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.

 

[dmens]

There's an army of die hard AMD fanatics who'll continue to reveal their intellectual maturity by posting the AVX-512 power draw without context. Typically, this crowd has little else to do than haunt tech forums/news sites comments section for hours per day, to defend their favourite company.

Have to remember they're only doing this as they feel threatened in some way, I'm not sure why...

The i9 with release Microcode/UEFI should take the gaming crown back, something I never thought possible while still on 14nm.

You have no idea why microcode patches are deployed. Engineers do not test something different from what they intend to release, as that would defeat the entire point of testing. Patches are applied as bugs are found after general sale, as @scineram pointed out. They almost always result in performance degradation. I have never seen a bug patch improve performance in my career.

As for BIOS itself, just look the timings and clock ratios. Again, all these electrical qualifications are done well ahead of release.

All this crying about BIOS/ucode this late in the game is just pure ignorance and FUD.

 

[Gideon]

The i9 with release Microcode/UEFI should take the gaming crown back, something I never thought possible while still on 14nm.

This post will age badly, I'm afraid.

 

[Mopetar]

The i9 with release Microcode/UEFI should take the gaming crown back, something I never thought possible while still on 14nm.

I don't really see that happening based on the AT review of the 11700K. There isn't enough clock speed headroom left and the increased latency seems to be a bigger hit to overall gaming performance than the increased IPC is able to compensate for. If you compare the results it looks like this:

Title / CPU	5800X	11700K	10700K	9900KS
Deus Ex MD (600p)	269.8	217.4	211.8	214.5
FF XIV (768p)	315.0	212.1	216.1	235.2
FF XV (720p)	220.3	199.0	179.9	186.8
World of Tanks (768p)	733.8	692.4	707.0	697.7
Borderlands 3 (360p)	214.9	172.6	163.6	175.9
F1 2019 (768p)	384.7	291.6	291.6	316.5
Far Cry 5 (720p)	188.3	178.3	169.8	181.5
Gears Tactics (720p)	389.2	310.9	309.9	306.2
GTA 5 (720p)	180.8	176.2	175.4	176.7
RDR 2 (384p)	190.7	149.8	157.4	167.1
Strange Brigade (720p)	637.2	435.5	463.1	513.3

Moving to 1080p or beyond leads to a GPU bottleneck in most titles in which case there isn't much of a gap between any of the CPUs and there are even a few cases where one of the Intel CPUs will wind up on top, but those are almost always within a margin of error. The 1080p Max quality benchmarks have the CPUs clumped up together in almost every title to the point where there's no difference what you go with, but in most cases the 11700K is still at the bottom.

The AVX results suggest that the i9 probably has some room left if it could draw up to say 275W in non-AVX workloads, but this thin is already pretty close to the edge and the increased clocks really only get it to where the 9900KS is already at. It's impressive that Intel has been able to push 14nm as far as they have, but I don't think an extra ~3% clock speed or any firmware tweaks are going to change what we're seeing now in a substantial way.

 

[lobz]

There's an army of die hard AMD fanatics who'll continue to reveal their intellectual maturity by posting the AVX-512 power draw without context. Typically, this crowd has little else to do than haunt tech forums/news sites comments section for hours per day, to defend their favourite company.

Have to remember they're only doing this as they feel threatened in some way, I'm not sure why...

The i9 with release Microcode/UEFI should take the gaming crown back, something I never thought possible while still on 14nm.

You can join the amd fanatics waiting for intrinsic shaders on Vega for a magical fix to your beloved vendor's product.

 

[Justinus]

You can join the amd fanatics waiting for intrinsic shaders on Vega for a magical fix to your beloved vendor's product.

And they can both join the people still waiting for Maxwell's async compute driver patch.

