Coffeelake thread, benchmarks, reviews, input, everything.

[PeterScott]

1280x720? ........

They were trying to make it a CPU benchmark, not GPU benchmark. The higher the resolution, the higher all the CPUs got to each other.

 

[epsilon84]

1280x720? ........

720P is the best way to test CPU gaming performance as the GPU is not the limiting factor. If you want 1080P (or higher) scores, they are in the link. But the difference between the 8600K and 8700K would be even smaller at higher resolutions.

 

[PhonakV30]

True but Should not use this as Metric , We live in reality Not 720p.Example 8350/8700K
720p : 8.2%
1080p : 4%
1440p : 2.1%

Because of 8.2% , Should i buy 8700K ? Hell no.

 

[GaiaHunter]

720P is the best way to test CPU gaming performance as the GPU is not the limiting factor. If you want 1080P (or higher) scores, they are in the link. But the difference between the 8600K and 8700K would be even smaller at higher resolutions.

Ah, the low res benchmarks.
They are a relic from the past when graphics cards were way simpler than they are today.
You might not even be taking advantage of a plethora of GPU trick and optimisations that will have an impact at higher resolutions.

I wouldn't use it to predict future CPU+GPU performance.

 

[eddman]

Great. Another thread derailed. I don't know why some people cannot understand the concept of "removing the GPU limit by using low resolutions to find out a CPU's true performance". It's not quantum science.

Everyone should buy a 7300. It's the same as an 8700K. /s

 

[Bouowmx]

This could be due to TDP throttling. Undervolting the CPU might let it run at 4Ghz in all states.

I don't think that is the cause because Intel overclocking motherboards don't set the power limit to the processor's TDP.

 

[GaiaHunter]

Great. Another thread derailed. I don't know why some people cannot understand the concept of "removing the GPU limit by using low resolutions to find out a CPU's true performance". It's not quantum science.

That assumes that lowering resolution has no other impacts on GPU and/or CPU performance.

You are not finding a CPU true performance with low resolutions benchmarks, you are trying to create "faster GPUs" and extrapolate how the CPU will perform with future "faster GPUs".

Sites love to do these low res benchmarks when new CPUs are out but they never revisit them to see how accurate a predictor of CPU performance they are when faster GPUs are released.

 

[PhonakV30]

Great. Another thread derailed. I don't know why some people cannot understand the concept of "removing the GPU limit by using low resolutions to find out a CPU's true performance". It's not quantum science.

Another thread derailed , Some people use this logic to argument their order buying.It's wrong.Period.

 

[eddman]

You are not finding a CPU true performance with low resolutions benchmarks, you are trying to create "faster GPUs" and extrapolate how the CPU will perform with future "faster GPUs".

Obviously you are not going to get the absolute, exact performance difference, but you can get very close. Using lower resolutions is the best way to find out a CPU's potential. Even digital foundry does it. Do you know a better way?

Yes, it simulates a "faster GPU" which is the whole point of it. Obviously a faster CPU would let a GPU stretch its legs more. It's very simple. How do you find out if a CPU is faster in games? Surely by not benching at 2160.

Another thread derailed , Some people use this logic to argument their order buying.It's wrong.Period.

First try to understand the point of low res tests before calling out other's logic.

 

[Despoiler]

Don't waste your time and energy; the sig tells us all we need to know and what he's advocating for in an Intel thread.

My sig indicates what I own. Not what I advocate. Remember, when I bought what I did Intel was simply not competitive.

The reason I want clock for clock is because Intel did what they've always done. Jumped ahead on process.

 

[raghu78]

My sig indicates what I own. Not what I advocate. Remember, when I bought what I did Intel was simply not competitive.

The reason I want clock for clock is because Intel did what they've always done. Jumped ahead on process.

