GTA5 CPU Benchmarks (I3 beats FX yet again)

Martini1991 · 25 Apr 2015 at 16:06

That's my i7 and 290X at 2560x1080 with everything on low.
4770K is 4.4, the GPU is at stock.

Talk about API bottleneck.

I'm just driving about in the dark on that.

ALXAndy · 25 Apr 2015 at 16:08

humbug said:
Tell ye' what.

In terms of performance i would have an FX-63## over any i3
For performance its a close call between a Haswell i5K and an FX-83##, For cost, Value for money its the FX-83## hands down.

For performance a Haswell i7 is a juicy prospect, but the price at x2 that of an FX-83## is bonkers, the i7 price IMO compare it to nothing is just plain bonkers.

But here is the cracker, what i really want is an Xeon 5650, more than a Haswell 4c 8t i7, i tried like hell to get one when they were going for £65 but a good X58 motherboard was hard to get and very expensive, the price of that CPU has now gone up, as have the boards. £150 for a good used one.

Any humble AMD user will tell you exactly the bloody same. IE - I would love an I7 but they're bloody expensive. It's not like they're deluded they know exactly what is what.

Yet still the slurs and the attacks continue with aplomb.

humbug · 25 Apr 2015 at 16:13

I know, right ^^^

Martini1991 said:
That's my i7 and 290X at 2560x1080 with everything on low.
4770K is 4.4, the GPU is at stock.

Talk about API bottleneck.

I'm just driving about in the dark on that.

That's not an nessesarily an API Bottleneck.

Core thread 1: 86%
Hyperthread 1: 45%

Core thread 2: 81%
Hyperthread 2: 38%

Core thread 3: 72%
Hyperthread 3: 36%

Core thread 4: 63%
Hyperthread 4: 28%

You have fairly high load on your 4 main threads there, there is quite a lot going on even if you can't see it. if it was an API Bottleneck only one of the main threads would have a high load on it.

Martini1991 · 25 Apr 2015 at 16:26

humbug said:
I know, right ^^^

That's not an nessesarily an API Bottleneck.

Core thread 1: 86%
Hyperthread 1: 45%

Core thread 2: 81%
Hyperthread 2: 38%

Core thread 3: 72%
Hyperthread 3: 36%

Core thread 4: 63%
Hyperthread 4: 28%

You have fairly high load on your 4 main threads there, there is quite a lot going on even if you can't see it. if it was an API Bottleneck only one of the main threads would have a high load on it.

Either way, all settings raised, CPU usage doesn't actually change much at all.
I also don't know which cores are which.

Martini1991 · 25 Apr 2015 at 16:39

And here's the settings ;

GTA 5's VRAM calc is way off according to AB.

At any rate, CPU usage barely moved from absolute lowest settings to pretty high settings. So yeah.

Scurburg · 25 Apr 2015 at 17:06

Martini1991 said:
If I owned a Titan X I'd be disappointed too.
I'm just sick of the current GPU performance from a single GPU. Going from a 7970 which debuted in the dying days of 2011, I don't feel we've gained anywhere near enough performance.

Now that I agree with. I was hoping that there would be something that would out do or at least perform on par with dual 280X's in xfire on a single gpu at a reasonable price by now (as you said, its 2011 tech, is that unreasonable?). Seems my only options are a fairly well clocked 780ti, a 980 or a titan. None of which I would call reasonable in terms of price right now. No way will I buy nvidia's 3 1/2 wheeler. Adding a 7990 for 3way madness actually seems like a more appealing option atm. Trying to patiently wait for AMDs new cards in the hope it will change the market in my favor, but I am struggling to wait!

humbug said:
Tell ye' what.
But here is the cracker, what i really want is an Xeon 5650, more than a Haswell 4c 8t i7, i tried like hell to get one when they were going for £65 but a good X58 motherboard was hard to get and very expensive, the price of that CPU has now gone up, as have the boards. £150 for a good used one.

I was exactly the same and thought it looked like a great option until the price rise, it is still possible if you are patient and vigilante. I think I got lucky while searching and ended up with a 3930k with mb for a similar price, although I really do hate this MB and/or the 2011 style of OCing. I've said it before but the FXs are more fun and flexible to OC IMO.

Anyway back on topic as I've blathered on enough. As someone said, we really do need more i3 benches for comparison along with frame times. Does AB measure these accurately enough?

humbug · 25 Apr 2015 at 17:27

Martini1991 said:
And here's the settings ;

GTA 5's VRAM calc is way off according to AB.

At any rate, CPU usage barely moved from absolute lowest settings to pretty high settings. So yeah.

