Coffeelake thread, benchmarks, reviews, input, everything.

[JoeRambo]

"25 percent more ST performance,

Really? Lake CPUs once on 5Ghz clock will have north of 30% better ST performance*. Yes, even in loads that favour Ryzens quad FPU pipes like Cinebench.

So when all is said and done, to undo this fact of ST gap, one needs "meaningful data". Cherry picked games and game settings, 3666Mhz memory for Ryzen, wattage comparison with Prime etc.

Well good luck with that, there are people around who value strong ST performance and honestly Kabylake @ 5Ghz was already good enough for them. CoffeeLake? Even more so, cause Ryzen's advantage in MT productivity is gone now, so no more tradeoffs for desktop CPU.

That is for the top end, enthusiast systems. For lower tiers, different battles of value, $ balancing between parts happen.


* Yes, 31% is north, cherrypicking on the value does not make the gap smaller

 

[DrMrLordX]

I'm not. I purposely brought that up to show the importance of process to the overall performance of a uarch. If you're a chip architect who's designing a uarch for a process that could scale to 5GHz with 40% more power than another process at 4GHz, how do you try to bridge the performance gap for the 4GHz chip if you have 40% more power at your disposal. Do you increase workload per cycle, hence stronger ipc? So if you were to compare the two chips at 4Ghz, who comes up tops? I know a test like that is useful to some, but for me it's meaningless. I don't know why I'm being called out on it.

Edited for clarity.

You have shifted the argument to one where you have a designer's discretion. The argument at hand is to compare two architectures, set in stone, to determine which has the higher IPC. In that scenario, the two architectures should have the same relative proportional delta in IPC at any shared clockspeed.

Coffeelake and . . . that other uarch are already taped out and in production. You can not assume that that . . . other uarch can't scale performance-wise beyond 4 GHz, because a). that isn't true (look at LN2 results, duh) and b). it would be inconsistent with every other uarch that has come before it.

For what it's worth, I'm sure Coffeelake scales just fine beyond 5 GHz as well.

If you look at LN2 results and compare them to air/water results for a number of architectures, you will probably notice that scaling DOES break down beyond a certain point, especially if cache and memory speeds do not increase along with clockspeeds. But that has less to do with the architecture and more to do with cache and memory subsystems.

Why is it that every Intel thread turns into an Intel vs AMD crap fest, but AMD threads hardly ever do?

Hey, some of us tried to stop it.

 

[Hans Gruber]

For what it's worth. Someone from the UK has been benchmarking the 8700K over @ http://cpu.userbenchmark.com The 7700K is the bench of 100%. When CPU's fall below it eg. R3 1200 @ 3.9ghz 84%. The 8700K did 119-127% meaning 20% above the stock 7700K. There was a 127% bench but I think that was OC'd to 5ghz.

 

[elhefegaming]

http://cpu.userbenchmark.com/Compare/Intel-Core-i7-8700K-vs-Intel-Core-i7-7700K/3937vs3647

There's one bench at 5 and one at stock speeds, I'm assuming they ate averaged to get that 23%
Looking good!

 

[StinkyPinky]

It looks real good. Aint competition grand.

 

[Dufus]

You're beating around a bush that's full of incoherent ideas regarding Instructions Per clock, Instructions per cycle(the two are not the same)

Instructions per clock cycle, wether using per clock or per cycle I've always taken it to mean the same and pretty sure can find Intel papers that mention either as the same context. Why do you see it as being different?

 

[TahoeDust]

I am sick of people discussing IPC when discussing Ryzen vs Intel. When comparing two processors with such vastly different frequency capabilities, it is largely a useless argument. Instructions per second (IPS) would be a much more meaningful discussion. When one has a ~25% frequency advantage, who cares if they are close in IPC?

 

[Arachnotronic]

I am sick of people discussing IPC when discussing Ryzen vs Intel. When comparing two processors with such vastly different frequency capabilities, it is largely a useless argument. Instructions per second (IPS) would be a much more meaningful discussion. When one has a ~25% frequency advantage, who cares if they are close in IPC?

There is a perception that clocking higher is easier than improving IPC, which is why people think when they are comparing at iso frequency they are doing "architectural" comparisons.

Truth is, they're all connected.

 

[TahoeDust]

Truth is, they're all connected.

Exactly. I could care less if the competition can match the IPC. Unless they can also match the clock, they can't match the performance.

 

[elhefegaming]

Stupid question I guess, on Oct 5 we'll be able to BUY (and get it home as soon as 1 day shipping) or will we be able to PREORDER and god knows when it ships?

