Mini-review: i3 4130

[Yuriman]

Snagged a fair deal on a Haswell i3 as an upgrade for my wife's aging Q6600 rig, and here's my results:

Normalized around the stock i5

Original data can be found here.

The basics:

Core i5 3570K, ASRock Z77 Extreme4, 2x 4GB DDR3 1600 9-9-9-24 XMP 1.25v, Gigabyte Radeon HD7850 underclocked to (860/1200), Catalyst 14.3 Beta, Windows 8.1 x64

Stock is 3.6/3.7/3.8/3.8
Overclocked is 4.6GHz (28%/24%/21%/21% higher)

Core i3 4130 (3.4GHz), MSI B85M-G41, 1x 4GB DDR3 1600 9-9-9-24 XMP 1.35v, MSI Radeon HD7850 @ stock (860/1200), Catalyst 14.3 Beta, Windows 8.1 x64

Q6600 (2.4GHz), Gigabyte EX38-DS4, 6GB DDR2 800 5-5-5-15 1.8v, MSI Radeon HD7850 @ stock (860/1200), Catalyst 14.3 Beta, Windows 8.1 x64

I'd like to highlight that the i3 is running in single channel because I got a screaming deal on the RAM. I figured it should be fine, since an i3 is roughly half an i7.

I only included the games she and I play regularly (had installed on both machines). All games were at max settings, 1920x1080 with no AA. Skyrim has around 50 mods installed and in places dips into the 20's, so I would expect it to be very GPU limited. Its data sample was taken in a particularly graphically intensive area. Guild Wars 2 FPS sample was taken in Rata Sum in a rather out-of-the-way spot where I was able to control the amount of activity around me to some extent, while still being much more CPU-heavy than a spot in the wilderness. Civ5 has the latest expenasion and data was taken from a save point 490 turns into an 8 player game.


In Handbrake, the Haswell i3 held around a 40% lead over the older Core2 Quad. Assuming perfect scaling, I'd have needed to clock the C2Q at 3.35GHz to match the Haswell dual. The i3 clocked in at 67% as fast as the previous generation i5, or the i5 held a 49% lead. Normalized for clock speed, the i5 is only 40% faster than the i3.

In the Cinebench CPU test, the i3 was around twice as fast as the C2Q and 63% of the i5. OpenGL rendering actually showed a small advantage to the i3, repeatable, though it was within 2% of the i5. The i3 completed the OpenGL test with 2.5x the framerate of the C2Q.

In Guild Wars 2, the i3 came out 75% faster than the C2Q with a 40% clock advantage, or from another perspective, you'd need a Kentsfield at 4.2GHz to match it, assuming perfect scaling. The i5, while clocked 6% higher, delivered 12% higher FPS at stock and did not scale with overclocking. Moving from C2Q gave an absolute improvement of 25fps to 43.5fps. There are definitely more CPU-bound areas, and I may revisit this test to see if the i5 might not pull further ahead in this situations.

In Civilization 5, the i3 delivered 50% higher framerates with 40% higher clocks than the C2Q. It's dead-even with a stock Ivy Bridge i5 clocked 6% higher, which itself scales linearly with clockspeed. I wish I had a method for comparing turn-processing time, framerate isn't even close to the entire story.

In Skyrim, the i3 system held a 3.5% (repeatable) advantage over the i5, both at stock and overclocked. I suppose platform improvements might be responsible for this? Regardless, an i3 delivers the same experience as an i5 in this (admittedly older) title.


____

Handbrake, Cinebench CPU and IBT all look to saturate all available threads, with IBT running a bit slower on the i3 presumably because hyperthreading can't extract anything out of pure-AVX workloads.

Cinebench OpenGL, SuperPi, seem to be single-threaded or two-threaded workloads (based on the graph, though I already know this). This could be deduced based on the overclocked i5 pulling ahead by ~21% rather than ~28%, which is its 2-core clockspeed advantage when overclocked. In lightly-threaded games where the CPU is not sufficient, I think you can expect the graph to resemble SuperPi's pretty closely.

Civ5 appears to be somewhere in between, and is probably pretty representative of how moderately threaded games will run.

Skyrim and Guild Wars 2 are examples of "good enough", with the stock i5 and i3 being just fast enough as to not cause a CPU bottleneck in the areas I sampled in those games. Note that not all areas are equally CPU bottlenecked though; I simply didn't capture it in my data.

