Upgrading - AMD FX8350 or Intel 4770K?

[Hagar]

I don't see why anyone would go AMD right now apart from loyalty tbh, I reckon those 8 slow cores will get thread limited in games in the near future if they aren't already in some games

my 4670k @ 4ghz is already pulling 73% cpu usage on the most used core in the most cpu intensive game I own.
surely the awful single core performance of the amd chips will be a huge issue very fast for gaming.
when in games that multicore well you're hitting 73% cpu usage and 50% on every other core you knows it's going to make an 8350 keel over because that 73% cpu usage clearly cant be split between threads any better than it already is.

Cinebench 11.5 - Single Threaded
Score in CBMarks - Higher is Better
fx 8350 = 1.1
haswell 4670k = 1.7

thats a huge difference in core for core performance

Agreed if you are only or mainly playing games that rely on the single core performance. There are plenty of benchmarks that show the 8350 on a par or slightly better than the 3570 for multicore game engines.

I am not going down the fanboy alley. Both are very good gaming processors with the caveat of lower single core performance on the piledriver.

The 8350 is far better than the 1090T which I had previously in the same machine. It maxes the (admittedly midrange) overclocked HD6950 at 99-100% GPU usage at 40-50fps in ARMA2. Very playable. And without the CPU bottlenecking.

For BF3 it is at least as good as the 3570 and close to the 3770. Normal resolutions.

I quote the last generation Intel as there is more data available.

My 8350 will overclock 25% above stock and game at 5GHz. For encoding and rendering tasks it is similar.

Prime or IBT will fall over above 4.7-4.8Ghz on eight threads but I have not yet seen many posted prime runs on Haswell to show stability when overclocking.

It is largely down to personal preference or pricing. Both are good gaming platforms IMHO.

 

[Martini1991]

Oh, I agree, if you want a meaningful core for core performance you've got to go Intel.
a lot of the newer games are GPU bottlenecked, so you'll them being "on par" but that doesn't mean they'd be on par with x or y GPU set up.

It's also very hard to say it's down to personal preference when one is factually better in the majority of games.
On the other hand, you could spend your life playing Y game which can use 8 cores and the FX would be the better choice.

 

[arknor]

There are plenty of benchmarks that show the 8350 on a par or slightly better than the 3570 for multicore game engines.

Did you look at the cpu usage in project cars? a next gen game

just incase you missed it somehow

Imagine an 8350 running that, imagine what would happen when one core wants the same cpu usage my haswell is getting
We all are in agreement
4 haswell cores are almost equal to 8 amd 8350 cores right? (with haswell winning more benchmarks than the 8350 but it's very close)
which means 1 haswell core is almost equal two 2x 8350 cores.

so lets take those pcars cpu usage figures and double them
82.4% , 146.8%, 71.2% , 95%
Bold core is massively thread limited surely no one can deny this?

if it were possible to distribute the load on that core the game would already be doing it look how even the other 3 cores are that's probably the most even multicore performance i have ever seen in a game

btw heres bioshock infinites cpu usage and that was on the boat at the start which is pretty much a cutscene so I doubt theres that much cpu calculations happening compared to when your rushing around inside the city area.

May aswell times them by 2 your self

Doesn't take a genius to figure out
2 cores are thread limited on an 8350 in bioshock , the game is not capable of spreading the load between more cores or it would be.

Btw I'm not cherry picking results these are just 3 shots I posted in the haswell thread the other day and I think the only 3 modern games I have installed that aren't indie games
the good news is skyrim will run great on an 8350

 

[Hagar]

Did you look at the cpu usage in project cars? a next gen game

just incase you missed it somehow

Imagine an 8350 running that, imagine what would happen when one core wants the same cpu usage my haswell is getting
We all are in agreement
4 haswell cores are almost equal to 8 amd 8350 cores right? (with haswell winning more benchmarks than the 8350 but it's very close)
which means 1 haswell core is almost equal two 2x 8350 cores.

so lets take those pcars cpu usage figures and double them
82.4% , 146.8%, 71.2% , 95%
Bold core is massively thread limited surely no one can deny this?

if it were possible to distribute the load on that core the game would already be doing it look how even the other 3 cores are that's probably the most even multicore performance i have ever seen in a game

I don't think it works like that, but I will try to locate a demo and post some results. the FX8350 scores 10/10 in their hardware checker.

