Is the i5 still enough? Bottlenecking explained with an RX-480.

[Majestic]

Bottleneck test.

 

I think the most predominantly asked question is: "Does my X bottleneck my Y?" And right after that is probably: "Is an i5 enough for gaming?". Most people would argue something like an i5-4460 or i5-6500 is more than plenty for modern games at 60 fps. So I've grabbed a set of modern games that i'm going to explore in-depth with graphs to see if this is actually the case.

 

Games tested:

Deus Ex: Mankind Divided (DX12), Settings: link

Hitman (DX12), Settings: link

Rise of the Tomb Raider (DX12), Settings: link

Watch Dogs 2 (DX11), Settings: link (overlay didn't want to co-operate)

GTA 5 (DX11), Settings: link, link and link

--New--

Mass Effect Andromeda (DX11), 1080p maxed settings (ultra).

Mirror's Edge Catalyst (DX11), 1080p hyper settings.

 

Games settings are not picked at random, or set to highest. They are set to 60fps is guaranteed (and only just). It's a mixture of High/very high, whatever the game allowed. Resolution is 1080p

 

I've tried enabling DX12 where ever possible to reduce CPU overhead as much as possible. The testplatform is tested at two settings.

1. Overclocked CPU and fast memory (4.3ghz, 2400mhz CL11)

2. Stock CPU and generic memory (3.6ghz, 1333mhz CL9)

 

Take note, this is still Haswell, so don't look at these memoryspeeds and think i'm deliberately nerfing the CPU. 1333 is entry for Haswell. 2400 is fast for Haswell.

 

Test System:

Intel Core i5-4670K 

16GB Corsair Vengeance Pro 2400-CL11

MSI Z87i Gaming AC

MSI RX-480 Gaming X 8GB (set to default 1266 to simulate generic 480).

Crucial BX100/MX100 250GB drives for windows and games.

Samsung C24FG70 144hz 1080p VA panel

 

Results (overclocked = 4.3ghz/2400, stock = 3,6ghz/1333).

 

Deus Ex: Mankind Divided. 

Test entailed starting from your appartment (poor), walking downstairs and making two circles around the appartment complex (outside).

Spoiler

Overclocked:

 

Stock:

 

Overclocked:

Framerate can be read as 20 fps/division, framerate remains above 60fps the whole time but only barely. CPU load is averaging 80%. So if you have an overclocked i5 this game should still be playable above 60fps at high to very high settings. This could be an AMD thing, as I've not seen this on my 980, but the framepacing was pretty bad. A few stutters can be visible in the frametime graph. But it's not as bad as some other games. Let's continue.

 

Stock:

It's no longer able to maintain a surplus of 60 fps at all times. Note the high CPU load, averaging 90-95% and clearly as can be seen from the inconsistent GPU load there is some clear CPU bottlenecking taking place. So a locked i5 is not able to keep this game at 60fps paired with a mid-range RX-480 at the settings this card was designed for. You need to drop geometry related settings, resulting in less graphical fidelity. You can also see regressed framepacing performance and infrequent frametime spikes. This was not a pleasing visual experience, even on a freesync panel. Frametimes ranging from 16.5ms to 20ms consecutively. So visually it's constantly switching between 50 and 60 fps frame transitions. It bad.

 

Hitman (2016):

Test involves the Paris mission, walking through the front door and making a circle through the various first-floor crowded rooms, and walking back to the start.

 

Spoiler

Overclocked

 

 

Stock:

 

Overclocked:

Fps was mostly 60-70 fps, with a few dips to high 50's at the start due to assets no doubt loading in. CPU load was at 85-90% average, so not much to spare. But it should be able to maintain 60 fps on an overclocked i5 yet again.

Framepacing wasn't steller again though, pretty sure it's just AMD's drivers and the lack of CPU overhead causing issues.

 

Stock.

Yes, it's as bad as it looks. Framerate might be at a constant surplus of 60fps, but the frametimes tell a different story, as do the GPU load graphs which are all over the place. CPU is pegged at 95%+ for the vast majority of the test and the frametimes suffer. Subsequent frametimes of 16ms to 26ms (60fps vs. 38fps) gives the visual experience of double-buffered v-sync 60hz with the frametargets being missed every other frame. It's jittery and nausiating to watch. A locked i5 paired with a 470 or 480 is not going to have a fluent experience running this game at the designed settings for these cards yet again. 

