[Ocaholic] Interesting Findings - 2500k/3570k/4670k Comparison

[Ventanni]

Over on the Planetside 2 forums, I saw a poster link this article comparing the performance between the 2500k, 3570k, and 4670k.

http://www.ocaholic.ch/modules/smartsection/item.php?page=0&itemid=1158

In the review, they utilize some low resolution settings to really place the focus on the CPU performance alone. Now, I know some of you guys find this method to be unrealistic, but to me it gives me an idea of how a CPU will perform in a largely CPU bound environment. If you play MMO's like I do, it can be relevant to what I can expect if I were to purchase such a CPU.

What the review found is that the Ivy Bridge 3570k performs, on average, better than the 4670k does, which he states in his conclusion. In some benchmarks, the Haswell chip performs as expected. In some, they perform about the same, and in others, Ivy pulls ahead. Now, even as an Ivy Bridge owner, my internal "Whaaat?" meter is going off the charts. It's like seeing actual magic happening, and I find it a little hard to believe given the fact that Haswell internally is a bit beefier than its Sandy/Ivy Bridge counterparts. In a nutshell, Haswell has:

-Improved branch predictor
-Widened execution ports (8 vs 6)
-Larger internal cache sizes
-Added instruction sets (kinda irrelevant for gaming)

So to me, there's hardly any way Ivy can pull ahead of Haswell so easily.

What I think?

In Tom's review of the 4770k, he does point out some cache bandwidth inconsistencies. From the link below, scroll down to the cache bandwidth section, he does go on to say this:

http://www.tomshardware.com/reviews/core-i7-4770k-haswell-review,3521-12.html

"L1D, L2, and L3 cache bandwidth are all up on our Core i7-4770K compared to my preview piece. However, only the L1D yields the doubling of bandwidth we were expecting. Given 64 bytes/cycle (compared to 32 for Ivy Bridge), this number should still be much higher than it is. There’s still no good explanation for that outcome."

What do you guys think? Fudgy results or is Ocaholic onto something?

 

[sm625]

Haswell has less TDP to work with. Even if the TDP of haswell and ivy are the same, they're really not, because haswell has to include its integrated VRMS in its TDP rating, while ivy does not since those parts are on the motherboard. If you were to strictly limit the amount of power going into an ivy motherboard to match the amount of power a haswell motherboard uses, its performance would be less than haswell.

 

[ShintaiDK]

What you need to ask is, is this review odd compared to other reviews in results. Or is it inline with the rest.

Your answer is there.

 

[Ventanni]

I find the results odd. In gaming benchmarks, it's showing Ivy is faster. In every single application benchmark though, Haswell is, as expected, 5-10% faster. And, in every other gaming benchmark I've seen, Haswell reigns supreme too, so I find the results of the article to be at odds with what I've seen elsewhere.

 

[mikk]

Over on the Planetside 2 forums, I saw a poster link this article comparing the performance between the 2500k, 3570k, and 4670k.

http://www.ocaholic.ch/modules/smartsection/item.php?page=0&itemid=1158

In the review, they utilize some low resolution settings to really place the focus on the CPU performance alone. Now, I know some of you guys find this method to be unrealistic, but to me it gives me an idea of how a CPU will perform in a largely CPU bound environment.

CPU tests from ocaholic are nonsense. They have to learn how CPUs should be tested properly. When they lower detail settings in the game they also lower CPU bound settings. First mistake. When they lower a resolution to a 5:4 or 4:3 format they also lower the CPU dependency since widescreen resolutions are more demanding for the CPU. 720p, 768p, 576p or any low 16:9/16:10 resolution is mandatory. This is their second mistake. In the end this test is far away from realistic, I would call it incompetent.

 

[GRAFiZ]

CPU tests from ocaholic are nonsense. They have to learn how CPUs should be tested properly. When they lower detail settings in the game they also lower CPU bound settings. First mistake. When they lower a resolution to a 5:4 or 4:3 format they also lower the CPU dependency since widescreen resolutions are more demanding for the CPU. 720p, 768p, 576p or any low 16:9/16:10 resolution is mandatory. This is their second mistake. In the end this test is far away from realistic, I would call it incompetent.