The 8350 has eight integer cores and shares four FP units with either eight 128bit instructions or four 256bit instructions. I believe the Haswell 4670 has four integer cores with four FP units I assume 256bit instructions.

 

[Martini1991]

Skyrim, at least at the FX8350's launch was pretty poor on the AMD CPU.

 

[arknor]

I don't think it works like that, but I will try to locate a demo and post some results. the FX8350 scores 10/10 in their hardware checker.

The 8350 has eight integer cores and shares four FP units with either eight 128bit instructions or four 256bit instructions. I believe the Haswell 4670 has four integer cores with four FP units I assume 256bit instructions.

but if you bother to look at benchmarks you will see in highly threaded applications they get the same score almost in pretty much everything.

haswell only has 4 cores
8350 has 8 cores

logic suggests a haswell core is equal to 2 8350 cores or in highly threaded benchmarks they wouldnt get almost equal scores.

doubling the cpu usage surely makes sense? if the game could split the load over the cores any better it would already be doing it on the other 3 cores on my haswell, pcars is surely is a good indicator of cpu usage in games that are in development now like that game is itself

btw forgot i had WOT installed which is a good example of a game that just doesn't want to multicore like pretty much most mmos and f2p games

I got 50% of that fps on my x6 1055t because I was thread limited so on an 8350 I would probably get about 60% of the fps in WOT than I do on my haswell.
but thats an extreme example of a game that only uses 1 thread and a tiny bit of a second thread.

simcity would also be a good example as that only uses 1 thread max, but I can't be bothered to launch that awful game and take a screenshot of my afterburner OSD stats

I just don't see why anyone who games would buy an AMD cpu right now and the only reason I would get one for rendering/transcoding is if I couldn't afford a 4770k (4770k's hyper threading makes it quite a lot faster than a 4670k in highly threaded applications.)

 

[Hagar]

but if you bother to look at benchmarks

I do bother.

logic suggests a haswell core is equal to 2 8350 cores or in highly threaded benchmarks they wouldnt get almost equal scores.

doubling the cpu usage surely makes sense?

The threads are scheduled by the OS and whether there are 2,3,4,5,6,7 threads available they will be allocated where possible and then the processor will do the work and pass it back.

How fast this is depends on the request, processor speed, cache hits and a lot of other factors such as wait times on the ram etc.

I got 50% of that fps on my x6 1055t because I was thread limited so on an 8350 I would probably get about 60% of the fps in WOT than I do on my haswell.
but thats an extreme example of a game that only uses 1 thread and a tiny bit of a second thread.

Subjective but possible.

I just don't see why anyone who games would buy an AMD cpu right now and the only reason I would get one for rendering/transcoding is if I couldn't afford a 4770k (4770k's hyper threading makes it quite a lot faster than a 4670k in highly threaded applications.)

Because they perform as a good gaming CPU, may not be the highest or fastest but considering the number of people who do game but also use their PC for other tasks. Cost may also be an issue and available equipment motherboards etc.

I am not getting into a argument on every little benchmark. But you seem to have suddenly seen the light and converted totally to intel based on a weeks experience with Haswell and from a 1055T which I agree was not the quickest AMD processor. I had one and soon swapped for a 1090T for overclocking.

I would quite happily use a 4670, 4770 or what I have and overall get a similar experience from the screen, mouse and keyboard which is the main thing.

Enjoy.

 

[arknor]

The threads are scheduled by the OS and whether there are 2,3,4,5,6,7 threads available they will be allocated where possible and then the processor will do the work and pass it back.

