Pentium G3258 - How can it be slower than an i3 @ -1GHz

[Chicken76]

Can someone please explain how can it be that in benchmarks (admittedly not all) the Pentium chip running 1 GHz faster than an i3 4330 be slower? And in the cases where it's faster, it's not by much.

Is that extra megabyte of L3 cache that important?
Can hyperthreading really make up for a 1GHz difference in clock in multi-threaded tasks?

Or is the Pentium chip crippled in other ways than those two? (OK, there's the lack of AVX, but how many of those applications used in benchmarks really use AVX?)

 

[ShintaiDK]

Could you link said benchmarks?

 

[SPBHM]

the extra MB of l3 is irrelevant, the big difference will only come from HT or something like AVX (not the case for current software as far as I know)

you can compare 3.5GHz i3 with 3MB vs 3.6GHz with 4MB here (4150 vs 4340) and the difference is not to amazing, even with 100MHz advantage.
http://www.xbitlabs.com/articles/cpu/display/intel-haswell-refresh_5.html#sect0

 

[Chicken76]

Could you link said benchmarks?

http://www.tomshardware.com/reviews/pentium-g3258-overclocking-performance,3849.html

Most of the time the i3 is faster than G3258 @ 4.5 GHz (or so they say. could their tests be flawed in some way?)

 

[ShintaiDK]

http://www.tomshardware.com/reviews/pentium-g3258-overclocking-performance,3849.html

Most of the time the i3 is faster than G3258 @ 4.5 GHz (or so they say. could their tests be flawed in some way?)

Depends it seems what apps and games are used. Example:
http://www.bit-tech.net/hardware/2014/06/24/intel-pentium-g3258-review/5

Hyperthreading would have to give something like 50% when you look at Metro and Grid 2. Perhaps these games simply spawn more threads that creates more context switching on the G3258.

 

[DrMrLordX]

Can someone please explain how can it be that in benchmarks (admittedly not all) the Pentium chip running 1 GHz faster than an i3 4330 be slower? And in the cases where it's faster, it's not by much.

Is that extra megabyte of L3 cache that important?
Can hyperthreading really make up for a 1GHz difference in clock in multi-threaded tasks?

Or is the Pentium chip crippled in other ways than those two? (OK, there's the lack of AVX, but how many of those applications used in benchmarks really use AVX?)

Hyperthreading. Intel's current implementation of HT is pretty strong on dual-cores, which is why they unlocked a Pentium, not an i3.

 

[sm625]

Any type of encryption benchmark will cause the i3 to put out big numbers vs the pentium. But if you're not using it then those numbers are meaningless. In real world usage I can tell you absolutely and positively that the pentium at 4.5GHz destroys every single configuration I've ever used, including an ivy bridge i7 at 4.1GHz. The 108 mS sunspider score is insanely fast. There are only very few moments where you can actually feel the lack of extra cores actually hurting it. Interstingly enough, one of those cases involves windows Snipping tool. There is an odd delay of about 0.5 seconds on the pentium. But I use a 3rd party screenshot tool anyway. With the advent of gpu based game recording, even that use case no longer requires extra cores.

 

[cbn]

Perhaps these games simply spawn more threads that creates more context switching on the G3258.

And I wonder if using a large video card with high resolution and detail settings makes a problem like that even worse?

Maybe testing at lower resolution and detail settings could narrow or eliminate the gap in certain scenarios where i3 has the advantage.

 

[ShintaiDK]

And I wonder if using a large video card with high resolution and detail settings makes a problem like that even worse?

Maybe testing at lower resolution and detail settings could narrow or eliminate the gap in certain scenarios where i3 has the advantage.

Size of the card, not really. Manufactor? Very so. nVidia for example perform better with a FX8350 than the same CPU does with AMD. Perhaps the G3258 would look alittle better when using an AMD card with less multithreaded drivers in DX.

 

[cbn]

Size of the card, not really. Manufactor? Very so. nVidia for example perform better with a FX8350 than the same CPU does with AMD. Perhaps the G3258 would look alittle better when using an AMD card with less multithreaded drivers in DX.

Thanks for reminding me about the driver differences between Nvidia and AMD.

However, I just wonder if using a smaller video card at lower resolution/detail settings vs. larger video card at higher resolution/detail settings (to keep the GPU balance relatively the same between the large and small video card) wouldn't show an improvement for other reasons besides graphics driver multi-threadedness?

All things being equal, isn't having to feed a large GPU (working at high resolution/detail settings) going to stress the limited resources of Pentium G3258 more than having to feed a small GPU (working at low resolution/low detail settings)?

If so, I wonder if we could see a narrowing (or elimination) of the gap between Pentium G3258 and Core i3 (by using a smaller video at lower resolution/detail settings) in situations where the game has too many threads for the G3258? (ie, Reducing one overhead problem to help at least partially fix another overhead problem). This without making the GPU the bottleneck.

 

[Flapdrol1337]

A quad is almost never at 100% in a game, even when you're cpu limited. A dualcore always is, so likely it would benefit more from hyperthreading than a quad.

The driver multithreading in the nvidia control panel is default set to "auto", so it would probably turn off automatically on a dualcore. A thing that might help is setting the "pre rendered frames" to 1 (the minimum). This might mitigate the poor scores in frametime consistency.

Using a lower object detail would probably work to the pentiums advantage, because the directx overhead would be relatively lower compared to the game simulation / scripts etc, which usually don't scale well to multiple cores.

Tests with different memory speeds and timings would be nice too, see if there's benefit to that when multitasking.

 

[crashtech]

The simple answer is that HT makes more execution resources available, such that when the apps + OS exceed a certain number of highly utilized threads, the i3 handles the load better. Same goes for comparing i3 to i5, 4 real cores is better than 2 plus HT. Same again with i5 vs i7, 4 cores plus HT is better than just 4, mostly.

In a way the popularity of the G3258 is going a long way to show just how far HT has come, since it is often compared to the i3, and how much HT does for the Haswell i3, which is really quite powerful for a dual core.

 

[Fred B]

The memory controller of the G3258 does 21,3 GB/s and the core i3 4330 25,6 GB/s and the 2130 Sandy 21 GB/s .

 

[SPBHM]

The memory controller of the G3258 does 21,3 GB/s and the core i3 4330 25,6 GB/s and the 2130 Sandy 21 GB/s .

it's 1333 vs 1600

it makes a difference, but not as much as HT
http://www.xbitlabs.com/articles/memory/display/haswell-ddr3_7.html

 

[DrMrLordX]

Note that the G3258 can run memory speeds other than DDR3-1333 on Z87 or Z97 boards. Granted, that sort of breaks the value proposition when you have to fork over $90+ for the board, but there are legitimate reasons to do that sometimes.