Speculation: i9-9900K is Intel's last hurrah in gaming

[epsilon84]

Good point. The two big open questions are: (1) Are AMD's gaming performance claims fair and representative? (2) How overclockable is the Ryzen 3000 series?

If matching or beating i9-9900K only requires toggling an overclocking setting in BIOS or the Ryzen Master utility, then it will be hard for Intel's status as champion not to be in dispute — especially if i9-9900K has higher power use and/or cost. For i9-9900K to really stretch its legs, it has to go well beyond its rated TDP, as thoroughly explored by reviewers, and discussed elsewhere on this forum.

Actually the 'locked 95W' 9900K doesn't suffer nearly as badly in gaming as it does in heavily threaded apps, which makes sense. Basically no game can exhaust a 9900K thread wise, so it's never truly running at 100% load when gaming.

 

[epsilon84]

Yeah, in the context of the thread poll, the question is how much frequency is required for Ryzen 3000 to take the lead in gaming. And, on the other hand, how much it will have to lose, for reviewers to still recommend Intel for "pure gaming".

Even if the 9900K ends up being 5% quicker, I think the lower price of Ryzen will get it favourable recommendations.

I'm not sure how important it is for AMD to get 'outright' gaming performance leadership, apart from fanboys. On a personal level it would be great if AMD can finally exceed Intel at gaming as its basically been an Intel dominated domain since Core 2 - that's 14 years! Dominance that has been far too long and one sided for my liking (imagine if nVidia beat AMD for 14 years straight?!), and I'm frankly getting a bit bored of 'Skylake IPC' and a lot of people probably are too.

I know AMD is basically 'close enough' even as it is right now even without 7nm, but I really want a reason to buy AMD for gaming apart from the 'its cheaper' line. I want cheaper AND faster Yes I'm hard to please...

 

[DrMrLordX]

It's a psychological issue, 5GHz that is. Hell, if we used base 9 we would be 5GHZ+ already with Ryzen 3000.

Imagine where we'd be in binary!

Back in 2013, AMD released a Piledriver CPU (the FX-9590) with a stock turbo clock of 5GHz. No one cared - because it sucked. Back when Intel was pushing the Pentium 4, they were the first to hit 2GHz - but AMD's Socket A Athlon chips were still better.

Not many people really obsess that much over clock rates.

It's more than that, though. Pinnacle Ridge, clock-per-clock, could hang with Skylake in a lot of applications (including games). What it couldn't do was clock higher than 4.3 GHz consistently. Give Pinnacle Ridge 5 GHz and it's competitive.

Now add +15% IPC (average) on top of that.

 

[rbk123]

Not many people really obsess that much over clock rates.

You're missing the point. If, with current lower clocks, AMD can only tie Intel, then they need to hit equal clocks to outright beat them.
DrMr and Vattila got it right.

 

[moinmoin]

Good point. The two big open questions are: (1) Are AMD's gaming performance claims fair and representative? (2) How overclockable is the Ryzen 3000 series?

If matching or beating i9-9900K only requires toggling an overclocking setting in BIOS or the Ryzen Master utility, then it will be hard for Intel's status as champion not to be in dispute — especially if i9-9900K has higher power use and/or cost. For i9-9900K to really stretch its legs, it has to go well beyond its rated TDP, as thoroughly explored by reviewers, and discussed elsewhere on this forum.

Regarding how overclockable Zen 2 on TSMC 7nm is, I'm starting to think the 5GHz rumor was actually well founded but not necessarily for the Ryzen 3000 series, instead for selected dies. So far 3950X appears to be the top end for the Ryzen 3000 series, and it boasts a boost speed of 4.7GHz. Taking the previous gens as an indication: Top speed Threadripper chips add 100Hz to the boost speed through better binned dies. XFR adds up to 200Hz on top of that. If all that's unchanged this gen some Threadripper 3 chips will be able to reach 5GHz at stock.

 

[exquisitechar]

That slide that shows the performance increase from PBO (up to 200MHz auto OC) gives a hint at the OC capabilities. The 3600 gets a big boost in CB single score, while the 3900x's score increases by a mere 8 points. Meaning, the 3900x's 4.6GHz turbo is likely close to its limit...