____


I've disassembled the Q6600 for sale, but I can still do comparisons between a Haswell i3 and an Ivy Bridge i5, if anyone would like to make requests.

I have Battlefield 3/4, Minecraft, Sim City Online, Dolphin (emulator), The Sims 3, and SoaSE I can easily do comparisons with, though coming up with a method of benchmarking that reflects any real situations might require some thought. My brother has Crysis 3 which I might be able to borrow, but it won't be in the next few days. I'm open to downloading free MMOs for testing.

 

[Yuriman]

Related thoughts:

My wife asked me to put the heat on today, which hasn't happened in a while. I think I'm going to appreciate the reduced heat output once it gets warm. Heat output when idle was very high with the C2Q, though it wasn't a problem if we put the machine to sleep.

The i3 has about 2/3 the TDP of my i5, and delivers very roughly 2/3 the performance in perfectly threaded workloads. It seems that with a power budget, you get what you pay for. In workloads which are lightly to moderately threaded, it seems pretty close to an i5, having the same single-threaded throughput (stock) while being able to run up to 4 threads in total. 30% seems a fair estimate of what hyperthreading can extract.

In two of the three games tested, the i3 gave a very comparable experience to my i5, and brought framerates up from the 20's (which couldn't really be raised by dropping settings) into mid 40's which is around the point I shoot for in non-twitch games as my GPU bottleneck. If the i3 isn't enough later on, I'll sell it, probably for close to what I paid, and drop in an i5.

 

[VirtualLarry]

So, according to your graphs, there's literally nothing that the Q6600 beats the i3 at. I wonder how the Q6600 would stack up, OCed to 3.6?

I sold a friend a really nice Q6600 gaming rig a few years back with a GTX460, for what I thought was a fair price. (And my friend was pretty ecstatic with it, when he got it.)

Just wondering if I should show those graphs to him, and talk to him about upgrading. Or would it be more of a side-grade (except for power-consumption), since his Q6600 is at 3.6?

Edit: btw, thanks for this thread. Seems few "pro" review sites bother with the slower chips.

Edit: If you scale the Q6600 by 50% linearly, it beats the i3 in handbrake, with a score of 0.72. Then again, since the scores aren't raw, but normalized to the i5, I don't know if that's a valid way to scale the Q6600 scores.

 

[Yuriman]

Well, what I've read and seen from my own testing so far is that a Haswell i3 is a little faster than an equally clocked C2Q in well-threaded tasks. If we assume you can get 30% from hyperthreading, that tells me that he stands to gain up to 30% in things that are heavy on 1-2 threads, give or take.

If he had been starting from scratch, the i3 would save money by requiring no extra cooling and allowing for a smaller power supply/cheaper motherboard. Moving over now will, at some point in the distant future, begin saving him money on his electricity bill over the cost of the upgrade. Though the i3 draws less power, it also dumps out a lot less heat and won't hit the air conditioner as hard in the summer. Remember that when overclocking, you often have to disable some power saving features in addition to increasing the draw of the chip itself. At 3.6, it wouldn't surprise me if it were close to 200w under load, and who knows what at idle.

I moved my Hyper212+ over when I upgraded, but it seems pretty unnecessary. I may try taking the fan off of it tomorrow and open up a game, just to see if I can, but the stock cooler is pretty darn quiet anyway.

 

[VirtualLarry]

At 3.6, it wouldn't surprise me if it were close to 200w under load, and who knows what at idle.

It has a Tuniq Tower 120 on it, but it OCed to 3.6 on pretty low volts, so I doubt it's pushing 200W. (Don't think the Tuniq is rated for 200W either.) I could believe 150-160W at load, and 100-110W at idle though.

The Haswell i3 would take considerably less.

 

[BSim500]

Yeah, they're pretty impressive for what they can do with only two cores.

In Skyrim, the i3 system held a 3.5% (repeatable) advantage over the i5, both at stock and overclocked. I suppose platform improvements might be responsible for this? Regardless, an i3 delivers the same experience as an i5 in this (admittedly older) title.

A lot of that is that the wider Haswell core has really benefitted Hyper-Threading on the i3-4xx0. Some games only show 5-10%, but others have shown up to 20% and beyond or so at same clock. Quite a jump up. The i3 shows the Ivy -> Haswell architecture improvement effect on HT far much more than the i7 (since the 3rd & 4th real cores swallow up any additional threads long before the effect of HT kicks in).

