Intel "Coffee Lake" Builders Thread

[PeterScott]

Why do people always make the same assumption that you get a 100% linear perf scaling with increased clock ? This never happens, atleast in majority of apps. 8700 at 4.3 Ghz with HT will definitely be faster than 8600k at 5.0 Ghz for MT workloads. Proof : have a look at 8600k at 4.8 Ghz not even close to 8700k (8700 has the same all core turbo)

https://www.techpowerup.com/reviews/Intel/Core_i5_8600K/9.html

Are you sure MCE isn't enabled. That is a problem with 8700K reviews.

 

[raghu78]

Are you sure MCE isn't enabled. That is a problem with 8700K reviews.

TPU scores are without MCE . HWC scores with MCE
http://www.hardwarecanucks.com/foru...el-coffee-lake-i7-8700k-i5-8400-review-5.html

 

[Justinbaileyman]

I am getting to the point that I am getting ready to buy what ever CPU comes into stock first I am sick of waiting.

 

[elhefegaming]

I am getting to the point that I am getting ready to buy what ever CPU comes into stock first I am sick of waiting.

That's why I got the 8400 😛

I have a felling that by the time the 8700k will be fully in stock and available, we will be very close to the upcoming z390 and gen 9

 

[crashtech]

Why do people always make the same assumption that you get a 100% linear perf scaling with increased clock ? This never happens, atleast in majority of apps. 8700 at 4.3 Ghz with HT will definitely be faster than 8600k at 5.0 Ghz for MT workloads. Proof : have a look at 8600k at 4.8 Ghz not even close to 8700k (8700 has the same all core turbo)

https://www.techpowerup.com/reviews/Intel/Core_i5_8600K/9.html

Because within the relatively tight frequency range within which most comparisons are run, performance does scale relatively linearly in many apps. From 2 to 3 to 4GHz might not be linear at all, but from 3.6 to 3.8 to 4.0GHz will appear much more so.

 

[krumme]

Honestly, I think the 8600K is in an awkward spot price wise and apart from gamers I really can't see its value when compared to other products around its price range.

As I said earlier, it's basically the same price as the 8700 non k once you factor in the price of a HSF, but even when overclocked to 5GHz it is only 400MHz faster than the 1C turbo of the 8700. Sure, it will be 700MHz faster when all 6 cores are under load, but if you are actually doing something that is pegging all 6 cores at full load you may as well get the 8700 because it has HT, and in terms of MT throughput a 6C/12T @ 4.3GHz >>> 6C/6T @ 5GHz

Imo add a cheap non z mb to the 8400 and 8700. Run your 3000c15 ram at 2666c14 😉 and they are the sweet gaming deal.
You cant buy them before dec january anyway.
Add you dont have to add some expensive or fancy cooling to both keep it silent and prevent trotling as you will have to do on oc 8600k or 8700k.

 

[raghu78]

Because within the relatively tight frequency range within which most comparisons are run, performance does scale relatively linearly in many apps. From 2 to 3 to 4GHz might not be linear at all, but from 3.6 to 3.8 to 4.0GHz will appear much more so.

There is a difference between linear and perfect 100% clock to perf scaling. There are very few cases where a 10% clock increase gives 10% perf gain . In most cases it’s probably between 6-8% perf gain for 10% clock increase.

 

[PeterScott]

There is a difference between linear and perfect 100% clock to perf scaling. There are very few cases where a 10% clock increase gives 10% perf gain . In most cases it’s probably between 6-8% perf gain for 10% clock increase.

Nonsense.

 

[crashtech]

There is a difference between linear and perfect 100% clock to perf scaling. There are very few cases where a 10% clock increase gives 10% perf gain . In most cases it’s probably between 6-8% perf gain for 10% clock increase.

Your example doesn't prove non-linearity, though. If perf increases 6% for a 10% clock increase, then another 6% for another 10% clock increase, that's linear, just that that slope is less than one.

 

[PeterScott]

Your example doesn't prove non-linearity, though. If perf increases 6% for a 10% clock increase, then another 6% for another 10% clock increase, that's linear, just that that slope is less than one.

60% is total nonsense to start with.

