Core i7 7700k benchmark

[NumLock21]

15 minutes ago, MageTank said:

"I saw someone compare" is not evidence. Give me more than that. Not only that, how did you know the memory speed? Do not even try to tell me you "extrapolated it from the bandwidth results" because the results are extremely inaccurate. My bandwidth is 30% slower on this test than any other bandwidth test I've used. 

 

There are only a few Geekbench submissions to choose from. In fact, I count exactly 8. https://browser.primatelabs.com/v4/cpu/search?utf8=✓&q=7700k

 

From these 8, only the Gigabyte results show a drastic improvement over Skylake. Feel free to pick the 4% improved IPC result out of those 8. Trust me, it doesn't exist because IPC was not changed. 

I like how both AMD and Intel have a 7700K.

And I still remember seeing the Core i7 6820HK. Thought it was the successor to the 5820K and now it's in a laptop, a 6 core 12 thread laptop?!

No, it was just a quad core.

 

[MageTank]

1 minute ago, NumLock21 said:

I like how both AMD and Intel have a 7700K.

And I still remember seeing the Core i7 6820HK. Thought it was the successor to the 5820K and now it's in a laptop, a 6 core 12 thread laptop?!

No, it was just a quad core.

 

I am just waiting for the poor soul that buys a Z270 board and AMD A10 7700k because he saw on a random youtube video that the "7700k is the fastest gaming CPU". 

[patrickjp93]

36 minutes ago, MageTank said:

"I saw someone compare" is not evidence. Give me more than that. Not only that, how did you know the memory speed? Do not even try to tell me you "extrapolated it from the bandwidth results" because the results are extremely inaccurate. My bandwidth is 30% slower on this test than any other bandwidth test I've used. 

 

There are only a few Geekbench submissions to choose from. In fact, I count exactly 8. https://browser.primatelabs.com/v4/cpu/search?utf8=✓&q=7700k

 

From these 8, only the Gigabyte results show a drastic improvement over Skylake. Feel free to pick the 4% improved IPC result out of those 8. Trust me, it doesn't exist because IPC was not changed. 

If your bandwidth isn't maxed out, that means the benchmark is more compute-heavy and currently is bottlenecked by the CPU. Do you even understand how these metrics relate to one another?

 

The Gigabyte result disagrees with you, but I can wait for wide release and 3rd parties to prove you wrong, again.

[MageTank]

Just now, patrickjp93 said:

If your bandwidth isn't maxed out, that means the benchmark is more compute-heavy and currently is bottlenecked by the CPU. Do you even understand how these metrics relate to one another?

 

The Gigabyte result disagrees with you, but I can wait for wide release and 3rd parties to prove you wrong, again.

How does it disagree with me? I clocked my rig at 4.5ghz and matched it. https://browser.geekbench.com/v4/cpu/631193

 

Are you telling me my IPC magically boosted by 4%? I think the one who does not understand is you. There is zero evidence supporting the 4% boost in IPC claim. Once again, you've seen the evidence but cannot provide it. I've proven several times over that IPC has not changed, and have even used my personal rig to do so. Also, have you even ran Geekbench? It does the tests one by one, one after another. It doesn't do compute while testing the bandwidth, so the CPU itself was not bottlenecking it. If that was the case, the other programs I use would indicate the same bottleneck, would they not? 

[patrickjp93]

3 minutes ago, MageTank said:

How does it disagree with me? I clocked my rig at 4.5ghz and matched it. https://browser.geekbench.com/v4/cpu/631193

 

Are you telling me my IPC magically boosted by 4%? I think the one who does not understand is you. There is zero evidence supporting the 4% boost in IPC claim. Once again, you've seen the evidence but cannot provide it. I've proven several times over that IPC has not changed, and have even used my personal rig to do so. Also, have you even ran Geekbench? It does the tests one by one, one after another. It doesn't do compute while testing the bandwidth, so the CPU itself was not bottlenecking it. If that was the case, the other programs I use would indicate the same bottleneck, would they not? 

With different memory speed and timings. Also, you don't know if the 7700k is maintaining 4.5 on all cores or not. Just as the 4790K goes down to 4.2GHz at 4-cores boost, it is possible Kaby Lake is doing the same.

 

You haven't proven anything because you lack the total evidence required. Their bandwidth test has a compute portion. Modern compilers will just erase operations that do nothing to actually make progress. If you just try to stream memory and do nothing with it, the compiler will optimize it away.

[MageTank]

1 minute ago, patrickjp93 said:

With different memory speed and timings. Also, you don't know if the 7700k is maintaining 4.5 on all cores or not. Just as the 4790K goes down to 4.2GHz at 4-cores boost, it is possible Kaby Lake is doing the same.

 

You haven't proven anything because you lack the total evidence required. Their bandwidth test has a compute portion. Modern compilers will just erase operations that do nothing to actually make progress. If you just try to stream memory and do nothing with it, the compiler will optimize it away.

That is exactly it. Finally, we agree. You do not know the clock speed of any of the 7700k results. Therefore, your claim that Geekbench proved a 4% increase in IPC is simply false, because it cannot be proven. I've said this exact same thing quite a few times in this thread too, have I not?