 

[JDG1980]

You're missing the point. If, with current lower clocks, AMD can only tie Intel, then they need to hit equal clocks to outright beat them.
DrMr and Vattila got it right.

In the Computex presentation, it was claimed that the Ryzen 7 3800X (8C/16T with 3.9 GHz base and 4.5 GHz boost clocks) can match or beat Intel's 9900K in single-thread performance, specifically including gaming. If that's true (and of course we will have to wait for independent benchmarks to verify it), then it follows that the 3900X, with an even higher 4.6 GHz boost, and especially the 3950X, with 4.7 GHz boost, should be able to beat the 9900K outright in ST/gaming even when falling short of the "magic" 5 GHz mark.

 

[DrMrLordX]

That slide that shows the performance increase from PBO (up to 200MHz auto OC) gives a hint at the OC capabilities. The 3600 gets a big boost in CB single score, while the 3900x's score increases by a mere 8 points. Meaning, the 3900x's 4.6GHz turbo is likely close to its limit...

Possibly. That contradicts some of what we're hearing about both the 3600/3600x and 3900x though. More testing required.

 

[ondma]

Actually the 'locked 95W' 9900K doesn't suffer nearly as badly in gaming as it does in heavily threaded apps, which makes sense. Basically no game can exhaust a 9900K thread wise, so it's never truly running at 100% load when gaming.

Yea, I have an 8700k, only six cores, but temps and power use in gaming or any other kind of everyday use is a complete non-issue. Overclocked to 4.7 all core, and uses only 60 watts at the max in gaming.

 

[rbk123]

In the Computex presentation, it was claimed that the Ryzen 7 3800X (8C/16T with 3.9 GHz base and 4.5 GHz boost clocks) can match or beat Intel's 9900K in single-thread performance, specifically including gaming. If that's true (and of course we will have to wait for independent benchmarks to verify it), then it follows that the 3900X, with an even higher 4.6 GHz boost, and especially the 3950X, with 4.7 GHz boost, should be able to beat the 9900K outright in ST/gaming even when falling short of the "magic" 5 GHz mark.

Sigh. If the 3900X and 3950X could beat Intel outright, they would have said so in the presentation, instead of settling for tie claim with the 3800X.
All AMD has to focus on now is getting their clockspeeds up and they win. Intel, on the other hand needs a new platform because a die shrink (and getting 10nm to work) isn't going to beat AMD's clock for clock superiority now.

 

[Markfw]

Yea, I have an 8700k, only six cores, but temps and power use in gaming or any other kind of everyday use is a complete non-issue. Overclocked to 4.7 all core, and uses only 60 watts at the max in gaming.

NO, at stock it consumes 86 watts, and overclocked to 4.7 on all cores would be around 110-120.

 

[maddie]

NO, at stock it consumes 86 watts, and overclocked to 4.7 on all cores would be around 110-120.

Well it depends on the GPU, resolution, game, etc. We have no idea as to details. The transistors can be switching states fast and still have very little data throughput.

 

[Markfw]

Well it depends on the GPU, resolution, game, etc. We have no idea as to details. The transistors can be switching states fast and still have very little data throughput.

At stock maybe, but overclocked ? you loose all power savings at that point. 4.7 on all cores is quite a bit. It takes chilled water to get to 4.9

 

[ondma]

NO, at stock it consumes 86 watts, and overclocked to 4.7 on all cores would be around 110-120.

I have measured it. You are free to believe me or not. And it does not require water cooling. Hyper Evo 212 IIRC is the cooler.
In fact, it will run prime 95 at 4.7 all core on that cooler.

Correction: Prime 95 actually runs at 4.4 due to the AVX offset.

 

[moinmoin]

Sigh. If the 3900X and 3950X could beat Intel outright, they would have said so in the presentation, instead of settling for tie claim with the 3800X.

At CES they just showed some 8 core chip that matched 9900k in Cinebench. At Computex they announced the 12 core 3900X as top end chip. At E3 they upped that to the 16 core 3950X chip but mainly compared their chips to Intel one's with mostly matching performance.

