Coffeelake thread, benchmarks, reviews, input, everything.

[epsilon84]

Did you miss the 8600k?

pretty much decimates the 7700k and its similar scenario, replaces 4 logical cores with 2 real cores.

IN HTT optimal workloads it just about beats a 7700k, in every other multi core workload (Which is almost all workloads) it completely overwhelms it.

People need to look past cinebench, it sadly is a very popular benchmark, but it is also very unrealistic.

Check even userbench.

8/8 without a doubt is better than 6/12. Too many people are obsessed with HTT, and its worrying, the amount of people wasting money on 8700Ks vs 8600Ks as an example.

The 8600K hardly 'decimates' a 7700K: https://www.techpowerup.com/reviews/Intel/Core_i5_8600K/18.html

At stock they are practically equal, in fact the 7700K beats it slightly, though it does have slightly higher turbo clocks (4.2GHz 7700K vs 4.1GHz 8600K all core turbo). It's close enough that I'd call it a draw.

 

[epsilon84]

8/8 will beat 6/12 hands down in games that can handle the extra cores, virtually no games even benefit from HTT.

In almost all gaming benchmarks 8600Ks match 8700Ks, 95% of steam games cant even handle more than 2 threads, never mind more than 6 or 8 threads. The ones that do have more than 8 threads wont benefit in 99% of cases if its fake/logical cores.

When a 8700k noticeably beats a 8600k it may be due to higher stock speed or higher cache size.

Hands down? At most, a 9700K will be perhaps ~5% faster than a 8700K, and that would be down to the higher clockspeed rather than 2 extra physical cores.

 

[PeterScott]

The 8600K hardly 'decimates' a 7700K: https://www.techpowerup.com/reviews/Intel/Core_i5_8600K/18.html

At stock they are practically equal, in fact the 7700K beats it slightly, though it does have slightly higher turbo clocks (4.2GHz 7700K vs 4.1GHz 8600K all core turbo). It's close enough that I'd call it a draw.

IIRC 7700K All core turbo is 4.5GHz single core, 4.4GHz all core.

Depending on the nature of the benchmark it could make a significant difference.

It should be possible to find a clock to clock comparison.

Hands down? At most, a 9700K will be perhaps ~5% faster than a 8700K, and that would be down to the higher clockspeed rather than 2 extra physical cores.

I agree there won't be much difference in most benchmarks, but we really don't know clock speed. I here 5.0GHz single core boost.

I would definitely prefer 8C/8T with higher clock speed over 6C/12T with lower, hands down.

 

[PeterScott]

People need to look past cinebench, it sadly is a very popular benchmark, but it is also very unrealistic.

Yeah it's the favorite application benchmark for a bunch of people that never use the application it is based on.

Kind of like Ashes of the Singularity is a favorite gaming benchmark, for a game that practically no one plays.

They both tend to show some advantage for higher core counts with HT/SMT. But both are probably poorly representative of what people are actually doing with their computers.

I noted getting ready to for my next computer, that I only do two activities that even tax my ancient PC (2008 vintage Q9400 at 3.2GHz).

Those two things are:

Gaming.
Video encoding.

So those are the benchmarks I actually pay attention to. I mainly use hanbrake for Video encoding, and the game I want to play is Witcher 3, so those are my preferred benches... The other benches are mainly curiosity.

 

[jpiniero]

https://browser.geekbench.com/v4/cpu/9128572

Geekbench 4 Single result nearing 10k with an 8086K.

 

[mikk]

At stock they are practically equal, in fact the 7700K beats it slightly, though it does have slightly higher turbo clocks (4.2GHz 7700K vs 4.1GHz 8600K all core turbo). It's close enough that I'd call it a draw.

Because of a clock speed difference, 7700k clocks higher. A typical multithread gain with SMT is something between 20-25%, a core increase from 6 to 8 gives 33% more cores resulting in 25-30% more speed in real world usually. Potentially 8/8 should be slightly faster than 6/12 in multithread applications. For gaming physical cores are much better, advantage should be higher.

 

[epsilon84]

Because of a clock speed difference, 7700k clocks higher. A typical multithread gain with SMT is something between 20-25%, a core increase from 6 to 8 gives 33% more cores resulting in 25-30% more speed in real world usually. Potentially 8/8 should be slightly faster than 6/12 in multithread applications. For gaming physical cores are much better, advantage should be higher.

