Discussion Intel current and future Lakes & Rapids thread

[DrMrLordX]

Cypress Cove. The IMC issues might be why Ice Lake only supports up to 3733.

Oh right I keep getting it mixed up with JAsper Lake.

 

[Hulk]

Intel's 19% number comes from here:

CES 2021 - 1 | Performance Index

I think people (A) ignore the "up to" portion and (B) ignore the fact that Intel was talking about only one specific benchmark and not 19% on everything.

That said, I'm going to wait a few weeks for the actual pricing and final performance benchmarks before making final conclusions. I'm not very hopeful, but maybe there will be a glimmer of good news there.

I was stupid and gullible. Intel's press info fooled me into thinking Rocket Lake would equal or beat Zen 3 clock-for-clock. You are absolutely correct. They tricked me but good! That's why I come to Anandtech for the truth. Anybody know where I can buy a video card?

 

[Hulk]

Just an FYI but Rocket Lake isn't Willow Cove - it's Jasper Cypress Cove, a backported Sunny Cove. With apparent L3 latency problems and some really stupid issues with the IMC.

Yes I do know that. I slipped! Thanks for the correction.

 

[Hulk]

10900 and 10850K are both priced at $320 at MC. I'm actually considering a build.
Which one would you choose?

Forget it. Just realized 10850K has higher all-core turbo clocks.

 

[coercitiv]

10900 and 10850K are both priced at $320 at MC. I'm actually considering a build.

If I were you I would aim for a 10600K or 9700K build, depending on the motherboards available too. They'll offer plenty of performance for now and if you decide to upgrade again in 2022 (ADL, Zen4) for a strong uplift in ST perf... the cost of today's build won't affect you much.

All things considered (pandmic, stock issues etc) Skylake 6c/12t or 8c/8t is a good place to be in 2021.

 

[jpiniero]

If I were you I would aim for a 10600K or 9700K build, depending on the motherboards available too. They'll offer plenty of performance for now and if you decide to upgrade again in 2022 (ADL, Zen4) for a strong uplift in ST perf... the cost of today's build won't affect you much.

All things considered (pandmic, stock issues etc) Skylake 6c/12t or 8c/8t is a good place to be in 2021.

Might be tough to get a 9700K for much less than $320 right now since Intel has stated to discontinue it. The extra 4 cores versus the 10600K might come in handy eventually.

 

[Mopetar]

Might be tough to get a 9700K for much less than $320 right now since Intel has stated to discontinue it. The extra 4 cores versus the 10600K might come in handy eventually.

Did you mean 2 cores? The 9700K is an 8C/8T chip and the 10600K is a 6C/12T chip. I'd probably get the 10600K anyhow. I don't think the 2 physical cores offer that much of an advantage over the 6 additional virtual ones.

 

[Zucker2k]

Did you mean 2 cores? The 9700K is an 8C/8T chip and the 10600K is a 6C/12T chip. I'd probably get the 10600K anyhow. I don't think the 2 physical cores offer that much of an advantage over the 6 additional virtual ones.

I believe he's saying @Hulk original choice of the i9 may come in handy in the future, so that's 4 cores.

 

[coercitiv]

Might be tough to get a 9700K for much less than $320 right now since Intel has stated to discontinue it. The extra 4 cores versus the 10600K might come in handy eventually.

I was looking solely at MC offerings since that's what sprung this conversation. The 10600K and 9700K are available at $190 and $200, and there's some additional savings available through bundles. (which is why I recommended choosing based on available motherboards as well).

The extra cores of the 10850K would come in handy eventually, but my argument was that by the time this happens... a value ADL or Zen4 build will bring more performance overall. Choosing 10600K over 10850K right now would save $130 from the start, and even the base ADL i5 SKU should be faster than 10850K in most tasks.

 

[Zucker2k]

Edit: Please ignore. That was a de-lidded 10900k.

 

[dullard]

10900 and 10850K are both priced at $320 at MC. I'm actually considering a build.
Which one would you choose?

I think you are prematurely rejecting Rocket Lake. One more week for some additional information and 3 more weeks until we get final reviews. Even in Anandtech's preview with potentially incorrect memory settings, the 11700K handly won over the 10700k in most tests. Only in gaming tests did the 10700k win very often and that was more of a neck-and-neck race.

