git gud git more CPU

But yeah, Source is a pretty CPU-intensive engine, so an i7 would always be better

If you have a choice go with the i7. Worst case it runs equally effective to an i5 in tf2 but will run everything else much better.

Multitask is king.

Technically, in some situations an i5 might be better, but the reason is a little finnicky. TF2 only uses up to three threads. An i7 is basically an i5 with hyperthreading, which means that while it still has four physical cores (for most i7s), each core shows up as two logical cores, with the second being used for smarter load distribution. Essentially, when the first thread is complete, there's a gap in time where the CPU has to fetch new instructions or resources for the next task in the first thread's queue, and during that time, the hyperthreaded thread is processed so that the CPU can get more done in the same amount of time. However, the gaps aren't perfect, and hyperthreads don't end up having the same performance as normal threads because of the fact that they have a lower priority. Because of this, if a game is set to run on hyperthreaded logical cores rather than the standard physical ones, you might actually get worse performance than if you had hyperthreading off (or had an i5).

This is evident in a couple of benchmarks with CPU-bound games that don't support many threads, like SC2.

http://images.anandtech.com/graphs/graph4083/35050.png

Something to note however is that these results are within the margin of error and that the i7 was running at a 100MHz higher clock (although you could very easily set the i5 to run at that).

However, issues like this are rare, and you can force your game to work around them, so I'd just get the i7 if you have a choice.

Both run TF2 more than good enough. TF2 can only use 3 threads.

i7 will let u stream just about anything

Is an i7 better than an equivalent i5 for TF2? Definitely.
Do you need an i7 if you're not streaming? Probably not.

The Xeon E3-1230 is worth a look, wait for the v4 or v5 though, should be released this summer. New i7s coming in August aswell.
FYI the E3-1230 is basically an i7 without the iGPU, but it's not like you were going to use it anyway, so why pay for it?
It's specifically the 1230 (there's also the 1231, an updated version in this generation, but usually it's just the 1230), the 1220 is only a Quad Core without Hyperthreading (=i5), for the 1240 and further you pay a huge premium for marginally higher clockrates and the version ending in 5 (or 6 aswell in this gen) are exactly the same as the ones ending in 0 (1) except they do have an iGPU and you obviously don't want to pay for that.

#5
Do I have to post the screenshot of TF2 using 8 threads again?
We've been over this before.

I'd like to see that screenshot, it you would

http://i.imgur.com/Eelw9IZ.png

That's pretty much just 3 threads of actual usage spread out over 8 logical cores.

Do you even know what threads are?

No, he doesn't.

But since when does TF2 support more than 3 threads? This is completely news to me.

I don't even know where that myth comes from, iirc it always supported at least 4 threads.
http://hardforum.com/showthread.php?t=1406716
That's right after the update that added multicore rendering.

It might be due to the 1 management/main thread + # of cores - 1 worker threads setup and Quad Cores being the max at that time. http://techreport.com/review/11237/valve-source-engine-goes-multi-core

But you've been able to force pretty much any number of threads with -threads for quite a while now.

Is there a way to see the % increase in performance for a i5 vs i7 (i know there are different cards in the series but just generally) for tf2?

0%

you won't get any benefit from an i7 in most gaming scenarios, and you'll get worse performance in some scenarios. spending the extra $100 to go from i5 to i7 for gaming is moronic.

I've been running an i5-3450 and a B75 mobo for nearly 3 years now and have considered upgrading to an i7 with the release of Skylake. I use my computer for games, 3D renders / visual effects, very large filesize digital painting, and heavy video editing.

Is it worth buying a Skylake i7 for the DDR4 memory, and other various chipset features? Would I benefit more from buying a similarly priced Xeon?

ddr4 has almost no current improvement for your workflow, but it's a necessary incremental upgrade for the future. As for the i7 itself, having more threads will always help with video editing and 3d rendering to a major degree.

regarding Xeon, you save money at the cost of lacking integrated graphics, which do occasionally have their uses - quick sync is neat and having an iGPU makes debugging a broken system easy if the GPU is the problem

Specific to this discussion, a single thread allows (at most) 100% usage of a single logical core of a cpu. The equivalent load may be spread out over multiple logical cores due to windows, however.

Just get an overclockable i5 (if you don't plan to stream) or an overclockable i7 (if you do stream). TF2 wants GHz, and you can easily feel a frame rate improvement just by overclocking to something like 4.2 GHz

Thanks for the advice. I am currently running pretty low speed RAM(1600mhz), so would an upgrade to DDR4 be worth it if I'm going for the 2400Mhz modules?