 

[Hulk]

I don't really see that happening based on the AT review of the 11700K. There isn't enough clock speed headroom left and the increased latency seems to be a bigger hit to overall gaming performance than the increased IPC is able to compensate for. If you compare the results it looks like this:

Title / CPU	5800X	11700K	10700K	9900KS
Deus Ex MD (600p)	269.8	217.4	211.8	214.5
FF XIV (768p)	315.0	212.1	216.1	235.2
FF XV (720p)	220.3	199.0	179.9	186.8
World of Tanks (768p)	733.8	692.4	707.0	697.7
Borderlands 3 (360p)	214.9	172.6	163.6	175.9
F1 2019 (768p)	384.7	291.6	291.6	316.5
Far Cry 5 (720p)	188.3	178.3	169.8	181.5
Gears Tactics (720p)	389.2	310.9	309.9	306.2
GTA 5 (720p)	180.8	176.2	175.4	176.7
RDR 2 (384p)	190.7	149.8	157.4	167.1
Strange Brigade (720p)	637.2	435.5	463.1	513.3

Moving to 1080p or beyond leads to a GPU bottleneck in most titles in which case there isn't much of a gap between any of the CPUs and there are even a few cases where one of the Intel CPUs will wind up on top, but those are almost always within a margin of error. The 1080p Max quality benchmarks have the CPUs clumped up together in almost every title to the point where there's no difference what you go with, but in most cases the 11700K is still at the bottom.

The AVX results suggest that the i9 probably has some room left if it could draw up to say 275W in non-AVX workloads, but this thin is already pretty close to the edge and the increased clocks really only get it to where the 9900KS is already at. It's impressive that Intel has been able to push 14nm as far as they have, but I don't think an extra ~3% clock speed or any firmware tweaks are going to change what we're seeing now in a substantial way.

Geomean added

Title / CPU	5800X	11700K	10700K	9900KS
Deus Ex MD (600p)	269.8​	217.4​	211.8​	214.5​
FF XIV (768p)	315​	212.1​	216.1​	235.2​
FF XV (720p)	220.3​	199​	179.9​	186.8​
World of Tanks (768p)	733.8​	692.4​	707​	697.7​
Borderlands 3 (360p)	214.9​	172.6​	163.6​	175.9​
F1 2019 (768p)	384.7​	291.6​	291.6​	316.5​
Far Cry 5 (720p)	188.3​	178.3​	169.8​	181.5​
Gears Tactics (720p)	389.2​	310.9​	309.9​	306.2​
GTA 5 (720p)	180.8​	176.2​	175.4​	176.7​
RDR 2 (384p)	190.7​	149.8​	157.4​	167.1​
Strange Brigade (720p)	637.2​	435.5​	463.1​	513.3​
Geomean	300.869​	246.4551​	244.5206​	255.913​
	22%​

 

[maddogmcgee]

All this discussion of who gets the gaming crown really counts for nothing. It doesn't matter if one cpu will get 5 fps higher than another at 720p. Like Ian says, people will but whatever they get get their hands on for a 'decent' price.

 

[moonbogg]

Intel should just license Zen3 tech from AMD and add their own IGP and compete that way. I wouldn't trade this mac & cheese with hotdog I'm eating for an 11700k.

 

[tamz_msc]

There's an army of die hard AMD fanatics who'll continue to reveal their intellectual maturity by posting the AVX-512 power draw without context.

Face it: a CPU that reaches 100 degrees and draws nearly 300W while running 25s of a scientific code is not praiseworthy by any standards, especially when it is well-known that Ian's 3DPM is an exception when it comes to scaling from AVX2->AVX512.

 

[Zucker2k]

Face it: a CPU that reaches 100 degrees and draws nearly 300W while running 25s of a scientific code is not praiseworthy by any standards, especially when it is well-known that Ian's 3DPM is an exception when it comes to scaling from AVX2->AVX512.

So it consumed "300W," reached 100 degrees celsius, and still did this:



How many 5800Xs would you need to reproduce that result? How many 5950Xs? Sometimes, you wonder why people make certain remarks out of context. Anandtech has a bad habit of separating power consumption figures from the benchmarks from which they were derived. With the image above, now 292W doesn't seem so bad for the result. Does it? No other desktop chip can produce this result, except RKL, and by a country mile, yet it becomes the point of ridicule for a yet to be released chip, on a secret motherboard with unlimited power characteristics no one is familiar with.
And, of course, you'd have folks ridiculing AVX-512 and its uselessness for the mainstream platform. Well, that makes the power consumption concerns moot then, doesn't it? Unfortunately, the same critics of AVX-512 are mostly the same critics of the PEAK power draw of the chip. Amazing.