Comparing 11700k to the 10700k:

• GIMP 2.10.18: 11700k faster
• 3D particle movement v2.1: tie without AVX 11700k won by a mile with AVX
• yCruncher: 11700k faster
• ApoA1 Simulation: 11700k faster
• AI Benchmark: 10700k faster
• DigiCortex: 11700k faster
• Dwarf Fortress: 11700k faster or a tie
• Dolphin Render: 11700k faster
• Blender: 11700k faster
• Corona: 11700k faster
• Crysis Render: 11700k faster
• POV-Ray: 11700k faster
• V-Ray: 11700k faster
• Cenebench R20: 11700k faster
• Handbrake: 11700k faster
• 7-Zip: 11700k faster
• AES Encoding: 11700k faster
• WinRAR: 11700k faster almost a tie though
• Kraken: 11700k faster
• Google Octane: 11700k faster
• Speedometer: 11700k faster
• CineBench R15: 11700k faster
• 3DPM: 10700k faster
• SPEC: 11700k faster
• Gaming: neck and neck, depending on the game or even the resolution within the game the 10700k or 11700k may be faster

If the 11700k is priced right, I might get it. If not, I'll strongly consider the now discounted 10700k. I'll consider AMD too if I can find it in an OEM.

 

[jpiniero]

and even the base ADL i5 SKU should be faster than 10850K in most tasks.

I dunno. I am thinking the base Alder i5 will be 6+0 (or possibly even 4+8 like the S BGA leak). But the cheapest i5 tends to be very stingy on clocks.

 

[Hulk]

I think you are prematurely rejecting Rocket Lake. One more week for some additional information and 3 more weeks until we get final reviews. Even in Anandtech's preview with potentially incorrect memory settings, the 11700K handly won over the 10700k in most tests. Only in gaming tests did the 10700k win very often and that was more of a neck-and-neck race.

Comparing 11700k to the 10700k:

• GIMP 2.10.18: 11700k faster
• 3D particle movement v2.1: tie without AVX 11700k won by a mile with AVX
• yCruncher: 11700k faster
• ApoA1 Simulation: 11700k faster
• AI Benchmark: 10700k faster
• DigiCortex: 11700k faster
• Dwarf Fortress: 11700k faster or a tie
• Dolphin Render: 11700k faster
• Blender: 11700k faster
• Corona: 11700k faster
• Crysis Render: 11700k faster
• POV-Ray: 11700k faster
• V-Ray: 11700k faster
• Cenebench R20: 11700k faster
• Handbrake: 11700k faster
• 7-Zip: 11700k faster
• AES Encoding: 11700k faster
• WinRAR: 11700k faster almost a tie though
• Kraken: 11700k faster
• Google Octane: 11700k faster
• Speedometer: 11700k faster
• CineBench R15: 11700k faster
• 3DPM: 10700k faster
• SPEC: 11700k faster
• Gaming: neck and neck, depending on the game or even the resolution within the game the 10700k or 11700k may be faster

If the 11700k is priced right, I might get it. If not, I'll strongly consider the now discounted 10700k. I'll consider AMD too if I can find it in an OEM.

That is a fair observation. Price-wise it seems the competition for the 11700K is the 10850K, not the 10700K. And those 2 extra cores do tend to level the MT playing field. I wonder if the 11700K will be priced at the current 10850K amount of $320 at MC?

 

[Zucker2k]

Read last line of footnotes


Cherry picked games?

https://twitter.com/x/status/1369442891198763011

 

[Zucker2k]

What this means is that the 11700k runs IMC in 1:2 mode at DDR4 3200MHz by default. Only the 11900k runs IMC in DDR4 3200MHz in 1:1 mode by default. Artificial segmentation? Binning? Who knows?

 

[jpiniero]

Artificial segmentation? Binning? Who knows?

The base clock on the 11900K is lower than the 10900K so both? A little strange that Ian didn't mention it in the review but perhaps that is NDA'd?

 

[JoeRambo]

What this means is that the 11700k runs IMC in 1:2 mode at DDR4 3200MHz by default. Only the 11900k runs IMC in DDR4 3200MHz in 1:1 mode by default. Artificial segmentation? Binning? Who knows?

Yeah, what that also means, is that Anandtech wasted a ton of effort on testing wrong "K" CPU in a mode that no enthusiast will run it on.

Still, 11700K is bad CPU, 1:1 mode goes up to 3733 currently, that is even worse scaling than on AMD chips.

 

[Zucker2k]

Yeah, what that also means, is that Anandtech wasted a ton of effort on testing wrong "K" CPU in a mode that no enthusiast will run it on.