Is there even such a thing as a 'typical'gain for SMT/HT? It seems to vary a lot based on the application, as for gaming, HT helps if you're thread limited, about 6 - 8 threads is where the point of diminishing returns kicks in though, so it's very useful for a quad core but has less effect as you increase physical cores.

Still, even at 6C, in thread heavy games like BF1 or AC:O you'll find a 8700K beating a 8600K clock for clock, but whether that will still be the case with a 8/8 9700K remains to be seen

 

[JoeRambo]

Still, even at 6C, in thread heavy games like BF1 or AC:O you'll find a 8700K beating a 8600K clock for clock, but whether that will still be the case with a 8/8 9700K remains to be seen

How much of that beating comes from "threads" and how much from 9MB vs 12MB of L3 cache? Games do love cache, and 9700K will also have 12MB of L3 that surely will help performance

 

[chrysalis]

The 8600K hardly 'decimates' a 7700K: https://www.techpowerup.com/reviews/Intel/Core_i5_8600K/18.html

At stock they are practically equal, in fact the 7700K beats it slightly, though it does have slightly higher turbo clocks (4.2GHz 7700K vs 4.1GHz 8600K all core turbo). It's close enough that I'd call it a draw.

try https://www.techpowerup.com/reviews/Intel/Core_i5_8600K/12.html (same source as you)
or
https://www.overclock3d.net/reviews/cpu_mainboard/amd_ryzen_5_2600x_and_ryzen_7_2700x_review/15 <- 8600k above 8700k

I am talking about gaming, not encoding, not cinebench, not 7zip.

99% of gamers, dont encode videos or stream.

When you get youtubers and the like promoting HTT, its because those guys sit editing videos all day, have niche workloads, then we get a masse of gamers buying into some kind of false dawn.

I can flip the coin if we want to play the e-dick benchmark game, on intel's XTU AVX test my 8600k scores higher than a 7700k by 54%, because of its 2 extra cores and the fact that test gains 0% from HTT. In real world workloads real cores mean vastly more than logical cores. All logical cores do is help fill "WAIT" gaps in cpu core time with more processing time, they work in scenarios where 100% cpu usage is reported to the OS for a core but the cpu itself is not quite loaded up, games dont generate that kind of load. Most desktop apps dont either, its a niche type of workload.

Also HTT unless it gets radically redesigned its days are numbered, there is vulnerabilities that are been found which will soon have public disclosures that will see that feature get disabled as default behaviour.

 

[epsilon84]

try https://www.techpowerup.com/reviews/Intel/Core_i5_8600K/12.html (same source as you)
or
https://www.overclock3d.net/reviews/cpu_mainboard/amd_ryzen_5_2600x_and_ryzen_7_2700x_review/15 <- 8600k above 8700k

I am talking about gaming, not encoding, not cinebench, not 7zip.

99% of gamers, dont encode videos or stream.

When you get youtubers and the like promoting HTT, its because those guys sit editing videos all day, have niche workloads, then we get a masse of gamers buying into some kind of false dawn.

What exactly am I supposed to see? The 7700K actually beats the 8600K in most of the gaming tests, and the second link shows an i3 8350K at the top of the charts along with the 8600K, so I have to question the validity of those results. I can see a 8600K beating a 8700K in niche situations where there is some form of performance hit from HT, but there is no way a 8350K, with only 4c/4t should come close, let alone beat a 8700K in gaming

 

[chrysalis]

The 8350k is up there because the majority of games cant do more than a 4 core load.

OC3D in his reviews say it right when per core performance is still king, that cpu has killer per core performance hence its up there.

Also look at the games that get reviewed, these reviews show high core count in the best possible light. The vast majority of games on steam (pretty much all indie games for a start) dont support more than maybe 1-2 cores, they coded for low spec'd machines, yet the reviewers for the games they bench use fancy triple AAA titles inside that small sub 5% of games painting a misleading picture for PC gaming.

 

[epsilon84]

The 8350k is up there because the majority of games cant do more than a 4 core load.

OC3D in his reviews say it right when per core performance is still king, that cpu has killer per core performance hence its up there.

Also look at the games that get reviewed, these reviews show high core count in the best possible light. The vast majority of games on steam (pretty much all indie games for a start) dont support more than maybe 1-2 cores, they coded for low spec'd machines, yet the reviewers for the games they bench use fancy triple AAA titles inside that small sub 5% of games painting a misleading picture for PC gaming.