Last Intel CPU I purchased was 2500k when it was already like 6 months old so I don't know how their releases have been lately.

 

[tamz_msc]

Instructions per clock cycle, wether using per clock or per cycle I've always taken it to mean the same and pretty sure can find Intel papers that mention either as the same context. Why do you see it as being different?

Because the clock frequency within a CPU is a reference of time. Cycles refer to periodicity of the waveform of the clock signal. If some instruction takes two cycles to execute, then it will still take two cycles to execute whether the CPU is at stock or overclocked.

 

[tamz_msc]

Really? Lake CPUs once on 5Ghz clock will have north of 30% better ST performance*. Yes, even in loads that favour Ryzens quad FPU pipes like Cinebench.

So when all is said and done, to undo this fact of ST gap, one needs "meaningful data". Cherry picked games and game settings, 3666Mhz memory for Ryzen, wattage comparison with Prime etc.

Well good luck with that, there are people around who value strong ST performance and honestly Kabylake @ 5Ghz was already good enough for them. CoffeeLake? Even more so, cause Ryzen's advantage in MT productivity is gone now, so no more tradeoffs for desktop CPU.

That is for the top end, enthusiast systems. For lower tiers, different battles of value, $ balancing between parts happen.


* Yes, 31% is north, cherrypicking on the value does not make the gap smaller

You keep forgetting how little of an impact a 10 percent overclock on the 7700k has on things like gaming, considering how eager some people are so eager to bring this point up under the garb of "gaming IPC", while conveniently forgetting that frequency scaling and software optimization is quite different across different architectures. Deliberately ignoring these issues, the drones of the juanrga army, which some people here seem to be a part of, waste no opportunity in saying things like Zen has sandy bridge gaming IPC.

What multithread advantage are you speaking of? Look at the Platinum 8160 and Gold 6154. Would you claim that the Platinum 8160 has no multithread advantage over the Gold 6154 because despite having 33 percent more cores, the 6154 has 33 percent higher clock speed?

 

[tamz_msc]

I am sick of people discussing IPC when discussing Ryzen vs Intel. When comparing two processors with such vastly different frequency capabilities, it is largely a useless argument. Instructions per second (IPS) would be a much more meaningful discussion. When one has a ~25% frequency advantage, who cares if they are close in IPC?

That's funny because five years ago 4GHz Sandy Bridge Vs 5GHz Vishera was decided on IPC.

Now it's become irrelevant. M'kay.

 

[Arachnotronic]

That's funny because five years ago 4GHz Sandy Bridge Vs 5GHz Vishera was decided on IPC.

Now it's become irrelevant. M'kay.

It’s perf/MHz and MHz together that matter.

 

[TahoeDust]

That's funny because five years ago 4GHz Sandy Bridge Vs 5GHz Vishera was decided on IPC.

Now it's become irrelevant. M'kay.

Well, my Sandy Bridge (2700k) did 5.1GHz, so I guess I skipped that conversation.

 

[tamz_msc]

It’s perf/MHz and MHz together that matter.

Then answer the Platinum Vs Gold issue. Same perf/MHz, more MHz Vs more cores.

What matters most for these chips, which have an architecture having worse than Sandy Bridge IPC in a certain class of applications?

 

[beginner99]

I think Ryzen is still a very compelling option, 1700+B350 @3.9GHz is likely a good bit cheaper and close enough for most MT tasks,
also Z370 + cooler + higher CPU price (maybe also delid and high end thermal paste?), it can become very expensive compared to how low you can go with Ryzen,

Fact remains that AMD will be the budget option. Good enough for a lower price. Or if you game at 4K the CPU also matters a lot less because you will be spending all your money ond GPUs to get that 60 hz.

One issue I have with Ryzen is that even if we assume same IPC, a 3.9 Ghz Ryzne vs 4.8 ghz Intel OC has the intel at 23% faster clocks. Again if we assume same IPC intel in ST is already 23% faster. That is like the difference from Ivy Bridge to Haswell. Meaning given the lack of ST gains in recent years, that 23% clock-difference which is even more in performance if IPC is factored in, is rather huge and can make your CPU last another 2-3 years longer.

 

[VirtualLarry]

Fact remains that AMD will be the budget option. Good enough for a lower price. Or if you game at 4K the CPU also matters a lot less because you will be spending all your money ond GPUs to get that 60 hz.