Still, 11700K is bad CPU, 1:1 mode goes up to 3733 currently, that is even worse scaling than on AMD chips.

Are you saying a site known for running RAM at manufacturer 'supported' speeds wasted time? That's their bread and butter. Nothing's changed. LOL

 

[JoeRambo]

Are you saying a site known for running RAM at manufacturer 'supported' speeds wasted time? That's their bread and butter. Nothing's changed. LOL

Yeah, but this generation is gonna be special (or specially retarded?). 11900K is gonna have quite much larger gap on "stock" settings than usual. The other chips are probably better off running 2933 instead of 3200 speeds also. 20-25% latency penalty is no joke.

What is sad, is that Anandtech probably won't retest 11900K and all those wrong IPC numbers will become gospel. So the world won't know what IceLake + desktop class MC was capable. For enthusiasts if there is a hard limit on 3733, that means that 10900K will still reign in gaming ( behind AMD ofc ).

But all of that is well deserved by Intel, they have shot themselves in the foot and i hope market will continue to punish them.

 

[tamz_msc]

Well this information is presumably under NDA so it wasn't mentioned in the review. I don't find anything wrong with testing the 11700K in Gear 2 mode as that's its default configuration, and Anandtech has a policy of not testing configurations that are technically 'overclocking', and I think Gear 2 -> Gear 1 mode on the 11700K falls under overclocking. So under default configuration the numbers Anandtech got are completely believable, and Ian is getting flak on Twitter for no reason.

 

[Zucker2k]

Yeah, but this generation is gonna be special (or specially retarded?). 11900K is gonna have quite much larger gap on "stock" settings than usual. The other chips are probably better off running 2933 instead of 3200 speeds also. 20-25% latency penalty is no joke.

What is sad, is that Anandtech probably won't retest 11900K and all those wrong IPC numbers will become gospel. So the world won't know what IceLake + desktop class MC was capable. For enthusiasts if there is a hard limit on 3733, that means that 10900K will still reign in gaming ( behind AMD ofc ).

But all of that is well deserved by Intel, they have shot themselves in the foot and i hope market will continue to punish them.

Yeah. It seems there's also talk, at least for the 11700k, that the uncore seems to be limited to 4GHz. If true, that's absolutely retarded. I think what most of us said about there being no difference between the 11700k and 11900k reached Intel, and they decided to draw a clear line between the two chips just so they can charge way more for the halo 11900k. Just retarded.

 

[JoeRambo]

I think what most of us said about there being no difference between the 11700k and 11900k reached Intel, and they decided to draw a clear line between the two chips just so they can charge way more for the halo 11900k. Just retarded.

No doubt that this is what happened. 14nm can clock uncore 4.5+ghz and memory well above 4Ghz in Comet Lake. It is pure segmentation decision.

But from technical and market situation it is completely retarded move. When your competitor has twice of L3 cache, you don't add 20-25% of extra memory latency by marketing decision.

I don't find anything wrong with testing the 11700K in Gear 2 mode as that's its default configuration, and Anandtech has a policy of not testing configurations that are technically 'overclocking', and I think Gear 2 -> Gear 1 mode on the 11700K falls under overclocking.

Yeah, but for core IPC comparisons you want best SKU. It is like testing Skylake core with Pentium SKU with 3MB of L3 and DDR4 2133C24 and claiming it has regressed everywhere.
So all that work is useless for IPC comparisons. And since Anandtech was already useless for enthusiasts due to running stock configurations, that means work wasted.

 

[Zucker2k]

Maybe Intel gave AT an 11900k for their review. Time will tell.

 

[tamz_msc]

Yeah, but for core IPC comparisons you want best SKU. It is like testing Skylake core with Pentium SKU with 3MB of L3 and DDR4 2133C24 and claiming it has regressed everywhere.
So all that work is useless for IPC comparisons. And since Anandtech was already useless for enthusiasts due to running stock configurations, that means work wasted.

I agree that it makes the IPC comparisons somewhat meaningless, and testing the 11900K with 3200 MHz memory in Gear 1 mode will result in some gains, though I doubt that the conclusion that IPC gains were more in FP than INT would change.

 

[JoeRambo]

Given how bad Gear 2 is in DDR4 4500 speed region, one has to wonder if penalty grows or diminishes @ DDR4 3200 speed. From people running things like Tomb Raider benchmark it seems to completely butcher performance.

This Gear 2 thing at such low speeds is moronic decision, that is a fact.