My i7 7700k Has Arrived - Insights, Benchies & Overclocks Inside (Now With Delid!)

[thesmokingman]

If this is with 150 samples than that's really impressive 20% lower then zen at 3,4Ghz and with only half the cores.

Impressive over Zen? It's running 1.8ghz higher than zen with half the core count. I'm not sure that's impressive?

My 6700k@4.7 is only 4 secs slower. 4 secs at a cost of 500mhz, for Krabby Lake. I'm not sure man.

 

[TheELF]

Impressive over Zen? It's running 1.8ghz higher than zen with half the core count. I'm not sure that's impressive?

My 6700k@4.7 is only 4 secs slower. 4 secs at a cost of 500mhz, for Krabby Lake. I'm not sure man.

Even impressier.
4,7 will be pretty much the stock speed of the 7700k.
Having close to the same performance with half the cores means almost double the IPC.

 

[thesmokingman]

Even impressier.
4,7 will be pretty much the stock speed of the 7700k.
Having close to the same performance with half the cores means almost double the IPC.

Seriously? It obvious it doesn't work that way. If you scaled up you'd end up with a gasp a 6900k ish, and that does what vs that raisin sample?

 

[Chrisch]

what do u expect? Kaby Lake has no ipc improvement compared to skylake.

4.4sec are ~10% and thats what my kaby clocked higher than your skylake.

Edit: btw idk if you can compare zen with kaby/skylake (or intel) on blender because blender uses different paths on intel & amd

https://twitter.com/FPiednoel/status/766336117025017856

 

[thesmokingman]

what do u expect? Kaby Lake has no ipc improvend compared to skylake.

4.4sec are ~10% and thats what my kaby clocked higher than your skylake.

Edit: btw idk if you can compare zen with kaby/skylake (or intel) on blender because blender uses different paths on intel & amd

https://twitter.com/FPiednoel/status/766336117025017856

lol, as if. What are you supposed to do then?

Btw, I've no issue with your results that wasn't my point. Your oc/score is what it is. My issue was with the other poster inferring some godlike ipc vs raisins lol.

 

[vissarix]

what do u expect? Kaby Lake has no ipc improvement compared to skylake.

4.4sec are ~10% and thats what my kaby clocked higher than your skylake.

Edit: btw idk if you can compare zen with kaby/skylake (or intel) on blender because blender uses different paths on intel & amd

https://twitter.com/FPiednoel/status/766336117025017856

Can you please run cinebench 11.5? I want to see how much of an improvement kabylake have over my i7 2600k at 5.1ghz...for reference
2.06 single thread
9.98 multithreaded score running at 5.1ghz.

 

[Chrisch]

 

[itsmydamnation]

Even impressier.
4,7 will be pretty much the stock speed of the 7700k.
Having close to the same performance with half the cores means almost double the IPC.

No it doesn't
1. They are both running significantly faster memory
2. IPC instructions per CLOCK

Honestly how do you get this trivial stuff just so wrong?

 

[TheELF]

No it doesn't
1. They are both running significantly faster memory
2. IPC instructions per CLOCK

Honestly how do you get this trivial stuff just so wrong?

Wow,seriously?
So what's your point here,that memory is executing instructions?
If faster mem helps you extract every bit of possible IPC from a core then that's a good thing.

 

[coercitiv]

Wow,seriously?
So what's your point here,that memory is executing instructions?

His point is you're having trouble distinguishing between performance and IPC.

To put your claim into perspective, you just said Kaby Lake has almost double the IPC of Broadwell.

 

[TheELF]

His point is you're having trouble distinguishing between performance and IPC.

Yeah I know,it comes from so many pages of people bickering over MT benchmarks and drawing conclusions on IPC (while ,without knowing what amd means with core or smt,you normally could only draw conclusions about performance. )

But you are quite right, the kaby core will only be almost twice as fast and not have almost twice the ipc.

 

[itsmydamnation]

Wow,seriously?
So what's your point here,that memory is executing instructions?

No execution units execute instructions. How does data get to the execution units? Prefetches and predictors determine that data is going to be used and move it from memory to cache. Now what happens when the prefetches and predictors miss or get it wrong? It has to be moved from memory to registers after the core has stalled! So guess what memory with faster access times does? That's right, moves it quicker. Guess what a CPU can do went its stalled waiting for memory? thats right, nothing.