Core thread 1: 93%
Hyperthread 1: 49%

Core thread 2: 70%
Hyperthread 2: 38%

Core thread 3: 70%
Hyperthread 3: 34%

Core thread 4: 79%
Hyperthread 4: 43%

Thats an API Bottleneck ^^^^

Core Thread 1 is significantly higher than most others and just about flatout, thats your Draw Call thread.

No amount of threads will cure that as you only get one Call thread in DX11.

Having said that its impossible to know IF that thread is shared with other tasks, if it is then it is so because that task has no where else to go given that all other Main threads are fairly high already, if you had more Main threads it would take the load off and reduce the latency in that thread thus improving performance.

mw8t · 25 Apr 2015 at 17:28

Wow wasn't expecting those results. I'm glad I haven't got an AMD chip.

Martini1991 · 25 Apr 2015 at 17:33

humbug said:
Core thread 1: 93%
Hyperthread 1: 49%

Core thread 2: 70%
Hyperthread 2: 38%

Core thread 3: 70%
Hyperthread 3: 34%

Core thread 4: 79%
Hyperthread 4: 43%

Thats an API Bottleneck ^^^^ Core Thread 1 is significantly higher than most others and just about flatout, thats your Draw Call thread.

No amount of threads will cure that as you only get one Call thread in DX11.

Having said that its impossible to know IF that thread is shared with other tasks, if it is then it is so because that task has no where else to go given that all other Main threads are fairly high already, if you had more Main threads it would take the load off and reduce the latency in that thread thus improving performance.

Assuming what you've said is correct, an AMD CPU core is no where near the performance of my Haswell core, so the AMD core's going to be maxing out much much sooner is it not? How can you call that an API bottleneck then? To me that's a CPU bottleneck if I had my draw call core maxed out (Inefficient API perhaps, but higher core performance would help)

To me, I'm GPU bottlenecked (As my GPU's maxing out, clocking it higher results in higher performance).

I was API bottlenecked before as nothing was maxed out (In my opinion and understanding...)

Bonjour · 25 Apr 2015 at 17:33

Martini1991 said:
I'd charge your battery

I don't need you to prove anything to me. I'm just a guy on the internet.

No worries, dude.

Reading back over my posts in the thread, I hadn't realised I come over so biased. I should rein it in a little..

Martini1991 · 25 Apr 2015 at 17:41

humbug · 25 Apr 2015 at 17:42

Martini1991 said:
Assuming what you've said is correct, an AMD CPU core is no where near the performance of my Haswell core, so the AMD core's going to be maxing out much much sooner is it not? How can you call that an API bottleneck then? To me that's a CPU bottleneck if I had my draw call core maxed out (Inefficient API perhaps, but higher core performance would help)

To me, I'm GPU bottlenecked (As my GPU's maxing out, clocking it higher results in higher performance).

I was API bottlenecked before as nothing was maxed out (In my opinion and understanding...)

If that 93% loaded thread of yours is purely the Draw Call thread, not shared with any other task then yes, that is absolutely correct.

If not, if there is a task or more also running on that thread the API would have anyone or all of another 4 threads to move tasks around to reduce the load on it.

Martini1991 · 25 Apr 2015 at 17:47

I've got some interesting results to show you.

humbug · 25 Apr 2015 at 17:49

Martini1991 said:
I've got some interesting results to show you.

Okay

Martini1991 · 25 Apr 2015 at 17:52

Had to take the picture on my phone as AB's being weird.
Same settings as previously. Just changed thread priority and unticked core 0 in affinity.

Now how'd you explain that

?

Scurburg · 25 Apr 2015 at 17:57

The bear seems to have taken on the work. But on a serious note, wth is going on there. Has it affected how it plays or anything?

humbug · 25 Apr 2015 at 17:58

Martini1991 said:
Had to take the picture on my phone as AB's being weird.
Same settings as previously. Just changed thread priority and unticked core 0 in affinity.

Now how'd you explain that ?

Is that exactly the same view in the game as the last one you posted?

Martini1991 · 25 Apr 2015 at 18:02

humbug said:
Is that exactly the same view in the game as the last one you posted?

Just driving through the dark in the same street as per the other 2 shots.
Anyway, MSI AB started working again, and then CPU usage started going up as it became lighter.

Hilariously frame rate's the same.

That CPU 1 core went back down quickly too.

I'm still pretty sure it's just a GPU bottleneck.
No API bottleneck.

Martini1991 · 25 Apr 2015 at 18:04

Scurburg said:
The bear seems to have taken on the work. But on a serious note, wth is going on there. Has it affected how it plays or anything?

Plays smooth as always at night.

gpuerrilla · 25 Apr 2015 at 18:08

So your 290x is the bottleneck with an i7. I thought you had an i5?