How fast this is depends on the request, processor speed, cache hits and a lot of other factors such as wait times on the ram etc.

windows does not determine the core load of games that is determined by the games developers and how much they are able to split up the games processes, the game tells windows how many threads it can take advantage of, windows distributes these threads to the cores , it can not alter the max amount of threads a game can use and it can not split these threads into 2 seperate threads.

so I'm not entirely sure what you are trying to get at.
if a game only uses 1 core it only uses one core regardless of what windows wants.

if your talking about core efficiency scaling or something then you can get as visible result with cinbench

3.91 efficiency if multicoring perfectly scaled then that would say 4 (1 core = 100 percent efficient x4 cores would be a max result of 4 which is impossible due to the nature of core scaling) (btw hyperthreading is designed to take advantage of that wasted core efficiency by using the other wise stalled cycles of each core whilst they are waiting for new work)

Subjective but possible.

Not really subjective though is it because if you disabled 4 cores on a 8350 so it has the same number of cores as a 4670k then it would be half as fast as a 4670k at best seeing as it's almost even with 8 cores

 

[Hagar]

Not really subjective though is it because if a 8350 had the same number of cores as a 4670k then it would be half as fast and that is a fact

Yes it is because it also depends on the frequencies, not only of the processor and cache memory but also the ram, the motherboard and disk access times. Delays due to waits from various interfaces not being ready will affect even a single core process.

 

[arknor]

yea im 4ghz @6.74

clock for clock haswell is faster so I don't get your point? what are the super PI times of a 8350 @ 4ghz with 1600mhz ram?

Spoiler

this will be interesting because haswell absolutely dominate ivy bridge/sandy bridge in super pi

EDIT found a benchmark at 3.9ghz and 1633mhz memory for the 8350

http://www.legitreviews.com/article/2055/11/
you really do not want to divide that time by 60 to work out how many minutes it took compared to how many it took a 4670k

Yes it is because it also depends on the frequencies, not only of the processor and cache memory but also the ram,

Btw superpi is all about memory/cache if you weren't aware.


anyway I have done my best to present a good argument as to why haswell is the better choice.
anyone wondering can look at the data and make up their own mind, I'm not trying to force anyone to go intel just giving them the chance to make an educated decision based on their budget.
if your spending haswell money then imo there's no choice to make.

if your budget does not allow haswell and you want brand new stuff then go amd and I'm sure you will be happy with the performance per £

 

[Phixsator]

Its funny how the arguement is still single core performance beats everything else when we are clearly starting to move away from that. With the increased diffficulty in adding further single core performance and the relative ease of just adding more cores single core performance shouldnt be the only thing to evaluate a purchase on, not even when it comes to games. Crysis 3 is not the only game that likes the AMD FX series just as much is it likes Intel, we got Battlefield 3 and even ARMA 3 likes the FX series. Look here for proof of that. So no, the lists of games that runs poorly on AMD cpus isnt that big to begin with and you would get more than enough FPS in pretty much every single GPU scenario.

 

[Martini1991]

Those results don't really show much, you can't work out if it's a CPU bottleneck that's hampering the results, as they're all stock.
It also shows core for core performance is still king.
The second test also shows it running poorly on AMD FX 8150, but we'll ignore that

 

[ussie]

Just get the 7970 and stick with the 9350, yes haswell is better, you are paying loads more money. But in reality it is not going to be a godly difference between the two that you'll see big frame rate differences for example in games.

For a more bang for buck get ur 7970, 8350 (nice cooler and overclock)
To further increase performance you can opt in for an SSD.

 

[Martini1991]

I don't understand advising a top end GPU that's bottlenecked and using the "You won't notice the difference" line.

May's well go lower GPU such as the 7950, saving more money.

7950 and FX8320 is a pretty potent combination.