 

[Timorous]

So it consumed "300W," reached 100 degrees celsius, and still did this:

View attachment 40724

How many 5800Xs would you need to reproduce that result? How many 5950Xs? Sometimes, you wonder why people make certain remarks out of context. Anandtech has a bad habit of separating power consumption figures from the benchmarks from which they were derived. With the image above, now 292W doesn't seem so bad for the result. Does it? No other desktop chip can produce this result, except RKL, and by a country mile, yet it becomes the point of ridicule for a yet to be released chip, on a secret motherboard with unlimited power characteristics no one is familiar with.
And, of course, you'd have folks ridiculing AVX-512 and its uselessness for the mainstream platform. Well, that makes the power consumption concerns moot then, doesn't it? Unfortunately, the same critics of AVX-512 are mostly the same critics of the PEAK power draw of the chip. Amazing.

So it has 1 niche use case. Question is how well does it do with a mixed workload where only a small % is AVX 512. Would the thermal and power draw in such a case cause a negative impact on the rest of such a test due to throttling?

 

[Zucker2k]

So it has 1 niche use case. Question is how well does it do with a mixed workload where only a small % is AVX 512. Would the thermal and power draw in such a case cause a negative impact on the rest of such a test due to throttling?

If power consumption/temps go down, frequency should ramp up, so no. These chips self-regulate to maximize loads, and that's a good thing.

 

[tamz_msc]

So it consumed "300W," reached 100 degrees celsius, and still did this:

View attachment 40724

How many 5800Xs would you need to reproduce that result? How many 5950Xs? Sometimes, you wonder why people make certain remarks out of context. Anandtech has a bad habit of separating power consumption figures from the benchmarks from which they were derived. With the image above, now 292W doesn't seem so bad for the result. Does it? No other desktop chip can produce this result, except RKL, and by a country mile, yet it becomes the point of ridicule for a yet to be released chip, on a secret motherboard with unlimited power characteristics no one is familiar with.
And, of course, you'd have folks ridiculing AVX-512 and its uselessness for the mainstream platform. Well, that makes the power consumption concerns moot then, doesn't it? Unfortunately, the same critics of AVX-512 are mostly the same critics of the PEAK power draw of the chip. Amazing.

Read my comment in full. Ian's 3DPM workload is NOT representative of typical AVX2->AVX512 scaling. It uses intrinsics and is optimized by an ex-Intel employee who is among the few who knows how to extract the maximum from AVX512. Even Alex Yee, author of y-cruncher, which is hand-tuned and comes in at 500 KLOC, is not able to make it scale beyond 2x and that too when memory bandwidth isn't a bottleneck. x265 uses hand tuned assembly and is not able to extract more than 5-7% extra performance with AVX512.

Bottom line: if not explicitly using the wider vector width of AVX512, the typical uplift is <2x, and not 6x like you have in case of 3DPM.

 

[tamz_msc]

So it has 1 niche use case. Question is how well does it do with a mixed workload where only a small % is AVX 512. Would the thermal and power draw in such a case cause a negative impact on the rest of such a test due to throttling?

I can give hints to a possible answer to this question. I have a TGL-U laptop and use Mathematica on it which uses Intel MKL. Guess what, the Apple M1 in a Macbook Air is faster than the TGL-U even though it runs under Rosetta so it cannot emulate AVX at all.

 

[uzzi38]

I can give hints to a possible answer to this question. I have a TGL-U laptop and use Mathematica on it which uses Intel MKL. Guess what, the Apple M1 in a Macbook Air is faster than the TGL-U even though it runs under Rosetta so it cannot emulate AVX at all.

Hmm, has the native MATLAB port for Apple silicon not launched yet?

 

[tamz_msc]

Hmm, has the native MATLAB port for Apple silicon not launched yet?