One issue I have with Ryzen is that even if we assume same IPC, a 3.9 Ghz Ryzne vs 4.8 ghz Intel OC has the intel at 23% faster clocks. Again if we assume same IPC intel in ST is already 23% faster. That is like the difference from Ivy Bridge to Haswell. Meaning given the lack of ST gains in recent years, that 23% clock-difference which is even more in performance if IPC is factored in, is rather huge and can make your CPU last another 2-3 years longer.

One factor that you're not considering, is that Ryzen gains more (IPC-wise) from their SMT implementation, than Intel does with HT.

So, on purely ST benchmarks, a clock-for-clock comparison might show Intel's Sky/Kaby/Coffee Lake (same core) ahead by a bit.

But expand that to include HyperThreading on Intel, and SMT on Ryzen, and run a core/core, thread/thread, clock/clock comparison, and see where the "IPC comparison" lies. It may even show Ryzen coming out ahead, in their "measured multi-threaded IPC", due to the superiority of their SMT implementation.

 

[Zucker2k]

One factor that you're not considering, is that Ryzen gains more (IPC-wise) from their SMT implementation, than Intel does with HT.

So, on purely ST benchmarks, a clock-for-clock comparison might show Intel's Sky/Kaby/Coffee Lake (same core) ahead by a bit.

But expand that to include HyperThreading on Intel, and SMT on Ryzen, and run a core/core, thread/thread, clock/clock comparison, and see where the "IPC comparison" lies. It may even show Ryzen coming out ahead, in their "measured multi-threaded IPC", due to the superiority of their SMT implementation.

Superior in what sense? Are you suggesting Intel's SMT implementation is inefficient? Or is this a case of a stronger, more efficient core leaving little resources for smt?

 

[TahoeDust]

One factor that you're not considering, is that Ryzen gains more (IPC-wise) from their SMT implementation, than Intel does with HT.

So, on purely ST benchmarks, a clock-for-clock comparison might show Intel's Sky/Kaby/Coffee Lake (same core) ahead by a bit.

But expand that to include HyperThreading on Intel, and SMT on Ryzen, and run a core/core, thread/thread, clock/clock comparison, and see where the "IPC comparison" lies. It may even show Ryzen coming out ahead, in their "measured multi-threaded IPC", due to the superiority of their SMT implementation.

Well, let's find out. Here is 1c/2t of a 7820x running 4.0GHz. Someone with a Ryzen chip run the same 1c/2t at 4.0GHz and see what is what.

 

[elhefegaming]

Can you guys stop it with intel vs amd for fraks sake?

I posted a coffelake related question (albeit maybe a stupid one) and was completely ignored, can we keep it on subject please?

 

[Zucker2k]

Well, let's find out. Here is 1c/2t of a 7820x running 4.0GHz. Someone with a Ryzen chip run the same 1c/2t at 4.0GHz and see what is what.

This is best case scenario for Ryzen.

 

[Zucker2k]

Stupid question I guess, on Oct 5 we'll be able to BUY (and get it home as soon as 1 day shipping) or will we be able to PREORDER and god knows when it ships?

Last Intel CPU I purchased was 2500k when it was already like 6 months old so I don't know how their releases have been lately.

It's hard to tell for now.

 

[TahoeDust]

This is best case scenario for Ryzen.

Absolutely. It is absolutely a useless comparison for actual use...still curious.

 

[VirtualLarry]

Well, let's find out. Here is 1c/2t of a 7820x running 4.0GHz. Someone with a Ryzen chip run the same 1c/2t at 4.0GHz and see what is what.

4.0Ghz might be a bit hard for me to do, I've only been up to 3.8Ghz, I'm on stock air cooling on the rig I was planning to test with. I can give you a benchmark @ 3.8Ghz in a bit, maybe.

Edit: I've got one of my Ryzen rigs booted at 3.80Ghz, with 2x8GB DDR4-3000 @ 2667, with a (2+0) CCX configuration. They didn't offer a (1+0) config in the UEFI.

Edit: Ok, I managed to get my Ryzen 5 1600 to run at 4.000Ghz, @ 1.400V (UEFI), or 1.488V (CPU-Z).
It only boots once out of like 5-10 tries at these settings.



If you look carefully, that's with a 2+0 CCX configuration, and the CB R15 MT is almost double yours. Btw, my RAM is dual-channel, at 2667, 18-18-18-38.

Yet, your ST is quite a bit higher.

So, why isn't your MT quite a bit higher? Because the SMT on my Ryzen made up for the ST deficit, and is practically matching your score for MT, if you cut my score in half, because I was running two cores rather than one.