Discussion i7-11700K preliminary results

[GunsMadeAmericaFree]

early benchmarks

more benchmarks

Just thought I would provide these so folks could draw their own conclusions. Evidently the processors have been getting sold for a few days now in Germany.

 

[Kocicak]

If you would like to sell your best 8 core CPU for price of the competing 12 core, you could make this 8 core stand out by sandbagging the lower model in the line. Intel may have intentionally castrated 11700K to make the reason why to pay more for the top model more obvious.

 

[tamz_msc]

Hardwareluxx has posted their (p)review.
https://www.hardwareluxx.de/index.p...-lake-s-als-core-i7-11700k-im-vorab-test.html

 

[coercitiv]

Andreas Schilling from Hardwareluxx just got a new BIOS revision (BIOS 0603 dated March 4th) and will test to see if performance changed.

 

[Zucker2k]

Andreas Schilling from Hardwareluxx just got a new BIOS revision (BIOS 0603 dated March 4th) and will test to see if performance changed.

Latencies are quite better, and the IMC is running 1:1
Write got a big boost; Copy as well



Edit: Wait! It's still in 1:2 mode.

 

[Hulk]

I just read that review with the new BIOS update. Performance improved a bit, but the 11700K is is decidedly beaten by the 5800X.

 

[Timorous]

I just read that review with the new BIOS update. Performance improved a bit, but the 11700K is is decidedly beaten by the 5800X.

That puts the Y-cruncher result in line with Ian's result. 9.1s for 250m vs 19.2 for 500m. The HardwareLuxx old result was 26.1s for 500m for the 11700K .

This makes me think the BIOS Ian used had some of the fixes others talk about.

EDIT: If you go through the review you can see the other gains. 0.5% improvement in gaming over their old review.

 

[TheELF]

There is no way in heaven or hell that mindfactory didn't make a good amount of money from each CPU they sold early, the 11700k will be in the same price range as the 10700k so about $380 while the 5800x sells for about $450.
The 11700k will be slower than the 5800x in many things and that's ok because it will also be at least $50-70 cheaper, scalpers not withstanding.

The real problem will be with the 11900k which won't have more cores and not much higher clocks but potentially be the same price as the 5800x if not a bit more.

 

[Det0x]

That puts the Y-cruncher result in line with Ian's result. 9.1s for 250m vs 19.2 for 500m. The HardwareLuxx old result was 26.1s for 500m for the 11700K .

This makes me think the BIOS Ian used had some of the fixes others talk about.

Apparently the new BIOS changed the boost behavior slightly and, depending on the boost status, also increased the multiplier by one step for a different number of cores. Accordingly, we see slightly improved performance in some multi-threaded tests.

Above all, however, the ASUS ROG Maximus XIII Hero does not seem to have set the AVX clock correctly beforehand, because we see significant improvements in benchmarks that use the AVX512. This applies above all to the Y-cruncher, which we want to emphasize again at this point.

The Y-Cruncher use the current SSE / AVX2 / AVX512 instruction sets. If these come into play, the Core i7-11700K makes a significant leap forward with the new BIOS. The Corona renderer, which uses Intel's Embree engine and thus also partly uses AVX512 instruction sets, is a plus, just like DigiCortex. We can also record a slight increase in performance in the games, but the jump is not that big here.

For a comparison, my 5950x used 10.5 seconds for 500m in Y-cruncher which is a AVX512 workload (!)

 

[coercitiv]

This makes me think the BIOS Ian used had some of the fixes others talk about.

Ian was careful not to divulge the motherboard brand and BIOS revision used in the review. Some accused him of obfuscating to hide poor testing methodology, but the more likely reason was protecting his source for the latest available BIOS at the time of the review.

 

[Timorous]

Ian was careful not to divulge the motherboard brand and BIOS revision used in the review. Some accused him of obfuscating to hide poor testing methodology, but the more likely reason was protecting his source for the latest available BIOS at the time of the review.

That is looking to be the case with a 2nd set of benches on the 0603 BIOS largely aligning with what Ian got.