Whereas Haswell i5's have typically shown 10% vs Ivy i5 of same clock, Haswell i3's have typically shown 15% or so vs Ivy i3 of same clock. It shows up better in many games than benchmarks:-

Sleeping Dogs:-
i5-4430 @ 3.1GHz = 123.7fps 11.5% boost with +100mhz clock (or nearer 8% at same clock)
i5-3330 @ 3.0GHz = 110.9fps

i3-4330 @ 3.5GHz = 109.1fps 16.8% boost (at same clock)
i3-3250 @ 3.5GHz = 93.4fps
http://www.xbitlabs.com/images/cpu/core-i3-4340-4330-4130/Charts-1/sleepingdogs.png

And a few games show massive improvement:-

Metro Last Light:-
i5-4670 = 76.9fps @ 3.4GHz
i3-4340 = 72.7fps @ 3.6GHz
FX-8350 = 62.5fps @ 4.0GHz
FX-6300 = 53.9GHz @ 3.5Ghz
i3-3240 = 51.2fps @ 3.4GHz
http://img.hardcoreware.net/wordpress/wp-content/uploads/2013/12/metro-last-light-fps2.png

72.7 vs 51.2fps = 42% difference (or 34% increase in fps adjusted for 200Mhz clock disparity). That's huge. Big enough that it's gone from being slower than an FX-6300 to faster than an FX-8350 in one jump. Whoever said Hyper-Threading was "crap"?... :biggrin:

Throw in the 54w @ 3.4-3.6GHz vs a Q6600's 105w @ 2.66GHz, and it's amazing how far stuff has come over the past 7 years. You still can't OC i3's but with the right motherboard you can still undervolt them, so that 54w may well fall even further to a real-world 38-45w or so (vs 150w or so for a heavily OC'd Q6600). Neat little chips if you can't afford an i5 / want lower TDP.

Be interesting to see what the 3.7GHz i3-4360 benchmarks like later on in the year. Pretty sad that only 3-4 tech sites have bothered to benchmark them (partly due to the later release date vs i5, and partly the "But it's only got 2 cores!" stigma).

 

[TeknoBug]

I've used a friend's i3 4130 and it's identical to my i5 3550's performance in just about everything and costs less and consumes less power. It'll do just fine for just about everything and can play many games since most games are still single-threaded, games like Battlefield 4 will struggle a little (what doesn't it struggle on?).

and ^ wow an i3 4340 getting more avg. frames than an FX 8350...

 

[frozentundra123456]

Nice info, but none of those games are well threaded, so it makes sense the i3 would compare well to the i5.

A very interesting test will be Watchdogs, I think. The recommended requirement (not sure if final) is a hyperthreaded quad.

If one can afford an overclockable i5, I still think that is the best choice, because even at stock with turbo it will be faster in single threaded and destroy the i3 in multithreaded productivity if one does that.

Where the i3 becomes much more competitive is vs the FX (FX 6300 and even in some games vs stock FX83xx) line and vs lower end i5 chips with lower clock and no overclocking available. It is too bad the i3 is not a bit cheaper though. Especially the higher end ones are getting close to i5 territory.

 

[BSim500]

Nice info, but none of those games are well threaded, so it makes sense the i3 would compare well to the i5.

A very interesting test will be Watchdogs, I think. The recommended requirement (not sure if final) is a hyperthreaded quad.

It remains to be seen, but I think a lot of "min requirements" are deliberately over-inflated as some "this game will be so bad-ass we say it needs an i7 so please pre-order it" fashion statement. Thief's "recommended" Intel CPU is an i7 and yet it runs flawlessly on i3's with barely 2-3fps difference vs an i7-4770K both with and without Mantle.

As for Watchdogs, either it will use 2 completely different sized maps (small for consoles + large for PC's), or in reality it won't need anywhere near an i7 to run if it actually wants to work on an XB1 / PS4 (6x 1.75GHz cores each with half the IPC whose combined CPU power barely registers a 3.0 in Cinbench 11...). Funny thing is the min requirement on Watchdogs "Intel Core2 Quad Q8400" & "Phenom II X4 940" are typically both around 15% slower than OP's new i3 even in near-perfectly threaded apps like Handbrake, so unless it's a really bad port, I can't see too many problems.

 

[crashtech]

Thanks for taking the time to make this information available. It confirms some of what I've been thinking about this generation of i3, namely that HT has made larger advancements this time than in previous generations. While there is a still a large performance gap between 2C/4T and 4C/4T, it's not as large as it was with Sandy and Ivy. Too bad we can't see what an i3 "K" could do, I bet it would be very interesting.