Well indie games are a far cry (excuse the pun) from AAA games like... Far Cry

Sorry just had to say it.

I play BF1 and there is no way I would go back to 4C/4T on that game, it's an absolute stutter fest on 64p maps.

Per core performance is still important of course, but for most popular games these days you need a minimum of 6 threads available to 'max out' the game, or not bottleneck the GPU

 

[VirtualLarry]

Also HTT unless it gets radically redesigned its days are numbered, there is vulnerabilities that are been found which will soon have public disclosures that will see that feature get disabled as default behaviour.

QFP, waiting for info...

 

[jpiniero]

Intel (or China, ha) is apparently doing another 200 series chipset rebrand for Coffee Lake's socket, called H310C. The Chinese site also mentions DDR3 support.

 

[VirtualLarry]

DDR3 on Coffee Lake would be interesting to see.

 

[PingSpike]

Also interested in DDR3 coffeelake. You'd be leaving performance on the table but with ram as expensive as it is it would be a nice option. Its pretty funny coffeelake still has the DDR3 controller though. Seems it was only put on there as a skylake stopgap for low power laptops until 10nm parts came out with LPDDR4 support. But skylake just keeps getting re-released so its still hanging around.

 

[JoeRambo]

The problem with DDR3 is that Skylake only supported DDR3L, so memory with 1.5V or those 1.65V OC sticks are no go. I am certain CFL has same if not tighter voltage limits, and personally i'd not go above 1.4V.

 

[epsilon84]

Interesting to see the performance hit from DDR3 compared to DDR4 for CFL. I remember Skylake comparisons and it was usually only a few % at worst, but that is so with only 4 cores. Is DDR3 level bandwidth enough to properly feed 6 or 8 cores? I have my doubts

 

[jpiniero]

https://videocardz.com/76910/intel-roadmaps-leaked-core-i9-9900k-coming-next-year

Intel may have delayed the Coffee Lake Refresh launch until next year. Also suggests there might be a locked i9 desktop part coming. It would make sense if they are really that capacity constrained and/or they know any 10th gen desktop parts aren't coming until 2020 so they want to get things aligned to a year.

 

[Shivansps]

DDR3 on Coffee Lake would be interesting to see.

where you even get a 1151 DDR3 board these days.

 

[PeterScott]

https://videocardz.com/76910/intel-roadmaps-leaked-core-i9-9900k-coming-next-year

Intel may have delayed the Coffee Lake Refresh launch until next year. Also suggests there might be a locked i9 desktop part coming. It would make sense if they are really that capacity constrained and/or they know any 10th gen desktop parts aren't coming until 2020 so they want to get things aligned to a year.

Strategically sound, but kind of a bummer.

 

[PingSpike]

where you even get a 1151 DDR3 board these days.

Straight from China it seems. The supply dried up on newegg for the 100 series DDR3 boards maybe 6 months ago. I was sort of keeping an eye on it because I kept thinking of giving it a try. Ultimately I'd always come back to the same thing, which was that I'd still only get 4 cores out of kabylake so might I well just forget it and just use an old DDR3 supporting platform. But a company called Onda seems to be selling boards that use DDR3 and advertise coffeelake support.

A user on reddit just posted his benchmark results currently a coffeelake quad with a Chinese motherboard.
http://www.userbenchmark.com/UserRun/9998679

He says they sell ones that claim to support 6 core as well. It not clear to me if that board is a H310 or a hacked/modified H110 in in H310 clothing. Its also not clear to me if that makes any difference in functionality though. If the stories are true they make weird x79 motherboards that use z68 and b75 south bridges and let you use cheap old Xeons off ebay.

 

[Shivansps]

this one i guess? https://www.newegg.com/Product/Prod...33&cm_re=h110_ddr3-_-1B3-007J-00004-_-Product

 

[PingSpike]

this one i guess? https://www.newegg.com/Product/Prod...33&cm_re=h110_ddr3-_-1B3-007J-00004-_-Product

This is the directly link I saw:
http://www.onda.cn/MotherBoard_Designs.aspx?id=397

If I run it through google translate The description for the board says "eighth generation support" or something like that. I can't really glean much though because I can't read a lick of chinese

 

[SPBHM]

if it supports the 8400... that's a great option for people with old ram...
ram is a bottleneck, but I doubt that it would make the 8400 slower than stuff older than haswell.