Clock speed scaling is much closer to 1:1

Here is the picture of 8600K OC to 5.5GHz.
http://abload.de/image.php?img=5500-5090-3866-cb15-1dnsw3.png
Cinebench @5.5GHz = 1385

Here is picture of the 8600K OC to 4.8GHz.
https://tpucdn.com/reviews/Intel/Core_i5_8600K/images/cinebench_multi.png
Cinebench @4.8GHz = 1196

5.5/4.8 = 15% clock-speed increase
1385/1186 = 16% performance increse.

There are NO scaling issues at all moving from 4.8GHz to 5.5GHz.

Performance did not fall off, as he keeps erroneously predicting.

 

[Justinbaileyman]

Any chance we could get like a low powered i7 8700t 25w or 35w part? I could for sure use a 6c/12t part but would love a energy efficient ultra low power version. I know the i7 8700 non "K:" version is 65w already so maybe if they drop the stock clock to something like 2.6Ghz with a turbo boost up to 3.6Ghz would be perfect for me.

 

[frozentundra123456]

7820X also has 4 more threads than the 8700K.

The 7820k also has faster ram, and could be running into a gpu limit on both, since everything tops out around 160 fps. Overall, seems a very poor comparison of apples to oranges: Overclocked 8700k vs (I assume) stock 7700k vs HEDT chips with faster ram. Hard to tell anything really, except that overclocking (obviously) helps the 7820x.

 

[Markfw]

Any chance we could get like a low powered i7 8700t 25w or 35w part? I could for sure use a 6c/12t part but would love a energy efficient ultra low power version. I know the i7 8700 non "K:" version is 65w already so maybe if they drop the stock clock to something like 2.6Ghz with a turbo boost up to 3.6Ghz would be perfect for me.

Undervolt/underclock it will you hit the number you want.

 

[Bouowmx]

Any chance we could get like a low powered i7 8700t 25w or 35w part? I could for sure use a 6c/12t part but would love a energy efficient ultra low power version. I know the i7 8700 non "K:" version is 65w already so maybe if they drop the stock clock to something like 2.6Ghz with a turbo boost up to 3.6Ghz would be perfect for me.

Get i7-8700. Set power limit to 35 W in BIOS or Intel Extreme Tuning Utility. No need to wait. Plus, power limit can be set back to 65 W, for flexibility of being able to use high frequencies, whereas the T models will be stuck at its rated frequencies.

 

[Justinbaileyman]

Undervolt/underclock it will you hit the number you want.

Yeah I kinda figured I could underclock, Just thought lower clocked lower power part would be a cheaper buy for me.

Get i7-8700. Set power limit to 35 W in BIOS or Intel Extreme Tuning Utility. No need to wait. Plus, power limit can be set back to 65 W, for flexibility of being able to use high frequencies, whereas the T models will be stuck at its rated frequencies.

Wow never know that was a thing? first time I ever heard you could change power and TDP limits. I Know you can add and take away voltages but not sure how one would go about changing wattage's from like 95w to 35w.
Any one with one of these cpu could you please test this and let me know what the max speed is running at 35w?

 

[raghu78]

60% is total nonsense to start with.

Clock speed scaling is much closer to 1:1

Here is the picture of 8600K OC to 5.5GHz.
http://abload.de/image.php?img=5500-5090-3866-cb15-1dnsw3.png
Cinebench @5.5GHz = 1385

Here is picture of the 8600K OC to 4.8GHz.
https://tpucdn.com/reviews/Intel/Core_i5_8600K/images/cinebench_multi.png
Cinebench @4.8GHz = 1196

5.5/4.8 = 15% clock-speed increase
1385/1186 = 16% performance increse.

There are NO scaling issues at all moving from 4.8GHz to 5.5GHz.

Performance did not fall off, as he keeps erroneously predicting.

Thats a ridiculously biased way to drive an explanation. You cannot arbitrarily take two different systems with different memory speeds and calculate core clock speed scaling. the 8600k at 5.5 Ghz is running at 1933 Mhz (DDR4 3866 Mhz) while the tpu review runs at DDR4 3200. Why don't you use the tpu 8600k results at stock with 4.1 Ghz all core turbo and 4.8 Ghz OC.

4.1 Ghz CB R15 MT score - 1051.66
4.8 Ghz CB R15 MT score -- 1196.14

1196.14/1051.66 = 1.137

4800/4100 = 1.170

So for a 17% clock speed scaling you get a 13.7% perf gain. 13.7 / 17 * 100= 80. Just like what I said. 80% perf gain for 100% clock gain. btw cinebench has very good perf scaling with core clock and maxes out the cpu at 100% usage. Not every app scales as well as CB. Now who was talking nonsense ?