 

[Yuriman]

I wonder if that voltage is being reported correctly?

 

[crashtech]

Seems really low. With that voltage, if accurate, passive cooling would likely be not problematic.

 

[jaydee]

Wonder where my Phenom II X4 @ 3.2 would fit here...

 

[Phynaz]

Wow, the Q6600 just gets smacked silly.

 

[crashtech]

Time to shake off the Core2 nostalgia.

 

[SPBHM]

"Q6600 (2.4GHz), Gigabyte EX38-DS4"

this thing is begging for some overclock

 

[escrow4]

The take away is if you want a faster box chuck anything Core 2 and welcome to 2014 with Haswell (or Broadwell). Modern platform that is substantially faster that dumps less heat and sucks less power.

 

[Maximilian]

The take away is if you want a faster box chuck anything Core 2 and welcome to 2014 with Haswell (or Broadwell). Modern platform that is substantially faster that dumps less heat and sucks less power.

New stuff is faster than old stuff? Whodathunkit!

 

[escrow4]

New stuff is faster than old stuff? Whodathunkit!

Substantial Its not just pffft 10 or 20%, its way faster and you get all the modern benefits. That is I'd say makes it essential.

 

[AkumaX]

"Q6600 (2.4GHz), Gigabyte EX38-DS4"

this thing is begging for some overclock

that's true. the minimum i ran any C2Q (Q6600, Q8300, Q9550) was 3GHz. Usually anywhere between 3.2Ghz-3.6Ghz.

on a sidenote, i also picked up an i3-4330 (cheapest processor that had HD4600 graphics, needed for Hackintosh )

 

[PliotronX]

These things are like unsung heroes in the enthusiast realm. I've been playing with my i3 4330 in a NAS I built around it and have been impressed with its performance. Like your finding, much faster in damn near everything than my Q6600 @ 3.6GHz while only politely asking for about the wattage of a 40W incandescent bulb. Simply amazing to me, I don't know about anybody else. It's only a shame we cannot find out how high of a clock these bad boys are capable of. I'd imagine 5GHz would be attainable for 90% of i3's.

 

[crashtech]

These things are like unsung heroes in the enthusiast realm. I've been playing with my i3 4330 in a NAS I built around it and have been impressed with its performance. Like your finding, much faster in damn near everything than my Q6600 @ 3.6GHz while only politely asking for about the wattage of a 40W incandescent bulb. Simply amazing to me, I don't know about anybody else. It's only a shame we cannot find out how high of a clock these bad boys are capable of. I'd imagine 5GHz would be attainable for 90% of i3's.

I have used many i3s and have been recommending them as a way to enter into the superior LGA1150 platform on a budget, but even I would not call them unsung heroes. There is a performance gap between them and the i5s in threaded loads that can't be ignored, and yet their lightly threaded performance is nearly indistinguishable from an i5 or i7. To use them effectively requires the kind of discipline in reducing unneeded processes that we used to have when only a single core was available. I do think it's a shame that Intel can't make room in their pricing structure for an unlocked i3, though again I am skeptical of any assertion that they would be better OCers than the i5s or i7s. It'd be enough to see one at 4.4GHz rocking games on the (relatively) cheap.

 

[PliotronX]

I have used many i3s and have been recommending them as a way to enter into the superior LGA1150 platform on a budget, but even I would not call them unsung heroes. There is a performance gap between them and the i5s in threaded loads that can't be ignored, and yet their lightly threaded performance is nearly indistinguishable from an i5 or i7. To use them effectively requires the kind of discipline in reducing unneeded processes that we used to have when only a single core was available. I do think it's a shame that Intel can't make room in their pricing structure for an unlocked i3, though again I am skeptical of any assertion that they would be better OCers than the i5s or i7s. It'd be enough to see one at 4.4GHz rocking games on the (relatively) cheap.

I refer to them as such because they go nearly ignored yet they harness most of the power in the Core series for a lot of tasks and sip power. I imagine their overclockability overshadowing i5's and i7's just due to the sheer power envelope disparity.

 

[crashtech]

FWIW, I mostly agree... but then again I like to teach my users some discipline when it comes to running processes. A machine trimmed down to run only what you want it to seems to do pretty well with 2C/4T for the majority of things.

 

[NewYorksFinest]

Well, the core i3-4130 is only ~10-15% slower than the i5-4440 IMO.