In the end though, it seems few people scale down their gaming experience to a level where modern CPU's demonstrate much impact... as long as the bandwidth support is there, the video cards provide the most influence.

While OCA might not structure their benchmarks or testing in a "scientific" manner, they do craft them more in line with an end users direct experience... thus, making their conclusions much more useful to their target demographic.

 

[amenx]

CPU tests from ocaholic are nonsense. They have to learn how CPUs should be tested properly. When they lower detail settings in the game they also lower CPU bound settings. First mistake. When they lower a resolution to a 5:4 or 4:3 format they also lower the CPU dependency since widescreen resolutions are more demanding for the CPU. 720p, 768p, 576p or any low 16:9/16:10 resolution is mandatory. This is their second mistake. In the end this test is far away from realistic, I would call it incompetent.

Incorrect. When resolution and game settings are lowered, the CPU has more of a workload. The higher the res and game settings, workload is passed off more to the GPU while the CPU less to do. Thats why when you see high res benches, the CPU results show less variance because the GPU is doing most of the work.

 

[ShintaiDK]

http://www.hardwarecanucks.com/foru...ntel-haswell-i7-4770k-i5-4670k-review-16.html

http://gamegpu.ru/action-/-fps-/-tps/battlefield-4-test-gpu.html

I pitty those that think IB would be faster than HW.

 

[ShintaiDK]

Haswell has less TDP to work with. Even if the TDP of haswell and ivy are the same, they're really not, because haswell has to include its integrated VRMS in its TDP rating, while ivy does not since those parts are on the motherboard. If you were to strictly limit the amount of power going into an ivy motherboard to match the amount of power a haswell motherboard uses, its performance would be less than haswell.

TDP is not the same. 77W vs 84W. Also Haswell uses lower voltage. Only in AVX2 loads does it get close to equal.

 

[TuxDave]

"L1D, L2, and L3 cache bandwidth are all up on our Core i7-4770K compared to my preview piece. However, only the L1D yields the doubling of bandwidth we were expecting. Given 64 bytes/cycle (compared to 32 for Ivy Bridge), this number should still be much higher than it is. There’s still no good explanation for that outcome."

From the test (L1$ bandwidth)
4770K: 3.5GHz (3.9GHz turbo), 4 cores, 982.15GBps
3770K: 3.5GHz (3.9GHz turbo), 4 cores, 516.35GBps

Back of the hand calculation makes us expect the part to run (best case) at:
3.9 GHz * 4 cores * 2 load ports * 16B per load = 499.2 GBps (IVB)

Kind of weird.

 

[Termie]

One possibility - while they have stated that Win7 was used in all tests, the use of Win8 on the Sandy/Ivy system could account for the higher speed. In at least one test (Crysis 3 at 1080p), the 2500K beats the 4670K. That's impossible under any fully-controlled circumstance. Crysis 3 is extremely CPU-dependent.

 

[mikk]

Incorrect. When resolution and game settings are lowered, the CPU has more of a workload. The higher the res and game settings, workload is passed off more to the GPU while the CPU less to do. Thats why when you see high res benches, the CPU results show less variance because the GPU is doing most of the work.

It is not incorrect, it seems you did not understand. Widescreen resolutions are more demanding and CPU heavier than 5:4 or 4:3. Every serious reviewer should not this but unfortunately most are not aware. Because almost everyone uses widescreen these days means 5:4 or 4:3 is wrong. You can lower the resolution to make it more CPU bound of course but you have to make sure a widescreen resolution is used.

 

[blastingcap]

I've been lectured by Toyota enough times on this that I have to agree with him, just lower the resolution more and more and keep it at the same aspect ratio.

In any case, does this even matter? We all know that Haswell was geared for energy efficiency and better iGPU. The difference between IB/Haswell is so small as to be insignificant by the time that most games are CPU bottlenecked by either one of them.

 

[Grooveriding]

I'm waiting for the other shoe to drop; Intel finally releases a new CPU that literally brings absolutely no improvements in CPU performance over their prior chip, just improved power consumption and better graphics performance.