[patrickjp93]

2 minutes ago, MageTank said:

That is exactly it. Finally, we agree. You do not know the clock speed of any of the 7700k results. Therefore, your claim that Geekbench proved a 4% increase in IPC is simply false, because it cannot be proven. I've said this exact same thing quite a few times in this thread too, have I not?

I do know 2 of them because they were done by enthusiasts I know personally. 4% uplift.

[Dabombinable]

58 minutes ago, patrickjp93 said:

That's up to the benchmarkers to handle correctly. There are two ddr4 2400 dual channel results I saw someone compare, and there's still a 3.8 (4)% IPC uplift.

 

It's bad for comparing across ISAs, not within them. No benchmark is perfect.

Considering the significant difference between each test that I've seen when running it on this computer. On a fresh install of Windows 10 with only the drivers installed. Its outright bad.

[MageTank]

Just now, patrickjp93 said:

I do know 2 of them because they were done by enthusiasts I know personally. 4% uplift.

This again? What is it with your evidence always being impossible to present? This is why nobody takes you seriously. First it was your secret intel shareholder phone conference proving Jim Keller was joining Intel, then it was the Skylake GT4e being faster than a GTX 950 (Which, it ended up being in between the GTX 750 and 750 Ti, exactly where I predicted). Also, you claimed you saw it in an updated Geekbench test. You understand the moment you run those, the results are published for everyone to search through, right? My result, one that I never gave consent to publish, is on there for everyone to search for and find. That means your 2 enthusiast friends that ran Geekbench, are one of the 8 results. Not a single one of those 8 results prove a 4% IPC increase.

 

Would you like to change your story again?

[patrickjp93]

2 minutes ago, Dabombinable said:

Considering the significant difference between each test that I've seen when running it on this computer. On a fresh install of Windows 10 with only the drivers installed. Its outright bad.

Funny, because I get less than 0.2% variance (within OS time slicing expectations) on my Haswell MBPr w/W10 and quite a bit of software installed. I think you have thermal throttling problems or bad drives/memory.

[patrickjp93]

5 minutes ago, MageTank said:

This again? What is it with your evidence always being impossible to present? This is why nobody takes you seriously. First it was your secret intel shareholder phone conference proving Jim Keller was joining Intel, then it was the Skylake GT4e being faster than a GTX 950 (Which, it ended up being in between the GTX 750 and 750 Ti, exactly where I predicted). Also, you claimed you saw it in an updated Geekbench test. You understand the moment you run those, the results are published for everyone to search through, right? My result, one that I never gave consent to publish, is on there for everyone to search for and find. That means your 2 enthusiast friends that ran Geekbench, are one of the 8 results. Not a single one of those 8 results prove a 4% IPC increase.

 

Would you like to change your story again?

It's not impossible to present, and it's already there for you to see. Hint: the two scores are less than 100 points apart, and they use the same 2400MHz CL 13 RAM, and you will find both of them published on overclock.net with AIDA64 graphs proving there's no throttling.

[MageTank]

3 minutes ago, patrickjp93 said:

It's not impossible to present, and it's already there for you to see. Hint: the two scores are less than 100 points apart, and they use the same 2400MHz CL 13 RAM, and you will find both of them published on over overclock.net with AIDA64 graphs proving there's no throttling.

So the two scores are the Gigabyte scores? Meaning this entire news article is from your 2 friend's test? I would ask how you know the ram speed and CL, but you will just tell me its because you know them personally and they told you. It's okay though, I can run my kit at 2400 CL13 too. Do I leave tertiary timings and RTL/IO-L on auto? Or am I allowed to tinker as much as physically possible? What will give you a fair fight on this one?

 

Either way, OCN has no source whatever on your claims with 7700k and Aida64 graphs. If you could provide a source (ha, this again) i'd appreciate it. 

[Dabombinable]

30 minutes ago, patrickjp93 said:

Funny, because I get less than 0.2% variance (within OS time slicing expectations) on my Haswell MBPr w/W10 and quite a bit of software installed. I think you have thermal throttling problems or bad drives/memory.

My 2TB HDD has had less than a month total powered on. My Boot SSD has around 6 months and is far from degrading. The RAM is less than 2 years old, and it has no problems at all (eg. a micro fracture that leads to random BSOD if it gets too cold). And my 4790K barely touches 70^oC under my Siedon 120V, and doesn't thermal throttle under the same cooler when run at 4.8GHz 1.3V There is nothing wrong with the components. Especially when all other benchmarks that I run are far more consistent.

[MageTank]

1 minute ago, Dabombinable said:

My 2TB HDD has had less than a month total powered on. My Boot SSD has around 6 months and is far from degrading. The RAM is less than 2 years old, and it has no problems at all (eg. a micro fracture that leads to random BSOD if it gets too cold). And my 4790K barely touches 70^oC under my Siedon 120V, and doesn't thermal throttle under the same cooler when run at 4.8GHz 1.3V There is nothing wrong with the components. Especially when all other benchmarks that I run are far more consistent.