Current release works under Rosetta, a native version is under development.

 

[Zucker2k]

Read my comment in full. Ian's 3DPM workload is NOT representative of typical AVX2->AVX512 scaling. It uses intrinsics and is optimized by an ex-Intel employee who is among the few who knows how to extract the maximum from AVX512. Even Alex Yee, author of y-cruncher, which is hand-tuned and comes in at 500 KLOC, is not able to make it scale beyond 2x and that too when memory bandwidth isn't a bottleneck. x265 uses hand tuned assembly and is not able to extract more than 5-7% extra performance with AVX512.

Bottom line: if not explicitly using the wider vector width of AVX512, the typical uplift is <2x, and not 6x like you have in case of 3DPM.

You're not seeing the forest for the tree here. Is there a relationship between that "300W" and that 3DPM result? Hint, POVRay is 225W (Peak).

 

[tamz_msc]

You're not seeing the forest for the tree here. Is there a relationship between that "300W" and that 3DPM result? Hint, POVRay is 225W (Peak).

This is actually more applicable to you. You saw 6x scaling and started running with the claim that despite needing 300W to do so, nothing even comes close to providing similar level of performance, when all I've done is to point it out to you that such scaling is ATYPICAL.

 

[uzzi38]

Current release works under Rosetta, a native version is under development.

Cool, was just wondering because I was talking about it with someone like a month ago now, was partially surprised it wasn't released yet.

 

[Zucker2k]

This is actually more applicable to you. You saw 6x scaling and started running with the claim that despite needing 300W to do so, nothing even comes close to providing similar level of performance, when all I've done is to point it out to you that such scaling is ATYPICAL.

So, let me ask you: What is your point in ridiculing a chip for consuming 300W and reaching 100 degrees Celsius while running atypical code?

My point is that 292W produced a rather comparatively outstanding return. So, what inspires your ridicule?

 

[Zucker2k]

So, you basically took the atypical consumption, and exaggerated it to make a false power consumption argument in order to justify your condemnation. Quite typical. This is exactly my problem with these 'power consumption' figures served out of context and which become fodder (pun intended) for the masses. A section of these same masses will strike out the AES portion of the GB score for Tigerlake chips because the score is, mind if I say, atypical? Yet, they have no problem using figures derived from Intel 'K' chips running AVX based P95 under unlimited power conditions to conclude that Intel chips run hot.

 

[tamz_msc]

So, let me ask you: What is your point in ridiculing a chip for consuming 300W and reaching 100 degrees Celsius while running atypical code?

My point is that 292W produced a rather comparatively outstanding return. So, what inspires your ridicule?

Because if you run typical code it will still draw the same amount of power and run just as hot while giving much less in return. To mitigate this if you restrict power usage then there is barely any uplift at all, as my submission in the 4K handbrake thread shows.

 

[Zucker2k]

Because if you run typical code it will still draw the same amount of power and run just as hot while giving much less in return. To mitigate this if you restrict power usage then there is barely any uplift at all, as my submission in the 4K handbrake thread shows.

Typical code like POVRay? Blender? Handbrake? Fine.

Show me those benchmarks consuming 300W. Anandtech showed POVRay consuming 225W, you can discard that number if you want, but show us an 11700K consuming 300W at stock while running those benchmarks.

 

[moonbogg]

I think the takeaway from all this is that the 10700K was a sick deal at $280. Alderlake will retake the crown in gaming by 40% with 5ghz base clocks. That's what people told me in digital space.

 

[Timorous]

Typical code like POVRay? Blender? Handbrake? Fine.

Show me those benchmarks consuming 300W. Anandtech showed POVRay consuming 225W, you can discard that number if you want, but show us an 11700K consuming 300W at stock while running those benchmarks.

The important part of the AVX 512 test are the accompanying graphs. If you look you see that as soon as the AVX 512 load starts it spikes in power and temp immediately. That shows if you are running a more mixed workload where only part is AVX512 enhanced you are still going to hit these temps and power numbers as soon as it starts to use these 512 features.