TFW you just got best RAM you could allow yourself to buy and you read next day that it's pretty low speed :/

For most tasks there is little difference between DDR3/DDR4. Furthermore, there is also very little difference between 1333~2400MHz on just DDR3. There are a few places where a higher frequency helps you (if you happen to be running an APU or professional programs for example), but as just a gamer you don't need that much.

Also, I'm not sure if you know this but an upgrade to DDR4 requires a new CPU and motherboard. You can't just stick a DDR4 DIMM in a DDR3 slot with a DDR3 CPU.

Yeah, I've been budgeting for that as well. I'm not too concerned with gaming, because I run my serious multiplayer games on ridiculously low settings. I've just been wondering if a RAM upgrade will help with software like After Effects, Premiere and Final Cut Pro as I know that at least AE uses massive amounts of RAM and has a pretty bad time with memory management.

Also, forgot to mention I'd ideally like to upgrade to hardware that is OS X compatible(and fairly hassle free) as I'm currently running a hackintosh both at home and at work.

This is going to be a long post.

#14
Benchmarks.
http://www.teamfortress.tv/thread/7598/tf2-benchmarks
Get someone with an i7 to timedemo the demo with and without Hyperthreading. The only difference between a true i5 and an i7 with Hyperthreading disabled is that the i7 has a bit more L3 cache so the result should be fairly accurate, if anything the "simulated i5" would actually score slightly better than a true i5.

I would do it but TF2 is incredibly bugged for me.
Here's a list of the average results (5 runs).
Multicore Rendering disabled: 83.07 fps
Multicore Rendering enabled and
-threads 1: 83.87 fps
-threads 2: 133.39 fps
-threads 3: 134.37 fps
-threads 4: 135.19 fps
-threads 5: 136.07 fps
-threads 6: 136.11 fps
-threads 7: 136.10 fps
-threads 8: 135.84 fps

I'm hoping this is just me, because otherwise this means two things:
1. I should've recommended the G3258 all along.
2. The TF2 engine is so far beyond saving that only a port to Source 2 can save us now. In that case I could totally understand Valve not fixing anything anymore, it'd be like spitting on a grease fire. Best case nothing happens and you just wasted time, worst case it blows up in your face.

#15
He asked specifically about TF2, which is the exception to the rule, unless... See the answer to #14.
I agree about the +100$ though. +50$ for a Xeon is ok, +100$ is meh, unless you absolutely need it.

#16
Very simple. Does it use more than 4 threads? -> You'd benefit from Hyperthreading.
At the same speed there's almost no difference between DDR3 and DDR4.
Can you name any chipset features that you're missing right now? I doubt it, but feel free to prove me wrong.

The idea behind buying a Xeon is actually not buying a similarly priced Xeon.
I'll show you with Haswell Refresh as an example:
i7-4790 4.0GHz 312$
Xeon E3-1271 v3 4.0GHz 339$
Xeon E3-1241 v3 3.9GHz 273$
Xeon E3-1231 v3 3.8GHz 250$

For comparison:
i5-4670/4690 3.8/3.9GHz 224$

I guess you can see now why the 1231 is sexy as fuck if the workload benefits from Hyperthreading.

There'd also be the option of running ECC RAM, but I don't know if you need that.

#17
Exactly.

#18
Not even close.
You're thinking of processes and even then you're far from the truth.

Let me break it down to ELI5-level.

A CPU is basically a really complicated calculator.

Now if you take a number of instructions, for example
[process]
1. 4 + 3 = x
2. x + 7 = y
3. y + 2 = z
4. display z
[/process]

That would basically be a process. It would also be a thread, you can follow the order of the instructions like a thread.
Now if a thread doesn't wait on it's own (e.g. "wait for 5 seconds") or has to wait (e.g. "wait for input") it'll run as fast as it can, just as if you'd be typing it into a calculator as fast as you can, meaning 100% core usage. On just one core. Any additional cores/calculators won't do anything.

Now if we want to take advantage of a multicore CPU we'd have to rewrite those instructions a bit.

