Will it be worth it to buy a Skylake i3?

[DrMrLordX]

People were overclocking their G3258s on H81 boards months ago. Intel may not have supported overclocking on H81, but the OEMs made it happen anyway.

The complaint is that Intel is using microcode updates to stop it from happening regardless of the motherboard OEMs.

 

[TheELF]

The complaint is that Intel is using microcode updates to stop it from happening regardless of the motherboard OEMs.

The thing is that oc never was in the specs for h81 OEMs "hacked" it so to speak,so intel does whatever update it has to do according to the specs and not
according to whatever hack an OEM did.

So it might just be a coincidence that they updated something that brakes oc,otherwise why would intel not have done this as soon as it came out.

 

[tipoo]

Not sure if you've all seen this but DF seems like it's a firm yes.

http://www.eurogamer.net/articles/digitalfoundry-2015-intel-core-i3-6100-review

 

[frozentundra123456]

Depends on the use case. We had a heated "discussion" on this in another thread, that discussed this Eurogamer review in detail.

For light use, the i3 is fine. Unfortunately the skylake i3 is hampered by the need for fast memory and the inability to use said memory except with Z170 motherboards. So my conclusion would be twofold.

1. It is fine for general use. However, in such cases, especially OEM, it will not be paired with a motherboard and memory that will show much performance gain over Haswell.

2. For gaming, you need fast memory and an overclocking motherboard. Considering that, it makes much more sense to spend the extra hundred bucks (or slightly more) for the 6600k. Four real cores and the ability to use the overclocking features you are forced to pay for in order to use appropriate ram for best performance.

The i3 has always been akwardly priced, IMO, and this memory issue just makes it even more so.

 

[dark zero]

The i3 has great price/ST performance.... however it didn't came with the correct techs. So Intel screwed it very hard.

 

[Hi-Fi Man]

Depends on the use case. We had a heated "discussion" on this in another thread, that discussed this Eurogamer review in detail.

For light use, the i3 is fine. Unfortunately the skylake i3 is hampered by the need for fast memory and the inability to use said memory except with Z170 motherboards. So my conclusion would be twofold.

I hate to let this spill in from the other thread but I already addressed the needing Z170 board issue in that thread. You can run the RAM at lower timings in a non-z motherboard to achieve the same actual latency (measured in ns) and thus performance.

 

[Justinbaileyman]

when are they coming out?

 

[Magic Carpet]

You can run the RAM at lower timings in a non-z motherboard to achieve the same actual latency (measured in ns) and thus performance.

You can, but for Skylake it is more beneficial to have higher bandwidth than lower latency. This is not a Thuban, heh.

 

[DrMrLordX]

You can, but for Skylake it is more beneficial to have higher bandwidth than lower latency. This is not a Thuban, heh.

Nobody has proven that one way or the other. Until someone actually runs benchmarks on the i3-6100 (or another Skylake i3) to test that theory, we won't know for sure.

 

[Erenhardt]

You can, but for Skylake it is more beneficial to have higher bandwidth than lower latency. This is not a Thuban, heh.

Nobody has proven that one way or the other. Until someone actually runs benchmarks on the i3-6100 (or another Skylake i3) to test that theory, we won't know for sure.

http://www.eurogamer.net/articles/digitalfoundry-2015-intel-core-i3-6100-review

Huge improvement from DDR3-2133 to DDR4-2666

http://pclab.pl/art65002-4.html

less than a 0,5% vs 11% in DF review

almost 0,6% vs 7% in DF review

almost 4% vs 10% in DF review

I do not believe i3 sees more benefit from faster RAM than i7\5

 

[ShintaiDK]

I do not believe i3 sees more benefit from faster RAM than i7\5

Your pclab benches actually backs up the case. Also it shouldn't be a surprise. The i3 in question got 3MB cache, i5 and i7 got 6 and 8MB cache. The 3MB i3 will have a lot more cache misses.

 

[Erenhardt]

Your pclab benches actually backs up the case. Also it shouldn't be a surprise. The i3 in question got 3MB cache, i5 and i7 got 6 and 8MB cache. The 3MB i3 will have a lot more cache misses.

Got any proper test showing that?

Those results actually back up that Santa is real. There is nothing in those numbers that point otherwise, so it must be true.

 

[ShintaiDK]

Got any proper test showing that?

Those results actually back up that Santa is real. There is nothing in those numbers that point otherwise, so it must be true.

The impact of larger cache have been shown over and over in the history. You may have forgotten the time where AMD sold you double the cache on K8 with a premium and it performed better as well.

Isnt Santa just a Red Team membership?

Note the low clock, Higher clocks would give higher difference. And remember all these are done on lower latency DDR3. And the uncore in Skylake runs faster.

 

[Raftina]

Where did this nonsense about Skylake i3 needing faster RAM come from? The vast majority of the benchmarks show about a 10% difference between the fastest configuration and DDR4 2133 (the fastest supported on a H110 mono).

 

[Erenhardt]

I know that cache size give some performance benefit.

We are discussing system memory frequency impact on gaming performance. Your graphs do not show any relation between memory speeds and L3 cache. There is no reason to believe i3 gets and order of magnitude more performance from going DDR3-2133 to DDR4-2666 than i5/7 counterparts.

 

[frozentundra123456]

Where did this nonsense about Skylake i3 needing faster RAM come from? The vast majority of the benchmarks show about a 10% difference between the fastest configuration and DDR4 2133 (the fastest supported on a H110 mono).