By now you should be used to AMD's sandbagging.

 

[rbk123]

At CES they just showed some 8 core chip that matched 9900k in Cinebench. At Computex they announced the 12 core 3900X as top end chip. At E3 they upped that to the 16 core 3950X chip but mainly compared their chips to Intel one's with mostly matching performance.

By now you should be used to AMD's sandbagging.

That's not sandbagging, that's slowly announcing different products.

 

[moinmoin]

That's not sandbagging, that's slowly announcing different products.

You said "if the 3900X and 3950X could beat Intel outright, they would have said so in the presentation" but they didn't show all their cards neither at CES nor at Computex. What makes you think they did that at E3 now?

 

[inf64]

That's not sandbagging, that's slowly announcing different products.

By now you should be able to easily extrapolate performance numbers for the whole lineup in most common desktop workloads. If you still think AMD is behind you suffer from a serious case of cognitive dissonance.

 

[epsilon84]

At stock maybe, but overclocked ? you loose all power savings at that point. 4.7 on all cores is quite a bit. It takes chilled water to get to 4.9

Are you talking about the 9900K or 8700K? Because my 8700K is at 5.0GHz on air, a 212+ no less, which is a $25 HSF. Power consumption is obviously higher than at stock, but nothing crazy, especially during gaming. Actually I've seen CPU power comparisons between the 9900K and 8700K and they are basically identical when overclocked (again, for gaming)

https://www.tomshardware.com/reviews/intel-core-i9-9900k-9th-gen-cpu,5847-11.html

 

[Panino Manino]

I know that Zen 2 has more than enough horsepower for gaming, but that memory latency... with 20-40ns more than Intel Processor Ryzen will never be as good as i9, right?
But after you pass the 60fps mark for minimal, does it even matter?

 

[scannall]

I know that Zen 2 has more than enough horsepower for gaming, but that memory latency... with 20-40ns more than Intel Processor Ryzen will never be as good as i9, right?
But after you pass the 60fps mark for minimal, does it even matter?

We don't know what the latency will be on Ryzen 2 yet. Or how much it will matter with such large caches. Reviews will tell us soon enough. Patience.

 

[Markfw]

Are you talking about the 9900K or 8700K? Because my 8700K is at 5.0GHz on air, a 212+ no less, which is a $25 HSF. Power consumption is obviously higher than at stock, but nothing crazy, especially during gaming. Actually I've seen CPU power comparisons between the 9900K and 8700K and they are basically identical when overclocked (again, for gaming)

https://www.tomshardware.com/reviews/intel-core-i9-9900k-9th-gen-cpu,5847-11.html

Sorry, I missed the word "gaming". Yes these look right in that situation, but possibly a best case scenario type game.

 

[PotatoWithEarsOnSide]

I know that Zen 2 has more than enough horsepower for gaming, but that memory latency... with 20-40ns more than Intel Processor Ryzen will never be as good as i9, right?
But after you pass the 60fps mark for minimal, does it even matter?

If they can go to memory less often due to larger L3 caches then their real world latencies drop. A number is just a number if you simply look at it in isolation.

 

[BigDaveX]

I know that Zen 2 has more than enough horsepower for gaming, but that memory latency... with 20-40ns more than Intel Processor Ryzen will never be as good as i9, right?
But after you pass the 60fps mark for minimal, does it even matter?

Memory latency isn't the be all and end all of things. Yes, the latency could hurt gaming performance and ensure that Intel still have the lead on that front... or it could be what means Zen 2 is "only" equal to or slightly better than the 9900K in games, instead of absolutely murdering it. It's really going to depend how well AMD have improved the other parts of the core, and whether that offsets the latency gap.

 

[exquisitechar]

If they can go to memory less often due to larger L3 caches then their real world latencies drop. A number is just a number if you simply look at it in isolation.

There's a reason why they called it GAMECACHE. Funny name that they came up with, but it's true. The effect of the doubling of L3 cache alone is huge according to one of the slides we got, bigger than I had thought. Even up to 21% performance increase (in CSGO).