[process]
1. 4 + 3 = x
2. 7 + 2 = y
3. x + y = z
4. display z
[/process]

This will obviously give you the same result. But it still won't use more than one core. And windows won't help you with that. It should be especially obvious with the first process, windows can't just willy nilly change your program based on what it thinks is right. If a program hasn't been programmed to use multiple cores it simply won't. Even if the programmer has already changed to order like in the second example windows is still not allowed to just rip the program apart and run it on two cores. It has to be done manually, a bit like this:
[process]
1. start 2nd thread
thread 1 runs here in the "main process" ______ thread 2 runs over here
2. 4 + 3 = x ______________________________ 7 + 2 = y
3. wait for thread 2 to finish _________________ end this thread
4. x + y = z
5. display z
[/process]

The only thing the operating system (windows) has to do with the whole multithreading is scheduling. Only one thread can run on each core at a time so someone has to organize that. Now if your OS is stupid or if you have only once core, it'll run step 1, then step 2 & 3 of either thread 1 or 2, then step 2 & 3 of the other thread and then step 4 & 5. That's 9 steps, way slower than variant 1 or 2 with just 4 steps. However if you have another unused core and the OS utilises it, it'll go like this: Run step 1 on core 1, run thread 1 step 2 & 3 on core 1 and at the same time run thread 2 step 2 & 3 on core 2, then run thread 1 step 4 & 5 on core 1. But it still takes 5 steps compared to 4. This is only worth the effort if step 2 takes far longer than step 1 & 3.

And now you know why multithreading is such a huge pain in the ass.
-worst case you basically have to rewrite the program
-it's not always possible
-it might not even be faster

#19
See the answer to #14.

#20
Considering that 1600MHz is the highest speed your CPU (and any other Intel CPU since) officially supports I wouldn't call it low speed.
Latency is going to pretty much the same so the only thing you gain from 2400MHz RAM, be it DDR3 or DDR4, is bandwidth. And you are probably not bandwidth limited. Going from single to dual channel rarely makes any difference and if it does, it's usually 5% or less, definitely never above 10%. That said, take what you can get, if there's no noticable premium on DDR4 and DDR4 mobos by the time you buy, you might aswell go for it.

#21
See the answer to #20.

EDIT: Added multithreading ELI5. Please don't kill me, I know it's horribly inaccurate, but it's the simplest way I could put it.

EDIT2: Fixed two typos.

I've always loved your posts. Thanks for the help.

Oh, and I definitely don't need ECC RAM.

Trying not to derail this discussion any further, but you're right, and sorry you had to type all that out (I'm quite familiar with c++ and multithreading). I was just trying to explain that lack of thread affinity does not mean that tf2 can simultaneously use 8 cores.

On another note, I'd be curious to see if the results of your testing with different -threads values were different when using threadpool_affinity 0.

My thought process behind buying an i5 over an i7 was to spend the extra money on a water cooling system and then OC to 4.6+ ghz. i7 is better just depends on how much money you have to spend.

#25
Thank you.

#26
Sorry, it just sounded so wrong after everything I'd read that day.
I see your point now, but the low CPU usage was probably due to fps cap I was using at the time (for streaming).
Also shuffling all threads around would be weird, even by windows' standards, but it would explain everything. It would mean that TF2 runs into a single threaded bottleneck at two threads already. G3258 confirmed best CPU for TF2. This is exactly the nightmarish shit that kept me from thinking about TF2 performance.
threadpool_affinity doesn't change anything for me.

Could you test -threads 1-4 aswell, if you have a Quad core? Or at least an i3/FX-4xxx, 4 logical cores should do just fine for this test.

#27
Why do people think watercooling is a good idea? Spending 140$ on an AiO makes no sense when a 70$ air cooler gets you the same performance and is quieter.
Assuming TF2 uses 8 threads, if HT just gets you +30% were talking 5+GHz for the i5 to pull even.
If TF2 only uses 2 threads a Pentium G3258 (65$) will get you the same peformance as a 230$ i5-4690K if you manage to get the same clockrate. And even with a 140$ AiO, which still makes no sense, it would be cheaper than the 4690K without a cooler. Or you know you could get a Xeon E3-1231 v3 for 250$ and still save money because there's no reason for a Z97 mobo anymore. Getting an i5 only makes sense if you're absolutely sure you need 3-5 threads. Only 2 and the Pentium will get you the same performance at a fraction of the price, 6 or more threads and you're better off with the Xeon.

I was wondering if any SSDs currently on the market benefit significantly from SATA express, as most of my rendering and file exports tend to get bottlenecked at the writing to storage stage, so having faster storage to me is a benefit.

Just get PCIe one :D

https://www.youtube.com/watch?v=slORqSKElRI