 

[Phixsator]

Those results don't really show much, you can't work out if it's a CPU bottleneck that's hampering the results, as they're all stock.
It also shows core for core performance is still king.
The second test also shows it running poorly on AMD FX 8150, but we'll ignore that

No offense meant but who gives a rats behind about the 8150?, the only interesting here is the fx-6300 and upwards and sure the bench does show that the game in its current form isnt that highly multithreaded above 4 cores but it also shows for the time being that AMDs lineup can follow closely to what intel has to offer in this specific title. And if you believe that this bench is bad for the AMD side i would argue its even worse for the intel with the difference i3 and i5 being 2 frames. I think the most important thing to remember is the "alpha" next to the Arma 3. Just a thought.

 

[Martini1991]

You can't ignore the FX8150.

I never said it's bad for the AMD side either, it shows core for core performance is the way to go, as it's obvious that engine isn't that heavily threaded (Hence the i3 and i5 result, although I'm unsure why you're trying to say that's a negative for Intel lol), the 3930k's cores have more cache a crap ton more bandwidth too than the 2600k.

It's your proof, not mine.

I wonder how things would look if everything was clocked as far as they'd go?

 

[Phixsator]

You can't ignore the FX8150.

I never said it's bad for the AMD side either, it shows core for core performance is the way to, as it's obvious that engine isn't that heavily threaded (Hence the i3 and i5 result, although I'm unsure why you're trying to say that's a negative for Intel lol), the 3930k's cores have more cache a crap ton more bandwidth too than the 2600k.

It's your proof, not mine.

I wonder how things would look if everything was clocked as far as they'd go?

We all know that the Intel lineup has stronger single core performance but that isnt showing in the bench, cause if it was even the 2500k should be ahead of the 8350. Also in Arma 2 the fx 8350 was able to keep up with the 3570k which is only be possible if the game has some kind of proper threading cause otherwise the 3570k should have ahead by a bit. My reasoning for bringing up Arma 2 is that i cannot see why the same thing wont end up in Arma 3, that being proper multithreading and if that is the case well then results will look different when we get to that point and what if Intel then only has single core performance to push its result?

 

[Martini1991]

Of course single core performance is showing in the bench, just look at the i3 2100's results.
There simply isn't enough data in that benchmark, if we see overclocked results, we might be able to draw proper conclusions.

http://www.behardware.com/articles/880-13/amd-fx-8350-review-is-amd-back.html

The FX8350 also isn't keeping upto the 3570k.

I'm not about to get into an AMD/Intel argument, you're the one who posted a link as proof for games not running poorly on AMD, yet the FX8150 is running rather meh (Oh, but we can conventionally ignore that, even though it's newer than the i3 2100 and has quadruple the core count, after stating that core for core performance isn't as important, as things are changing).

ARMA 3 doesn't look significantly more threaded than ARMA 2. Or the giant over 50% exta potential of the 3930K would be realised over the 2500K.

 

[Phixsator]

Of course single core performance is showing in the bench, just look at the i3 2100's results.
There simply isn't enough data in that benchmark, if we see overclocked results, we might be able to draw proper conclusions.

If thats the case then why isnt the 2500k ahead of the 8350 like the 2600k is? While i agree that OCing will give a better picture of what is going on i dont think we can get a proper conclusion until the game is out and their design choices of the engine is final.

 

[Martini1991]

If thats the case then why isnt the 2500k ahead of the 8350 like the 2600k is? While i agree that OCing will give a better picture of what is going on i dont think we can get a proper conclusion until the game is out and their design choices of the engine is final.

Why isn't the 3930K leagues ahead of the 2500K? (It's 4 FPS higher on average, with over 50% more performance in total, yet the higher clocked 3970 is 6 FPS higher than the 3930K, with only cache/clock being the difference)

The 2600K is no more ahead you're also forgetting it has higher clock/more cache = More core for core performance.

It's definitely CPU limited. Reason for the FX8350 even with its IPC deficit could be instruction sets as we've seen with Crysis 3.

I mean the i5 to i3 difference proves it's not massively threaded, HT on a Sandy i3 would give a possible maximum performance boost of 30%, whereas the 2500K has 4 higher clocked cores, giving it well over 70% performance that it can achieve, but that's translating to what, 5% performance increase? That 5% difference is pretty much the clock difference.