It's coming, it's practically already there right now looking at the meager improvements in CPU we've seen the past two releases, and increasingly worsening temperature characteristics of overclocked Intel chips compared to last gen Intel parts.

 

[frozentundra123456]

One possibility - while they have stated that Win7 was used in all tests, the use of Win8 on the Sandy/Ivy system could account for the higher speed. In at least one test (Crysis 3 at 1080p), the 2500K beats the 4670K. That's impossible under any fully-controlled circumstance. Crysis 3 is extremely CPU-dependent.

Not really convinced of the validity of their tests either. Looks like most of the 1080p benchmarks are gpu limited, and in a lot of cases, overclocking to 4.5ghz gives the same result as stock, which makes it look like something other than the cpu is the limiting factor.

If you accept the results of this test, it looks like bottom line you can expect a 10% or so improvement from SB by going to Haswell/Ivy, but is not clear if Haswell is faster than Ivy. But personally I dont trust this data.

 

[videogames101]

this data is bullshit, simple as that

 

[glugglug]

Heat sink is not mentioned in their test setup.

While they are probably the easiest to install and remove to go from testing one CPU to the next, Intel stock heatsinks are garbage, and the defective by-design plastic pins will stretch a bit more and get even worse-performing with each successive installation and removal.

If they are using the garbage stock Intel heatsink, that would explain it. Haswell has more heat to dissipate at the CPU itself due to FIVR. All 3 platforms will thermal throttle once they hit 100C or possibly 95C, not sure on that. Which will happen pretty easily with Intel's junk heatsinks. Only marginally better than running with no heatsink at all.

 

[SammichPG]

In the end though, it seems few people scale down their gaming experience to a level where modern CPU's demonstrate much impact... as long as the bandwidth support is there, the video cards provide the most influence.

While OCA might not structure their benchmarks or testing in a "scientific" manner, they do craft them more in line with an end users direct experience... thus, making their conclusions much more useful to their target demographic.

OP may be a wow player, that game would love an 8ghz dual core haswell.

It starts as a 2004 game with refreshed textures and a bit of directx11 effects but the end game is one of the most cpu bound gaming scarios still relevant to the gaming industry.

You need heavy LUA addons to play end game and the fight themselves are very script heavy with thousands of events goin on, you could to be CPU bound even with an overclocked sandy bridge i7.

It's a bad game engine overall, there are 3 main threads, but one is much heavier than the others.

 

[Ventanni]

For the record, I don't really trust the data, but I brought it here to AT as a topic of discussion. In every application benchmark, Haswell is faster than Ivy. And, in every game test I've seen, Haswell is faster, except these of course.

As for being a WoW player, I used to be a long time ago, but not anymore. I played for about 4 months after I bought my 3770k. I used my Core 2 listed below prior to that and never really had any fps issues, and the 3770k just never flinched period. These days I play Planetside 2 which is just starting to see optimizations. With the exception of Amp stations, framerates are extremely fluid, albeit I'm heavily GPU bound all the time (PS2 will tell you in game if you're CPU or GPU limited). I also play Path of Exile, and that game will run on anything.

I have no emotional ties to my CPU, lol, and I'm with ShintaiDK when I also say that I pity anyone who thinks Ivy is faster. Like I said though, I brought it to AT as a topic of discussion.

Question - Does CoD:Black Ops, Skyrim, and Far Cry 3 have in-game benchmarks, or are they games you have to test yourself using FRAPS?

 

[ElFenix]

When they lower a resolution to a 5:4 or 4:3 format they also lower the CPU dependency since widescreen resolutions are more demanding for the CPU. 720p, 768p, 576p or any low 16:9/16:10 resolution is mandatory. This is their second mistake.

wat


edit: probably referring to field of view and how lazy developers lock the vertical angle of view regardless of aspect ratio, so wider aspect ratios have much greater field of view. frankly i think it's stupid that 1920x1080 shows more on screen than 1920x1200.

 

[mikk]

More objects yes. It is often the case that 1280x720 noAA runs slower than 1280x1024 noAA. Less pixel to render but slower because more demanding for the CPU.