 

Rise of the Tomb Raider:

Test involved Geothermal valley, running (normal pace) from the village to the camp and back again. Panning the camera a bit to load in additional assets.
 

Spoiler

 

Overclocked

 

 

Stock

 

 

 

 

Overclocked:

Brief periods of CPU bottlenecking with erratic frametimes. But a mostly saturated GPU and frametimes remaining fairly consistent (14-16ms). FPS at 60-70fps average and remaining above 60fps the whole time. DigitalFoundry already pointed this out in the 2500K video they did, that Rise of the Tomb Raider suffers from massive driver overhead from AMD, even in DX12. And this is quite evident in these results. It's mostly a solid experience, but not flawless. I toggled off Tessellation for good measure, this being an Nvidia title, perhaps it was causing havoc. But this really didn't seem to fix it the overhead.

 

Stock:

Wew lad, this is not looking great yet again. Lots of erratic framepacing, a GPU load that is predominantly unsaturated and framerates that barely skimp 60fps. Frametimes can be as excessive as 52ms (<20fps). CPU load seems pegged to 95%-100% and is clearly the predominant performance denominator. It explains the stuttering and crappy frametimes. DigitalFoundry has repeated this multiple times aswell, you generally want to be GPU bottlenecked, as being CPU bottlenecked introduces stuttering. This is again a game that is not going to be fully enjoyable on a locked i5 with an AMD midrange card. I have some results for my 980 on this title. But as this is a i5/bottlenecking thread and not AMD vs. Nvidia, I will save it for the footnotes.

 

Watch Dogs 2:

An infamously CPU heavy game I put in here for good measure. Gamers Nexus has already showed that i5's are out of their league in this game, and my results show nothing different. The test involves driving around the park outside the dedsec hideout twice, and walking around the park with panning camera.

 

Spoiler

 

Overclocked

 

Stock

 

 

 

 

Overclocked:

Fps mostly above 60 fps, but the frametimes range from 15 to 20ms consequtively. The game however has some excellent motion blur that manages to make these sub-average framepacing look fluent still. CPU is struggling however, and there are some gnarly GPU load drops already visible, but this is not directly visible in the frametimes or pacing. This game, unlike many people proclaim, has an excellent engine that manages to drive consistent performance even under a struggling CPU at 60fps. This much can't be said for the other games.

 

Stock:

Wew² Lad³, this is bad. CPU flatlined at 100% and fully driving performance. FPS sub 50fps most of the time and GPU load is literally all over the place. Here I have to admit the pollingfrequency of the capture is probably not fast enough. Because the frametimes don't look half as bad as they felt during gameplay. It's safe to say that for Watch Dogs 2 an i5 is woefully inadequate, especially with slow ram and no overclock.

 

Grand Theft Auto 5

 

This is an older game, and doesn't really fit the narrative of the test. But I've included it for comparisson purposes. The test involves starting at micheal's house and driving around downtown from dawn to midday.

 

Spoiler

Overclocked

 

Stock

 

Overclocked:

Framerate is certainly high enough and the frametimes are good enough to blend well with Freesync. Which is to be expected from an older game, and one that was recently removed from the test lineup at Gamers Nexus because the game engine starts to buckle beyond 180 fps. CPU is driving performance sometimes though, but increasing the AA settings would tone that down a bit, as the framerates would become a little lower.

 

Stock:

CPU is driving performance, saturation takes a hit and a poorer frametransition can be noticed (even through freesync). It's still certainly still playable, and again increasing fidelity would increase the CPU overhead. And alas, the game does not scale beyond 4 cores, so there wouldn't be any way of fixing a CPU bottleneck even if you wanted to.

 

--Updated--

Update is a small set of extra tests of games I've found the i5 to be lacking on. These are only in overclocked configuration and with the GTX 1070 only. The results speak for themselves though.

 

Mass Effect Andromeda

 

Test involves running around the Nexus Area's, where there is both a pretty high CPU load aswell as it not being cherrypicking seeing as you'll spend quite some time there. It is after I finished the game, so only Operations and the Docking Bay area. Spoiler alert.