Your example doesn't prove non-linearity, though. If perf increases 6% for a 10% clock increase, then another 6% for another 10% clock increase, that's linear, just that that slope is less than one.

I never argued against linear perf scaling. I said perf never scales 1:1 with clock increase and thats a measurable fact.

 

[BrandonT]

I could for sure use a 6c/12t part but would love a energy efficient ultra low power version.

Change the lightbulb in your bedside lamp to an LED for $10. Boom, done.

I don't mean to single you out, but after months of reading hysterical comments on "power hog" cpu's I think we've lost a wee bit of perspective on just how much energy we're talking about.

 

[PeterScott]

Thats a ridiculously biased way to drive an explanation. You cannot arbitrarily take two different systems with different memory speeds and calculate core clock speed scaling. the 8600k at 5.5 Ghz is running at 1933 Mhz (DDR4 3866 Mhz) while the tpu review runs at DDR4 3200. Why don't you use the tpu 8600k results at stock with 4.1 Ghz all core turbo and 4.8 Ghz OC.

4.1 Ghz CB R15 MT score - 1051.66
4.8 Ghz CB R15 MT score -- 1196.14

1196.14/1051.66 = 1.137

4800/4100 = 1.170

So for a 17% clock speed scaling you get a 13.7% perf gain. 13.7 / 17 * 100= 80. Just like what I said. 80% perf gain for 100% clock gain. btw cinebench has very good perf scaling with core clock and maxes out the cpu at 100% usage. Not every app scales as well as CB. Now who was talking nonsense ?

Because at stock, speeds don't stay uniform across all cores, all the time. You can get single core and quad core speeds faster than all core speed, depending on how the load shifts.

So to properly check clock speed scaling, without the confounding factor of turbo boost, you have to run at speeds where the clock speed doesn't change with active core count, so above the single core turbo speed.

Which is why Comparing 4.8 vs 5.5 is a better scaling comparison, than anything at stock speeds with shifting turbo speeds.

If scaling was really as bad as you say, then 5.5GHz on another system should easily show the negative effect, regardless of other incidental differences.

Here is a CPU that doesn't have Turbo Boost to mess up the results (8350K).

https://tpucdn.com/reviews/Intel/Core_i3_8350K/images/cinebench_multi.png
@4 GHz: 683.88 cb
@4.5 GHz: 764.32 cb

12.5% clockspeed vs 11.8% performance. Very close to one to one.

Nowhere near the nonsense 60% scaling you claim.

 

[BrandonT]

In keeping with the spirit of this being a build thread, what would be the sweet spot for RAM specs for an 8700k, with minimal to no overclocking (though running XMP profile would be acceptable)?

 

[Crono]

In keeping with the spirit of this being a build thread, what would be the sweet spot for RAM specs for an 8700k, with minimal to no overclocking (though running XMP profile would be acceptable)?

This is probably my sweet spot. 3600, CAS 15 (drop down to 14 if you don't mind dropping frequency), 2 X 8GB, $220. Price becomes prohibitive (for me) for faster + lower latency, which of course would be preferable otherwise.

 

[Bouowmx]

Stock: 2666 MT/s C15
Low latency (my sweet spot): 3200; 14
Even lower latency: 3600 (this frequency may not be compatible with some motherboards); 15

 

[PeterScott]

This is probably my sweet spot. 3600, CAS 15 (drop down to 14 if you don't mind dropping frequency), 2 X 8GB, $220. Price becomes prohibitive (for me) for faster + lower latency, which of course would be preferable otherwise.

I am cheaper. I think 3200 offers a lot more competition, so cheaper pricing and very close performance:
2x8 $160
https://www.newegg.com/Product/Product.aspx?Item=N82E16820231977

 

[Justinbaileyman]

$220 for 2 8GB sticks?? that is just insane, they need to stop all this price gouging already. I think my sweet spot is going to be 4 stick of 3200 if prices ever come down but for now I am stuck with 4 sticks of 2400 🙁

 

[StinkyPinky]

I went with 3200/CL14 for my build. I think either 3600/CL16 or 3200/CL14 is the best option although you do pay a premium for it.

 

[TheF34RChannel]

Ehm just got a shipped notification for the CPU 😀 Since there's no Apex boards I think I'll need to go with the Hero as that's the only board in stock or wait until November...