Intel Kaby Lake Core i7-7700K Sample Tested

That heat is rather offputting, provided that those temperatures are accurate and not just a faulty chip. :/ It looks like Zen is going to be an attractive alternative for a fair number of folks.... at least at this point.

 

I was a early 4770K adopter too, bought it by rls and payed 295€.. 3-4months later it was already 320-330€.. lol now after 3.5yrs still ~ 330€ :grin:

 

Well I'm pretty sure most of us realized by now that this chip really only brings a new MB platform with mostly enhancements for SSD's, and a few other motherboard related features.

I wasn't expecting anything dramatic from the specs. For those that do not OC, sure maybe, but I'm good for a while it looks like.

 

Absolutely! If you want to stick with Intel the first worthy upgrade looks likely in late 2018/early 2019. Coffee Lake may be okay, even though it is a second interim processor they have enough time to tweak it. I wouldn't expect huge advantage from it though. The 6-core version will probably cost a lot more and I suspect clocked lower (or have a lower potential overclock) than the 4-core variant.

 

EVEN LOWER PERFORMANCE PER CLOCK IN SOME TEST lol

 

The 128mb L4 do a world of difference for the i7 5775c. 1GB HBM might either be a quantum leap or just hit the ugly diminished returns roof. Either way, should be interesting.

 

128MB for a 4c/8t? I don't think so. People seem to forget why caches exist in the first place and why they're so small. The bigger it is, the slower it becomes. If we keep adding more levels of cache and keep making them bigger, it eventually defeats the purpose of having them, or RAM for that matter.

Also not sure if this changes anything but the 5775c has 128MB of eDRAM, which to my understanding is functionally similar to an L4 cache but specific to the IGP. The IGP is why something like an L4 cache would even be considered necessary in the first place.


Obviously, there are some situations where having a giant L4 (or even L3) cache would make a huge performance difference. But I figure that would really only apply in synthetic benchmarks.

 

No, it's just a matter of feasibility. If your whole system could be RAM, then the benefits would be tremendous, it's just not economically feasible. That HBM won't have the same latencies as the L3 cache, but it should be twenty times faster than normal RAM, and its size would mean that there would be no code that the CPU would have to go to RAM for. The gains (depending on the application) would be insane. The $10,000 GPU+SSD combo that AMD is selling is getting 3x rendering increases just by having an SSD on the same PCB as the GPU.

Intel did something new for their CPUs by mistake almost. That CPU has 35% lower TDP than other i7s, and yet it's faster than all of them in the same Hz, with lower power consumption. The L4 is also used by the CPU, not just the GPU. In a "normal" system with a separate GPU, it would all be for the CPU use.

Again, no. The reason that companies like HP push for stuff like memristors is that even current CPU/GPUs are massively underutilized due to I/O restraints. The more caches the better.

From Anandtech:

Take a look at their general performance tests too, where things are not GPU-bound. It's in the same ballpark as CPUs with 30-40% more power consumption, all really due to the L4, since the IPC leap is only 3.3%. It seems to destroy the rest in things like web browsing etc.

 

Bristol Ridge will have better igpu performance than any Intel but if Zen gets Ivy/Haswell IPC, GCN4.0 and HBM Intel never catch AMD and in 2nd gen Zen aka "Zen+" AMD Will get the IPC and GFX crown! Mid Range AMD competing with i7 on IPC and iGPU? Now "Dual graphics" Will show big difference thanks to Zen IPC than Bulldozer ark does'nt had

 

You correct, but broadwell was disapointed, that cache elimininated hugher frecuencies/OC capabilities

 

Because it was eDram, and part of it was also that it was the first 14nm Intel CPU. HBM (if it happens), is not on die.

 

Is Intel still pursuing HBC memory or have they dropped that in favor of HBM? Hopefully the latter.

 

I guess you mean HMC. I don't know what kind of future that would have, since HBM is here for more than a year and sold in relatively large quantities already.

 

What's wrong with that power consumption?
It was supposed to be lower, dear lord.

 

Damn.

If that wasn't the case, the 7700K would've been a great CPU for those that don't OC. But for people that OC there's literally 0 point in upgrading. Then again, it's been like that for years. I hung on to my 2600K for ages, only to finally swap when my motherboard and PSU died.

 

NVIDIA and AMD sell us whole multi-layered PCBs, with custom power delivery systems, coolers and ultra-fast memory on top of 4Bn+ transistor GPUs, at prices lower than Intel is selling us 1,75Bn transistor CPUs which don't include the dignity of their own feet for the motherboard.

No matter the scenario, this is the tenth time that Intel sells a quad core CPU as an enthusiast performance part. Even if the CPU was the second coming of Christ, it would be too much at this point.

Spoiler

 

Oh man... you're bringing back memories of bulldozer hype. But just like we saw a few years back, the above is nothing more than a pipe dream.

AMD is claiming a 40% increase in architectural IPC. The major caveat here though is that 40% architectural IPC gain will only translate into an actual 40% performance boost if we’re looking at chips with equivalent clockspeeds. That IPC improvement is compared with the existing Excavator cores, used in the Athlon X4 845 for example. That’s a 65W quad-core CPU running at between 3.4GHz and 3.8GHz and delivers a Cinebench single-threaded index score of just 96.

Doing some rough, back-of-the-napkin maths, if a Zen quad-core is running at 3.8GHz then we could expect to see a score or around 134 in Cinebench’s single-threaded benchmark. Extrapolating that further you’d see the current engineering samples, running at an operating peak of 3.2GHz, performing somewhere around the 113 mark. That’s only a potential performance boost of less than 19% over the Excavator core. They're going to have to ramp the clocks up a lot more in order to get close to Intel.

 

Hey dude, there are some of us with Intel CPUs that have been around long enough to be able to conceive that Intel can lose in a cycle. I'll agree about the pipe dream once benches are in, but until then Bulldozer has been AMD's only launch failure since the 286 clones (and in modern games it's surprising a lot of people on multithreaded benchmarks).

Intel losing their way is more of the norm, in total time, than the opposite.