We already know Intel is ahead clock for clock against Ryzen. Ryzen IPC is Haswell range for non AVX workloads and slightly below Broadwell. Coffelake is on avg >= 10% higher IPC than Ryzen. In memory latency sensitive workloads its definitely > 10%. btw Intel's process lead especially with transistor performance is a big factor in their current leadership position in desktop CPUs. If AMD want to compete with Intel's best products they need their foundry partner GF to pickup its execution and deliver on 12LP and 7LP with competitive transistor performance which would allow them to compete on clocks too against Intel.

 

[slashy16]

My point was, that some people do use their CPU's@100%, and as such, running over like 75c for long periods or thereabouts is NOT good.

The esxi DEV box I use at work currently has a 4790k soon to be replaced with a 7940x(yay). The temps I get running prime are way higher than I could ever achieve running 12vms and be pushing the cpu to it's the limit. CPU's are designed to run hot, The best example is the 200 or so PentiumD desktops we had at work until recently. They ran at 90c all the time and never had any problems. Heat is only a factor when you are pushing for those last few mhz. I have yet to see any reputable reviewer say they have had a problem with the 8700k and throttling while not overclocked. Having to shell out money dollars for a good cooler is a small price to play for the chance to own the fastest CPU available today.

 

[slashy16]

1280x720? ........

How else are you going compare gaming performance? This isn't a video card review.

 

[PhonakV30]

First try to understand the point of low res tests before calling out other's logic.

Try to understand my logic , NOT base on PRICE.

 

[eddman]

Try to understand my logic , NOT base on PRICE.

What are you even talking about?! What price? No one mentioned the word "price". Do you even have a point? Did you understand the concept of low res tests?

 

[BSim500]

I posted this over on the other thread. I'm going to re-quote it below as it's amazing how many people still don't get it after all these years...

For those still struggling with how benchmarks work:-

CPU Test = You minimize being bottlenecked by using the fastest GPU you can find and then lower the resolution if that's not enough, to load the (CPU) cores before the GPU hits 100% usage. Anything else just results in different CPU's idling / downclocking from Turbo waiting on the GPU's at which point you're not properly testing the CPU at full load. Lowering the resolution gives a better idea of how a CPU that's bottlenecked by a GPU today will perform in the future on a stronger GPU. The reason why no-one tests 1440p / 4K on CPU reviews is the same reason why no-one tests 5GHz i7's on GTX 1030's.

GPU Test = You minimize being bottlenecked by using the fastest CPU you can find and then raise the resolution if that's not enough, to load the (GPU) cores before the CPU hits 100% usage. Anything else just results in different GPU's idling / downclocking from Turbo waiting on the CPU's at which point you're not properly testing the GPU at full load. Raising the resolution gives a better idea of how a GPU that's bottlenecked by a CPU today will perform in the future on a stronger CPU. Again, it's the same reason why no-one reviews GTX 1080 Ti's on Celeron's.

Real World tests (eg, "But I play at 1440p / 4K / with 60fps VSync on and only want to see how that limitation affects CPU/GPU balance so I don't overbuy"), etc, is an entirely fair personal metric by itself to gauge the best average CPU/GPU pairing for a given tier of hardware / budget / constraints. But in arguments with other people of different brands, it often gets abused into cherry picking when bottlenecked low fps numbers get discarded one minute when one's favored "team" is behind, after previously being included when it was "the other way around". That's not aimed at anyone here personally, but is a highly visible and tedious trait amongst brand fanboys who seem to populate Youtube comments section and certain other clickbaity "rumor" sites...

In short, 4K / 1440P gamers should still take note of the 720p benchmarks as they basically show how much headroom your CPU has after your next GPU upgrade (ie, longer lifespan between upgrades). The real-world ones like this recent one from Techspot, are there to primarily help lower requirements gamers avoid huge mismatches. And even those are only a rough guide as people can and do play on Med/High vs Ultra which reduces GPU bottleneck (often up into the next tier, eg, in Witcher 3, 1050Ti Med = 1060 Ultra) which then makes even a budget CPU's limitations more pronounced.

 

[PhonakV30]

What are you even talking about?! What price? No one mentioned the word "price". Do you even have a point? Did you understand the concept of low res tests?