If faster mem helps you extract every bit of possible IPC from a core then that's a good thing.

Yes faster memory or a lower cpu clock speed with the same memory will help increase average IPC because the core will be stalled for less clock cycles.

But you really should try to compare apples to apples and Zen and broadwell-E used DDR4 2400 in the test runs.

But you are quite right, the kaby core will only be almost twice as fast and not have almost twice the ipc.

Care to put a sizable wager that kaby lake per core will be twice as fast? I take that to be based off fastest stock single core ?

 

[vissarix]

nice score, whats your max stable overclock? and what voltage is considered safe for kabylake?

 

[Chrisch]

idk whats the max voltages for kaby lake, i think its the same as for skylake. but in my opinion i wont go over 1.35v for 24/7 or Prime.

max what i tried is this

i think this one can do stable 5.2GHz sub 1.4v

 

[Charlie22911]

I'd be personally leery of pushing a chip that far without some below ambient cooling, especially having removed the IHS and with it any hope of an exchange.

 

[Madpacket]

Wrong thread, delete.

 

[thesmokingman]

idk whats the max voltages for kaby lake, i think its the same as for skylake. but in my opinion i wont go over 1.35v for 24/7 or Prime.

i think this one can do stable 5.2GHz sub 1.4v

I would guess that you are well under the safe limit.

From the ROG guides regarding Skylake:

There is no official word on a safe CPU core voltage. But it is widely accepted that you shouldn’t exceed 1.45V. If you’re the type of user who likes to play it safe, then keep it under 1.4V.

 

[FlanK3r]

idk whats the max voltages for kaby lake, i think its the same as for skylake. but in my opinion i wont go over 1.35v for 24/7 or Prime.

max what i tried is this

i think this one can do stable 5.2GHz sub 1.4v

u kiding me man Wow! Gold sample I thinking. I remember you from Bloomfiled times (or only the same nick ??)

 

[Chrisch]

Hi FlanK3r,

yes thats me. Good old bloomfield time

lets see if it is a golden sample, will get 3 other 7700k this week and u will see the results here

 

[TheELF]

lets see if it is a golden sample, will get 3 other 7700k this week and u will see the results here

You guys are killing me.....
Can someone at least try some qsv x265 transcoding and tell us what he thinks about quality,time needed and cpu usage?
I mean (sadly) that's the biggest feature of kaby.
Info about EU count and Mhz would also be nice.

 

[Chrisch]

if you can tell me a benchmark for x265 transcoding i will test it.

 

[mikk]

You guys are killing me.....
Can someone at least try some qsv x265 transcoding and tell us what he thinks about quality,time needed and cpu usage?
I mean (sadly) that's the biggest feature of kaby.
Info about EU count and Mhz would also be nice.

Unchanged 24 EUs and 1150 Mhz for i7-7700k. 8 Bit HEVC encoding quality seems unchanged as well, speed is roughly 10% better. 10 Bit encoding is not yet support of any tool.

 

[thesmokingman]

if you can tell me a benchmark for x265 transcoding i will test it.

Handbrake will do the trick.

Unchanged 24 EUs and 1150 Mhz for i7-7700k. 8 Bit HEVC encoding quality seems unchanged as well, speed is roughly 10% better. 10 Bit encoding is not yet support of any tool.

Is it still unreliable in Handbrake? Half the time Skylake encodes don't work, or well. Because of this I switched NVENC, so much better.

 

[mikk]

Is it still unreliable in Handbrake? Half the time Skylake encodes don't work, or well. Because of this I switched NVENC, so much better.

I prefer QSVEnc via Staxrip for Quicksync because I have more options and runs faster. Handbrake should work depending on the settings.

 

[thesmokingman]

I prefer QSVEnc via Staxrip for Quicksync because I have more options and runs faster. Handbrake should work depending on the settings.

I've been meaning to try stax but haven't had a chance. I read it does NVENC pretty well too. In Handbrake I found that qsync had issues with decoding certain combinations of encodes and it was hard to figure out what the culprit was, like banding... big ass bands across the finished encode. When it did work, qsync x265 did cut 3 hour cpu times down to 1-1.5 hours. NVENC under Mediacoder however cuts that down even more to like 45 minutes. HDTV shows take 12 minutes to re-encode lol. I'd have to have a good reason to go back to qsync at this point.