 

Spoiler

Full resolution (right-click, open in new tab)

 

GPU load is not 99%, and there are the occasional spikes to 90-100% CPU load when loading in scenes. i5 seems to be mostly hanging in there (average 70% load), but take into account this is a 20% OC with fast memory. A stock configuration would be having a much harder time. Still very playable, I'm not certain however the spikes are elleviated with an i7, need someone to verify.

 

 

Mirror's Edge Catalyst

 

Test involves running around the city from the first safehouse to the first mission. Traversing the city in a fast pace and giving the engine a good workout. 

 

Spoiler

Full resolution (right-click, open in new tab)

 

As is evident, even at a moderate framerate of only 70-80 fps (144hz panel now) the CPU is having clear issues keeping up with the 1070. The framerate is predominantly limited by the CPU's ability to generate drawcalls and the GPU is not locked at a 99% load even when the settings are set to hyper. The result is an experience that is very smooth when the CPU is at 80-90%, but starts to spike in frametimes and give visual stuttering when it's being maxed. This is, like Deus Ex and Tomb Raider, a game that simply can't really perform well enough to saturate a mid/high-end GTX 1070 in a free range, and requires hardlocking to 60 to smooth out.

 

 

 

Verdict.

 

I hope these results are Real World enough for everyone. I didn't use a Titan X or super low resolutions or settings to prove my point. This is the experience the average user would be getting with their midrange i5 and graphics cards in the modern titles at a more in-depth and personal level. Average framerates simply don't tell you the entire story and this is something which always bothered me about outlets like Tech City or Hardware Unboxed. They go through all these benchmarks that ultimately tell you jack shit. Gamers Nexus are sufficiently better with 1% and 0.1% lows, but they still test GPU's with high-end CPU's, and CPU's with high-end GPU's. Never putting mid-tier hardware side-by side. Frankly, I have to admit that my i5-4670K is definitely on it's last legs after 4 years, and the i5's that are sold right now are, to me, woefully inadequate for future performance and should stop being recommended (or with the explicit caveat of them not lasting for years). Intel hasn't improved much with Skylake and Kaby Lake, so my overclocked performance should still be a rough indicator for locked i5's of the newer generations. 

 

These new console titles are beginning to require more than quadcores, and to me this should be a more compelling argument for new chips like Ryzen's 5 and 7. 

 

This is also a more in-depth analysis of visualizing bottlenecking. If you witness any of these pictures with your system when running Rivatuner (MSI Afterburner) in the background, it's safe to say your CPU is bottlenecking the hell out of the GPU.

 

So tell me what you guys think, I took the time to make this as inclusive as possible and I will probably be tinkering it a bit based on feedback and no doubt forgetting stuff. If anyone would supply me with results of Nvidia hardware on core i5's with the settings provided, that would be grand. You can look in the graphs which parameters I'm testing, pollingrate is set to 100ms (fastest possible).

 

 

 

 

 

[Majestic]

Reserved for Nvidia vs. AMD results discussion.

 

As was said in the original post, there was also the (in)famous driver overhead to discuss, but I asked for some feedback, lest I had to purchase another card. Though there wasn’t any substantial evidence to consider. Well since I did away with the AMD card for this exact overhead, I can redo the tests with a GTX 1070. A (much) more powerful card, but during a full CPU bottleneck at the same settings, a driver overhead can still be distinguished if present. Hence this is what I did. During the 4.3ghz tests It proved to be the case that the extra GPU headroom proved to result in higher results due to the RX-480 being mainly GPU bottlenecked before. So we should only really look at the 3.6ghz results. Since that’s where I actually saw 100% CPU load scenario’s together with a floating GPU load at the same time.

I only tested the four main games this time, which promted me to do the test in the first place. Deux Ex, Hitman, Tomb Raider and Watch Dogs 2.

 

Deus Ex: Mankind Divided

 

 

 

Overclocked

 

 

Stock

 

 

Hitman 2016.

 

 

 

Overclocked

 

 

Stock

 

 

Rise of the Tomb Raider.

 

 

 

Overclocked

 

 

Stock

 

 

Watch Dogs 2

 

 

 

Overclocked



 

Stock

 

 

 

[The Sloth]

1 minute ago, Majestic said:

I didn't use a Titan X or super low resolutions or settings to prove my point. This is the experience the average user would be getting

yup, this is what i hate about reviews sometimes, lets all find a person who would buy titan x sli than play games in 480 or 720 lowest settings.  