 

[PhonakV30]

Here :

Intel Imitates AMD's Memory Overclocking Approach With Rocket Lake CPUs

Team Blue may have its own Infinity Fabric-like feature.

www.tomshardware.com

now read comments :

Overclocking 11700k/11900k results, bins and discussion

www.overclock.net

One said :

I even got bios today from Shamino, and I tried to overclock memory and cpu/ring "to the max"
It's impossible to fix Rocketlake. Well it's pretty fast with 3733c14 tweaked (why not faster than 3733mhz *** !) in 1:1 mode with 5.1 ghz all core, but it's not even close to "max" overclocked 10900k in games with 4700c17 memory.
L1 cache is about 20% slower ! than 10900k LOL ***?
I can run 4533c14 1:2 mode, but still latency is pretty bad and bandwidth is average. Maybe I'm not a pro Rocketlake overclocker, but it looks like it's unfixable.

If you compare anadtech/Hardwareluxx , Not much different in game performance.Those buyer said they can't do OC above 3733 with 1:1 mode then It's officially DOA for gaming

 

[scineram]

If you compare anadtech/Hardwareluxx , Not much different in game performance.Those buyer said they can't do OC above 3733 with 1:1 mode then It's officially DOA for gaming

That conforms with what Ian measured: 1 more cycle latency for L1 and L2.

 

[Mopetar]

I'll read one or two to see if the numbers line up with what Anandtech just published with their 11700k. It's not likely I'll go through a half-dozen of them though, unless I feel the need to get into some kind of pointless nerd fight over why Rocket Lake-S is actually a bad product.

The question is whether having this early review would change your normal reading behavior. Do you normally go through half a dozen, but now you won't because of the AT article?

For me it changes nothing and I don't think it matters for anyone that's a fan of a particular site or prefers their reviews. I'd still read an AT review of the 3080 even though it would be months late. Hell, I'd even read a 960 review if they published one.

Guessing whoever runs Pat's twitter account at Intel's social media team isn't aware of the review.

Or Pat's demonstrating some maturity and showing leadership by not burning bridges over something that he realizes may be Intel's fault. Throwing a Twitter tantrum won't fix any of the problems that make Intel look bad that have come from this.

If you would like to sell your best 8 core CPU for price of the competing 12 core, you could make this 8 core stand out by sandbagging the lower model in the line. Intel may have intentionally castrated 11700K to make the reason why to pay more for the top model more obvious.

The number of those top chips will be small. It isn't worth ruining a model that will sell in far greater numbers just to make another that will only sell a fraction as many units look more appealing.

Maybe they could get away with that when they were on top, but with AMD being competitive or in some cases having the better product it makes even less sense.

 

[DrMrLordX]

The question is whether having this early review would change your normal reading behavior. Do you normally go through half a dozen, but now you won't because of the AT article?

If it were a major launch that dominated the semiconductor enthusiast scene for weeks/months then I would go through half a dozen such articles or more. If it were a dud of a launch, maybe one or two. The difference here is that by publishing early, the ONE article I for-sure would read (actually already read) on Rocket Lake-S is Dr. Cutress' early one.

 

[Kocicak]

The number of those top chips will be small. It isn't worth ruining a model that will sell in far greater numbers just to make another that will only sell a fraction as many units look more appealing.

And what if 11900K will be competitive (ignoring power consumption) and Intel with their superbly optimised 14nm process can make substantial number of them now? Then they would want it to stand out.

 

[Reliant]

Is this memory ration 1:2/1:1 a RKL only thing? I see the screenshots of "gear1 and gear2" for MSI, but I have a ASUS Z590 and don't see it. I am running 10th gen but wanted to see what's up.

Also, in HWIFO64 or CPU-Z, how can you tell if you are running 1:1?

 

[Panino Manino]

Cap is still on twitter "claiming victory" with the improved bios numbers failing to realized that the actual performance still falls in line with Anandtech's review. And there are people jumping together to talk about others being "blinded".

Well, no matter which one is better anyway, what matters is that both are competing hard, meaning that AMD will have to give everything they can to remain back in the game. Zen 4 probably will not be weak.

 

[lobz]

The chip was sold without the approval of Intel. The whole world knew about this so why would a professional reviewer exploit this loophole because of the technicality that arose out of an illegality, and an NDA signee at that?

I wonder what were your comments on that French tech magazine releasing a review of a Zen ES.

 

[dmens]

The most amusing thing out of all of this is the fact that none of these people have any clue what the BIOS actually does. It is like pro athletes with their superstitions about their equipment or shaving their beards, except at least those superstitions could actually affect their mindset. Just look at the control registers. The end.

 

[Mopetar]

And what if 11900K will be competitive (ignoring power consumption) and Intel with their superbly optimised 14nm process can make substantial number of them now? Then they would want it to stand out.