Exactly. And 10% is nearly all the gain seen from Haswell to Skylake. So that is exactly the point. Using slow ram pretty much negates any gain from Haswell, at least in the desktop segment.

 

[Magic Carpet]

Nobody has proven that one way or the other. Until someone actually runs benchmarks on the i3-6100 (or another Skylake i3) to test that theory, we won't know for sure.

Where did this nonsense about Skylake i3 needing faster RAM come from? The vast majority of the benchmarks show about a 10% difference between the fastest configuration and DDR4 2133 (the fastest supported on a H110 mono).

We are discussing system memory frequency impact on gaming performance. Your graphs do not show any relation between memory speeds and L3 cache. There is no reason to believe i3 gets and order of magnitude more performance from going DDR3-2133 to DDR4-2666 than i5/7 counterparts.

For example.

 

[Hi-Fi Man]

If you take a look at those pclab results it proves my point that Skylake responds to lower absolute latency more so than bandwidth. This isn't some Thuban or Deneb thing this is just how all if not most CPUs have responded to memory. Since CPUs aren't highly parallel (relative to GPUs) and usually execute serial instructions they don't need a lot of bandwidth and are more sensitive to RAM latency (actual latency measured in nanoseconds not timings). GPUs have always been the opposite.

Keep in mind the pclab test only tested the i5 and i7, these CPUs require more bandwidth because they have more cores.

 

[AtenRa]

Increasing memory bandwidth also decreases memory latency, because more data can be accessed at the same time.

So, going higher bandwidth(higher speed ram) vs faster ram (lower latency) should be very close.

 

[Magic Carpet]

If you take a look at those pclab results it proves my point that Skylake responds to lower absolute latency more so than bandwidth.

One their slides is showing that too, there are games like GTA 5 that can clearly extract more frames per second from higher-bandwidth/lower latency DDR4/3 RAM, DDR4 2666 Mhz seems to be the way to go if gaming is a priority in my opinion, i5/i7 or i3. Doesn't matter. Skylake can always use some extra bandwidth, expect the gap to get bigger in the future.

Keep in mind the pclab test only tested the i5 and i7, these CPUs require more bandwidth because they have more cores.

The YouTube video I linked earlier, clearly shows the higher-bandwidth DDR4 memory showing tangible gains in GTA 5 over its slower DDR4 part, yes, even with a "lousy" i3.

This isn't some Thuban or Deneb thing this is just how all if not most CPUs have responded to memory.

No, Thuban was special. Normally CPUs benefit from both latency and bandwidth, but my Thuban (AMD CPUs of that era in general) showed no real benefit above ~1600Mhz, but very much "liked" and preferred ultra low latencies (gains of up to 20% in some apps, e.g. Handbrake). Intel CPUs didn't have the same pattern at that time, at least in my rigs (latencies didn't affect them as much as pure Mhz). Different memory controller, whatever, I don't know. But I very clearly remember what I saw. Same today, my Haswell doesn't care much if I set my RAM to 1600 CL8 or CL7, but I know with 1866Mhz CL9 I get better results.

Increasing memory bandwidth also decreases memory latency, because more data can be accessed at the same time.

So, going higher bandwidth(higher speed ram) vs faster ram (lower latency) should be very close.

Usually, it's like this.

Anyway, it's pretty clear DDR3 is reserved for more budget oriented builds (where cost is important) and the difference of 1-10 frames per second is not a big deal. Those DDR3 memory kits that can actually compete with higher-clocked DDR4 will be "out of budget" for most people. And people with money won't bother with them, why? they will most likely be overclocked and overvolted. Why bother when you can get some nice 1.2-1.35V DDR4 today (last I checked, prices were reasonable). Safer for your CPU too. And if you are building a gaming rig, you might as well get a decent Zee board with a decent memory kit and maybe upgrade to an i7 later down the road, that i3 does surely look like a good start. Unless of course, you are a secret Zen follower, haha.

Again, as always, in just my humble opinion

 

[ShintaiDK]

Increasing memory bandwidth also decreases memory latency, because more data can be accessed at the same time.

So, going higher bandwidth(higher speed ram) vs faster ram (lower latency) should be very close.

Not really no. Just look up old memory tests and you see. It depends on the individual application.

 

[Magic Carpet]

Not really no. Just look up old memory tests and you see. It depends on the individual application.

The real question is, how fast we "run out" of regular dual-channel memory bandwidth for the common apps. There are apps like WinRAR that like bandwidth and extra cache.

Source.

Extra cache for the win

 

[ShintaiDK]

Remember you get both lower latency and higher bandwidth with EDRAM. But ye, I wish we had more of that.

 

[DrMrLordX]

Look, all I know is that someone needs to get a "cheap" H150 board that has enough UEFI options to change memory timings, and then take an i3-6100 and bench DDR3-2133 CL13 vs CL14 vs CL 15 in some applications/games. We can extrapolate and speculate all we want, but as others have noted, different memory controllers respond to memory bandwidth and timings . . . differently. We are not dealing with Stars or Haswell here; we're discussing Skylake.

The eurogamer tests leave us with more questions than answers, since it benches a Skylake with DDR4 vs. a Haswell with DDR3, which proves next to nothing about whether memory clockspeed or memory timings benefit Skylake more. Adding to the confusion is the use of an i3 on a motherboard chipset that is (mostly) meant for unlocked i5s/i7s . . . in other words, it's a CPU/mobo combo that is not likely to show up "in the wild".