I guess You're limited to anandtech forum.Some use it as Perf/price.

 

[eddman]

I guess You're limited to anandtech forum.Some use it as Perf/price.

That's not exactly the subject here but ok, let's compare (based on newegg prices); in BF1 multiplayer at 1080p, an 8400 is 23% faster than a 1600X, and its 99th percentile is 10% higher. A 1600 costs $215, a 1600X $220, and an 8400 $190.

Now obviously this is only the CPU. As of now the only compatible boards are Z370s and they cost more than cheap AMD boards, so a MB+CPU combo is about $35 cheaper for AMD (with a B350 board for OCing).

190 + 120 = $310
215 + 60 = $275

The intel combo is about 13% more expensive. Is it worth it? That's up to you. This would change once intel's B and H boards arrive.

EDIT: Let's add RAM to the calculations. The cheapest 16 GB (2x8) that is not below 2666 right this moment is $143. This would bring the price difference down to ~8.5%.

P.S. the AMD boards have the advantage of supposedly supporting up to Zen 2. Intel's 300 boards' future support is unknown at this point.

 

[Sonikku]

The i3 8350k price is way to high. I rather get the low end 6 core i5 at that price. They need to lower it to $150 bucks or something

It's for high end WoW players. You basically can't get 144fps in the game it's so single threaded cpu bound.

 

[Bouowmx]

I think another interesting analysis in the future is:

AMD B350, Ryzen 5 1600, 3200 MT/s C14 RAM
vs
Intel H310, Core i5-8400, 2666 MT/s C15 RAM, since cache/ring frequency is not tied to memory frequency

 

[epsilon84]

That's not exactly the subject here but ok, let's compare (based on newegg prices); in BF1 multiplayer at 1080p, an 8400 is 23% faster than a 1600X, and its 99th percentile is 10% higher. A 1600 costs $215, a 1600X $220, and an 8400 $190.

Now obviously this is only the CPU. As of now the only compatible boards are Z370s and they cost more than cheap AMD boards, so a MB+CPU combo is about $35 cheaper for AMD (with a B350 board for OCing).

190 + 120 = $310
215 + 60 = $275

The intel combo is about 13% more expensive. Is it worth it? That's up to you. This would change once intel's B and H boards arrive.

P.S. the AMD boards have the advantage of supposedly supporting up to Zen 2. Intel's 300 boards' future support is unknown at this point.

IMO you also need to include the price of DDR4 RAM into the equation, as its part of the overall platform cost. This further dilutes the difference in cost between platforms.

 

[reb0rn]

Is it fact that 8 core we gona see next year coffee lake, no chance it gona be ice lake 10nm?

if 10nm is still not ready and delayed why not ice lake 8core on 14++nm??

Would also like to know best ram for intel at concerning perf/price is it 3000-3200Mhz?

 

[jpiniero]

Is it fact that 8 core we gona see next year coffee lake, no chance it gona be ice lake 10nm?

Need more info/rumors, but it does make sense given how bad things are going with 10 nm.

 

[reb0rn]

I agree, due delay on 10nm we have same 4 generations with no new tech??, its kinda insane not to push new architecture to 14nm then to wait fro AMD to take even a lot more market share

 

[VirtualLarry]

I think that the i5-8400 solution will get a lot more price-competitive once the "H" and "B" boards come out for it. But by that time, we might not be far from Zen+ or Zen 2, or Zen 12nm, or whatever you want to call it. If they can pull off 4.5Ghz, or possibly 5Ghz (OK, a dream, but they got this far already with Zen), then AMD will be sitting pretty again, and all of those people that opted for Zen the first time around, will just have to upgrade their CPUs, not their whole platform.

It's entirely possible, that when Intel comes out with the 8-core consumer chips, that they might require the purchase of a Z390 board, thus preventing any future-proofing of the Z370 platform. I mean, it's pretty much up in the air at this point, but just pointing out, that this is Intel we're talking about, they already pulled that once, with Z370 / CFL, being not forward/backward compatible with other 1151 CPUs.