It's not the raw number, but the relative amount. Unless the bin in a very non-traditional manner, there will be more 11700K parts than 11900K parts. Say for example that the top chip represents the 10% best silicon. If you produce twice as many wafers, you still have only half as many chips as those that get put in the bin representing your next 20% best silicon. If your process matures and gets better over time, it just means your 10% best silicon performs better, not that there's more of it.

Look at the statistics (https://siliconlottery.com/pages/statistics) that silicon lottery has provided as an example of why what you're proposing won't work. Start with Skylake and work your way through the various Intel chips released on their 14 nm process over the years. It did improve and the fastest parts always got just a little bit faster, but the absolute best performance possible could only be achieved by a tiny number of CPUs from any already small number of CPUs that even qualified to be sold as the top-end part.

Even if we lived in a universe that didn't produce those results and instead gave a uniform distribution in terms of yield with respect to performance capability, companies still wouldn't do what you're suggesting. In fact we live in a universe where companies routinely artificially limit the performance capabilities of chips that could be sold as a higher bin part at a higher price because there's a limit to what the market will buy. Even if the 11900K is only 2% better than the 11700K, there are a certain number of people that will spend an extra $200 just to get that last little bit.

Again, look at Silicon Lottery where getting a 5.1 GHz 10900K or 10900KF instead of a 5.0 GHz one costs an extra $200. That 2% difference in clock speed will be purchased by someone. However, a glance back at the binning statistics shows that only 1% of 10900K and 2% of 10900KF processors can hit that 5.1 GHz mark. Rather than sandbagging with the less capable (but vastly more numerous) chips to make the very limited top-end chip look better, you just charge more for the very limited top-end chip. The more rare it is, the more you can charge for it.

Also, the idea of your approach fails for other reasons because using the same logic it makes sense for Intel to sandbag the lower performing parts even more to make the 10700K look more appealing. Of course since the 10700K would have been artificially limited, it means everything below that has to be even more limited to account for that. Why pointlessly cripple 90% of your CPUs just to make the top 10% look more appealing. You'd have to be able to sell them for a lot more than Intel can charge to make up for the difference. Never mind that sandbagging at all when you're behind your competition is flat out stupid.

 

[zir_blazer]

And what if 11900K will be competitive (ignoring power consumption) and Intel with their superbly optimised 14nm process can make substantial number of them now? Then they would want it to stand out.

Performance competitive consuming two or three times the power, even if the Processot price is the same, is a heavy lose. You're going to spend significantly more in Motherboard as you need far beefier VRMs to supply those 200-300W, plus a bigger sized PSU. And a nuclear cooling tower sized Heatsink, with a Fan that should be significantly more noisy that if you're cooling around 65-105W.
Just because it is a desktop, doesn't means than the sky is the limit.

 

[IEC]

Spoiler alert: Ian's review already had the BIOS fixes.

 

[Mopetar]

Cap is still on twitter "claiming victory" with the improved bios numbers failing to realized that the actual performance still falls in line with Anandtech's review. And there are people jumping together to talk about others being "blinded".

Honestly it just makes if funnier when the actual reviews come out and the results haven't changed much out of the usual variance between different reviews and methodology.

It's a lot like all of the AMD diehards that were holding out hope for Vega when all signs were pointing towards disappointment, but this should at least be amusing since the AMD diehards were pretty used to disappointment by the time Vega actually did launch and even though AMD was able to come out on top the CPU market in a way that hadn't been seen in decades with Zen 3, Intel faithful were always able to hold out some hope that Intel's next CPU would regain supremacy.

I suppose that at least they'll still have UserBenchmark.

 

[Racan]

A Reddit user tested his 11700K vs 10900k

Borderlands 3:

1080p Very Low DX11
10900K 5Ghz all core 4000Mhz CL17 - 194 average
11700K 5Ghz all core 3600Mhz in 1:1 mode with CL15 - 209 average
Squad -- Tested on Firing Range : This was tested by sitting on the same AA gun and looking downrange at same spot. Incredible gain!! Will have to do more UE4 testing.
10900K 5Ghz - 158 fps
11700K 5Ghz - 199 fps
Far Cry 5 Built In Benchmark:
10900K 5Ghz - 197 average, 262 max, 157 min
11700K 5Ghz - 210 average, 275 max, 162 min

https://www.reddit.com/r/intel/comments/m2bvqy/_/gqiv18m

 

[AdamK47]

Yep. Results all over the place.