustered indexes, RAID, enough RAM, disk access speed, etc). CPU cores make a big difference in compiling binaries as well. These are tasks that grind the CPU by orders of magnitude more intensely than Battlefield 4 ever will.

When it comes to 3D graphics, given a reasonable spec CPU, the GPU is the next most important thing to throw money at.

With all of that said though, I agree with Skoup's last paragraph about spend over time. Getting onto newer architecture now will mean that you have more freedom to upgrade later at a lower cost. You'd be able to pick up a second graphics card later at a pretty decent price and keep up to acceptable frame rates. If you want MOAR FPS NOW, drop more on the graphics card now and worry about newer architecture when it becomes a barrier/bottleneck.

[Paul]

Geez never thought this would stir up such debate. Thanks for all the input. After reading all I am leaning towards the higher end GPU with a mid range CPU.

Mega, that is why I am only upgrading in Feb/March, I hope that new info on mantle, new architectures etc are out before then. However I also beleive to put a stake in the ground and upgrade when needed instead of waiting. There is always going to be something better just around the corner.

So for now I am looking at the i5-4760 with a decent enough Mobo (suggestions) and either R9 280x or 770 (depending on price)

[Cat Nipples]

Nimbzy what did the SanDisk cost you? (sorry for derailing) Just asking out of curiosity

[Paul]

I got an 128Gb SSD for R1500 about 6 months ago, it is a Corsair so a good make

[Cat Nipples]

Oooh... Hmm..

I think I need to go into male stripping, need some cash..

[kmckelvin]

Nimbzy what did the SanDisk cost you? (sorry for derailing) Just asking out of curiosity

256GB for R2400 at Matrix (got it at rAge, so it might have been a discount)

[KakHard]

So for now I am looking at the i5-4760 with a decent enough Mobo (suggestions) and either R9 280x or 770 (depending on price)

4670? If so, rather pay the R120 extra and get the 4670K so that you can overclock if you later feel like doing that.

[Skouperd]

The place I've really seen HT kill performance is in running virtual machines.

Ahem to that! I think I know why that is happening but yes, I have seen that too.

Bottom line about my comment though is that I'm running a 2 year old CPU architecture with a top of the line gfx card, and the bottleneck is still the GPU, not the CPU or memory bus. And if there's any question about disk speed, running a mid-range Sandisk SSD I'm almost always the first person to spawn in on a map load.

While my main gaming rig is leased out to a company requiring a bit more processing power, I had to revert back to my LAN pc which sports an i7-950 CPU, (released Q2-2009) and I paired that with a GTX780ti. I agree with you that a better GPU will increase frames but I disagree that your CPU is not bottlenecking you. How can I explain this, ok, let's take a hard drive (say a 7200RPM drive) for instance which I think we will all agree is one of the slowest components in your computer. Run a benchmark on your harddrive, (copy files between two folders) your throughput on your harddrive is say for argument sake 100MB/s. Now, underclock your cpu with 50%, run the same benchmark on your harddrive, your throughput will still be 100MB/s. Increase your CPU overclock to say 50% above normal, your hard drive will still only give you 100MB/s. Doing that test you know that your harddrive is indeed the bottleneck in your system. When you drop your CPU to something stupid like 50MHz or something, then your CPU will become the bottleneck and then that is the slowest component.

Now, let's do the same thing with your GPU as you say your GPU is your bottleneck. If your GPU is the true bottleneck, then what one should expect is that as we underclock the CPU, at some stage the CPU will become the bottleneck and no longer your GPU. (obviously the CPU will become a bottleneck much sooner than the HDD example when it only becomes the bottleneck at say 50Mhz). Now, if you are saying that your GPU is bottlenecking you, then if you double up the speed of your CPU you should NOT see any difference in performance of the GPU. (the hard drive did not increase from 100MB to 120MBs just because we overclocked the CPU). However, I am prepare to bet you my rAge2014 ticket that when you increase your CPU speeds on your current rig that you will get better performance from your GPU thus your GPU is NOT your bottleneck.

It is very dangerous to assume just because your CPU is sitting on 30% util and your GPU on 99% that your CPU (or something else) is not also holding you back. Building a fast system, and understanding what is causing what bottlenecks is a very complex situation. You can only say that your GPU is bottlenecking you if you see no improvements when you increase the speed of your CPU.

Back to my original statement above, as mentioned, I have a GTX780ti running in an old i7-950. Let's compare that to BadA$$ machine, he currently has a GTX660ti paired with a socket 2011, 3820 CPU. Running one of my favourite graphics benchmarks, he scored 28,089 points, compared to my score of 27,578. BadA$$ with a bottom of the range LGA2011 (non overclocked) and a comparitively no-starter GPU (compared to my GTX780ti) still manage to out scored one of the fastest GPU's on the planet with the main difference the CPU. Saying that, I hit 200FPS, dropping to 120 on occasions, in BF4 (that was before I stripped my moer for EA and Dice) with everything on Ultra apart from the last three things in the GPU options.

Either way, Paul, good luck with your decision, and I am looking forward to recall this thread in 3 years time when you start the thread again by saying... "ooo.... guys help I need to upgrade my computer as my game sucks and I am not able to get anything more then 40FPS...."