The test is extremely inconsistent. I've ran it multiple times in a row (have timestamps on my runs to prove it) and the multi-core score changes anywhere from 100 to 400 points at random. This is with a fixed voltage and core clock. It is by no means an accurate test. Nobody that I know of, especially the enthusiasts I know, use it to benchmark performance. Certainly not people that just got their hands on unreleased CPU's. 

[patrickjp93]

51 minutes ago, MageTank said:

So the two scores are the Gigabyte scores? Meaning this entire news article is from your 2 friend's test? I would ask how you know the ram speed and CL, but you will just tell me its because you know them personally and they told you. It's okay though, I can run my kit at 2400 CL13 too. Do I leave tertiary timings and RTL/IO-L on auto? Or am I allowed to tinker as much as physically possible? What will give you a fair fight on this one?

 

Either way, OCN has no source whatever on your claims with 7700k and Aida64 graphs. If you could provide a source (ha, this again) i'd appreciate it. 

Stock xmp for G.Skill Ripjaws at 2400.

[MageTank]

Just now, patrickjp93 said:

Stock xmp for G.Skill Ripjaws at 2400.

Mine does not have a 2400 XMP. Either way, I'll leave everything on auto, dial in 2400 C13 (do you know the entire string of primary numbers?) and give the test a go with CPU clocked at 4500.

[patrickjp93]

24 minutes ago, MageTank said:

The test is extremely inconsistent. I've ran it multiple times in a row (have timestamps on my runs to prove it) and the multi-core score changes anywhere from 100 to 400 points at random. This is with a fixed voltage and core clock. It is by no means an accurate test. Nobody that I know of, especially the enthusiasts I know, use it to benchmark performance. Certainly not people that just got their hands on unreleased CPU's. 

You can't say it's inaccurate. You can say it's imprecise.

[MageTank]

1 minute ago, patrickjp93 said:

You can't say it's inaccurate. You can say it's imprecise.

Fine. Either way, I can't find a G Skill DDR4 kit that runs at 2400mhz C13. None for sell either. Are you certain its G Skill? The only 2400 C13 kits I can find in DDR4 are Mushkin (Hynix IC's). My kit has Samsung B Die, so simulating the tertiary timings that Hynix enjoys is going to be slightly difficult if left on auto. 

 

Lowest latency G Skill DDR4 2400mhz kits I could find, are C14-14-14-34. Either your friends also have access to unreleased ram, or you've mistaken the exact kit they are using. Give me a specific set of timings and clock speed, and i'll run it. 

[patrickjp93]

3 minutes ago, MageTank said:

Fine. Either way, I can't find a G Skill DDR4 kit that runs at 2400mhz C13. None for sell either. Are you certain its G Skill? The only 2400 C13 kits I can find in DDR4 are Mushkin (Hynix IC's). My kit has Samsung B Die, so simulating the tertiary timings that Hynix enjoys is going to be slightly difficult if left on auto. 

 

Lowest latency G Skill DDR4 2400mhz kits I could find, are C14-14-14-34. Either your friends also have access to unreleased ram, or you've mistaken the exact kit they are using. Give me a specific set of timings and clock speed, and i'll run it. 

Apologies, it's overclocked 2133 (to 2400) kept at 13-13-13-31.

[MageTank]

36 minutes ago, patrickjp93 said:

Apologies, it's overclocked 2133 (to 2400) kept at 13-13-13-31.

Kept at 13-13-13-31? G Skill does not make 2133 C13 memory. PANRAM, Kingston, Corsair, and A-Data do. Basically anyone that uses SK Hynix on their lower speed kits. Either way, I'll dial in those timings at 2400mhz, and leave everything else on auto. I'll get screenshots from the BIOS to confirm, and run the test with Aida64/CPUZ confirming my speeds. I'll edit this post with the results in about 10 minutes. 

 

EDIT: Now for the flood of information. First, my BIOS settings (including the poor RTL/IO-L auto training values):  http://imgur.com/a/E1n0z

 

Now for the geekbench results: https://browser.geekbench.com/v4/cpu/644243

 

Finally, the confirmation from multiple programs showing my settings are as promised: http://imgur.com/a/Lvov3

 

My score with this kit falls in line with 3 out of 8 of the 7700k results, and exceeds 3 out of those 8 scores. The only two scores that beat this handily are the Gigabyte results, the ones that you claim are from your friends. I do notice their bandwidth is faster by 1-2GB/s. Either their board is training their kit better than mine, or their speed is higher than claimed. Either way, this score matches/exceeds a majority of the 7700k results. That leads me to believe my previous statement regarding the non-existent IPC increase to be true. Like you said before, we do not have sufficient evidence to make any claims like this until the final product is released and is reviewed by third party reviewers. The good news is, one of us will be wrong come January, so we won't have that long of a wait. 

[MageTank]

@patrickjp93 Any additional tests before I do away with this awful memory profile? I don't want my ram thinking it deserves to run this slow at all times. 

[patrickjp93]

17 minutes ago, MageTank said:

@patrickjp93 Any additional tests before I do away with this awful memory profile? I don't want my ram thinking it deserves to run this slow at all times. 

Nope.