[Majestic]

Just now, nerdslayer1 said:

yup, this is what i hate about reviews sometimes, lets all find a person who would buy titan x sli than play games in 480 or 720 lowest settings.  

There is a place for those tests to measure CPU performance in absolute terms. But when making the case for bottlenecking or a more subjective experience, they fail to add anything tangible.

[Guest]

This guide / review / benchmark is @Majestic

I have a 480 4gb with i5 4460 and my cpu usage never go high on game like crysis 3

 

[Majestic]

2 minutes ago, deXxterlab97 said:

This guide / review / benchmark is @Majestic

I have a 480 4gb with i5 4460 and my cpu usage never go high on game like crysis 3

 

You missed the part where I said I was aiming for 60fps, not 30. That video is showing performance sub 30fps. And even then the CPU is sitting at 70%. Meaning it won't to 60fps.

[david cassar]

16 minutes ago, Majestic said:

 

  Reveal hidden contents

Overclocked:

 

Stock:

 

  Reveal hidden contents

Overclocked

 

 

Stock:

  Reveal hidden contents

 

Overclocked

 

 

Stock

 

 

 

 

  Reveal hidden contents

 

Overclocked

 

Stock

 

 

 

 

 

snip

Interesting results. Would definitely say that i5s nowadays (especially i5s at stock frequencies) get maxed out in a majority of modern titles. The one which comes to mind is easily battlefield 1 64 player multiplayer. You need an i7 to play bf1 at a constant 60 fps.

[Guest]

5 hours ago, Majestic said:

You missed the part where I said I was aiming for 60fps, not 30. That video is showing performance sub 30fps. And even then the CPU is sitting at 70%. Meaning it won't to 60fps.

So this?

Spoiler

 

 

[Majestic]

Just now, david cassar said:

Interesting results. Would definitely say that i5s nowadays especially i5s at stock frequencies  get maxed out in a majority of modern titles. The one which comes to mind is easily battlefield 1 64 player multiplayer. You need an i7 to play bf1 at a constant 60 fps.

Yeah, I think the i5 is now bad value to be honest. I'm waiting for Ryzen 5 to see how they do in terms of overclocking. I'm also a little annoyed that all Ryzen benchmarks were done on Nvidia graphics cards. Would have liked to see how AMD cards perform on those chips, especially in Vulkan or DX12.

[Majestic]

3 minutes ago, deXxterlab97 said:

So this?

 

Yes, at 3:20 you have the Geothermal Valley part. Your CPU is also pegged at 100% and sub 60fps. But these types of benchmark runs are less CPU intensive, due to the fact there is no gameplay engine, NPC's or unknown camera panning.

[david cassar]

Just now, Majestic said:

Yeah, I think the i5 is now bad value to be honest. I'm waiting for Ryzen 5 to see how they do in terms of overclocking. I'm also a little annoyed that all Ryzen benchmarks were done on Nvidia graphics cards. Would have liked to see how AMD cards perform on those chips, especially in Vulkan or DX12.

Well to be fair I think that a highly overclocked skylake i5 should still be fine and for a majority of modern titles an i5 is still enough (for honor for example is pretty well optimised). Also I agree especially since nvidia gpus tend to run better with lower end cpus due to their multi threaded drivers.

[Majestic]

Just now, david cassar said:

Well to be fair I think that a highly overclocked skylake i5 should still be fine and for a majority of modern titles

Which is pretty much what rolls out of the test. My CPU, due to ITX (3phase motherboard) at 4.3ghz manages to mostly skimp along 60+fps. So a 4.5-4.8ghz skylake/kaby lake will do ok. But not for long tbh, it's now pretty much the bar.

[david cassar]

Just now, Majestic said:

Which is pretty much what rolls out of the test. My CPU, due to ITX (3phase motherboard) at 4.3ghz manages to mostly skimp along 60+fps. So a 4.5-4.8ghz skylake/kaby lake will do ok. But not for long tbh, it's now pretty much the bar.

True. I was recently doing a budget build and was gonna go with a locked sandy bridge i5. Decided to step up and get an i7 2600 instead.

[i_build_nanosuits]

17 minutes ago, david cassar said:

Interesting results. Would definitely say that i5s nowadays especially i5s at stock frequencies to get maxed out in a majority of modern titles. The one which comes to mind is easily battlefield 1 64 player multiplayer. You need an i7 to play bf1 at a constant 60 fps.

...and watchdogs 2...that's a CPU pig...even my overclocked core i7 see 90-95% load pretty much consitently at high framerates.

[VerticalDiscussions]

Rip me, using I5-4670. Luckily, still using it with a GTX 770.

[Majestic]

5 minutes ago, VerticalDiscussions said:

Rip me, using I5-4670. Luckily, still using it with a GTX 770.

*presses 'F' to pay respects*

[FellTheSky]

In all honesty i think this results are a bit dubious.

Ive build a lot of machines with low end i5's and i never saw anything remotely close to that stuttering mess.

Would you try 1 game with stock clocks and 1866 rams?

[RGProductions]

On 3/9/2017 at 11:42 AM, Majestic said:

Tech City

He;s said before that he's OCD about stutter, and he tunes settings to eliminate it.

[Tellos]

even if I5s are becoming limited perhaps we should ask why? is it that CPU performance per core isint high enough or is it something hyper threading/SMT can help or is it something else? Seems like we are moving even with intel toward more physical cores as games can handle more threads thrown at a CPU.

[Froody129]

I'm running a 1060 at 1440p, so whatever bottleneck there would be with my i5 it isn't bothering me  

[ybriK]

I can confirm with my i7-2700K + R9 290X that I too get high CPU usage in those games - 60%-80% on all 8 threads. If I were to slap a 980 Ti, Fury X, 1080 Ti or any other high-end GPU solution then I wouldn't be surprised to see stuttering/bottlenecking. I'm definitely going to be investing towards Ryzen 7 1800X or "1900X" instead of an i7-7700K or "8800K". It's funny how some people paint Ryzen 7 as a bad solution for gaming when they don't see the full picture.

[Flavio hc 16]

In the next week if possible I will try the i5 7600k+gtx 970+ram at 3000mhz and see what happens

[Teddy07]

3 minutes ago, Flavio hc 16 said:

In the next week if possible I will try the i5 7600k+gtx 970+ram at 3000mhz and see what happens

If you buy this as new PC then it is a bad choice

[Flavio hc 16]

2 minutes ago, Teddy07 said:

If you buy this as new PC then it is a bad choice

No it's  not a new pc, and i need it for some heat test (I'm designing  a small form factor case)

[Majestic]

3 hours ago, FellTheSky said:

In all honesty i think this results are a bit dubious.

Ive build a lot of machines with low end i5's and i never saw anything remotely close to that stuttering mess.

Would you try 1 game with stock clocks and 1866 rams?

If they are dubious, you have some evidence of your own to present? Or a flaw you'd like to point out in the test methodology? Not trying to be an ass, and I'm glad you're taking interest in the topic, but much like the previous person who just said they're weird/off/dubious, he didn't conduct testing of his own to this degree. 

 

3 hours ago, RGProductions said:

He;s said before that he's OCD about stutter, and he tunes settings to eliminate it.

That may be, but his results are still minimum/average framerates in the video's. So he has two faces apparently.

 

2 hours ago, Tellos said:

even if I5s are becoming limited perhaps we should ask why? is it that CPU performance per core isint high enough or is it something hyper threading/SMT can help or is it something else? Seems like we are moving even with intel toward more physical cores as games can handle more threads thrown at a CPU.

Well, you're seeing all cores being maxed out, i'd say it's lack of threads. Especially Watch Dogs 2 can use multiple threads.

49 minutes ago, ybriK said:

I can confirm with my i7-2700K + R9 290X that I too get high CPU usage in those games - 60%-80% on all 8 threads. If I were to slap a 980 Ti, Fury X, 1080 Ti or any other high-end GPU solution then I wouldn't be surprised to see stuttering/bottlenecking. I'm definitely going to be investing towards Ryzen 7 1800X or "1900X" instead of an i7-7700K or "8800K". It's funny how some people paint Ryzen 7 as a bad solution for gaming when they don't see the full picture.

Yeah, Ryzen caught my interest with the "for honor" minimum framerates. Aswell as JayzTwoCents's remarks on how smooth the framerates were in GTA 5 compared to Intel parts. Honestly, I think Ryzen has some nice potential to be competitive on a different level than we're used to. We just need to adjust our testing methodology to highlight this.