Intel Core i7-11700K "Rocket Lake" CPU Outperforms AMD Ryzen 9 5950X in Single-Core Tests

[RandallFlagg]

No one said anything about AES when Zen 3 was far faster than Comet Lake. Seems like a tremendous level of hypocrisy.

Look at your URL bar, every place it says https and not http (with attendant insecure site browser warnings), you are using AES. Encrypted files, filesystems, and so on use AES. 5% weight seems like a reasonable number for a client system.

Ivy Bridge and Haswell made huge improvements on AES. Haswell was about 4X faster at AES vs Sandy Bridge and about 60% faster than Ivy. The difference in that case is palpable while using a browser on modern AES encrypted websites, especially with multiple encrypted tabs/connections. I will admit there is not as much of a user feel difference after that, beyond a point a user doesn't feel the difference, but AES is most definitely a thing that affects the user experience - and you can feel it if you swap between SB and Haswell boxes.

 

[dragontamer5788]

No one said anything about AES when Zen 3 was far faster than Comet Lake. Seems like a tremendous level of hypocrisy.

Not only that, but AMD pushed Zen3 into AES-XTS first.

AMD Zen 1 started the trend, pushing 2-AES pipelines per clock tick, doubling its AES-performance over Intel. Intel pushed back with their own 2x AES pipeline design, then AMD Zen3 allowed a 2nd doubling, allowing their 256-bit vectors to perform 2xAES per pipeline per clocktick (or now 4x 128-bit AES calculations per clocktick). Intel is pushing back by making AVX512, or 4x AES on 512-bit registers to be pushed.

If anything, these CPU-manufacturers prove how important AES is to today's workloads, be it server or client. Apple has extremely fast AES units, AMD has extremely fast AES units, Intel has extremely fast AES units. Everyone's optimizing AES here. This isn't an Intel or Geekbench thing: literally the entire industry is pushing faster-and-faster AES performance.

Youtube? Delivered by HTTPS. Netflix? Each 30-minute to 60-minute episode at 1080p is 4GB to 10GB. All of which has to be AES decrypted as you're watching. Everything you're doing on the internet is HTTPS encrypted and goes through that AES core before any further processing (decoding, or rendering) can be done. Its a non-trivial bottleneck in all web-based applications.

Seeing 5% weight to it is fine.

 

[Vya Domus]

If anything, these CPU-manufacturers prove how important AES is to today's workloads, be it server or client.

Yes it has it’s importance but lets not kid ourselves too much, stuff like improving AES performance are easy wins to make a CPU look better in synthetic test.

 

[efikkan]

CLang is probably not so common, but probably is representative of Javascript. SQLite is in a bunch of random stuff, so its probably a good benchmark today.

The primary problem with Geekbench is that it's synthetic, it doesn't translate into any real world performance. As with all benchmarking, the most important thing is what the benchmark is actually benchmarking. Geekbench isn't useful at all for any consumers, perhaps somewhat useful for some scientific work, but it's really benchmarking the implementation not the hardware here. People should never use Geekbench to extrapolate generic real world performance, no matter how "balanced" the workload may be. Real applications should always be the way we measure CPUs.

BTW, Clang is the C/C++/ObjC frontend for LLVM, a very common compiler, nothing to do with JavaScript.

 

[dragontamer5788]

The primary problem with Geekbench is that it's synthetic, it doesn't translate into any real world performance. As with all benchmarking, the most important thing is what the benchmark is actually benchmarking. Geekbench isn't useful at all for any consumers, perhaps somewhat useful for some scientific work, but it's really benchmarking the implementation not the hardware here. People should never use Geekbench to extrapolate generic real world performance, no matter how "balanced" the workload may be. Real applications should always be the way we measure CPUs.

BTW, Clang is the C/C++/ObjC frontend for LLVM, a very common compiler, nothing to do with JavaScript.

Javascript is a C-like language, composed of for-statements, and {} brackets. Parsing Javascript code probably is very similar to Clang compiles.

Especially when you consider that modern Javascript is JIT-compiled into assembly language these days.

 

[AleXXX666]

I wonder how many new security flaws this generation will have.

Wait, are these jokes still actual? Come on, stop living in 2019...

 

[pexxie]

Amazing what a little extra CPU cache can do.

Per-Core L1-L2 Caches:
10700K: 32KB-32KB-256KB
11700K: 32KB-48KB-512KB
5950X: 32KB-32KB-512KB

Although Intel now offering a bigger L1 Data Cache (48KB) than AMD's flagship (32KB) looks like a bit of desperation, and/or the inability to make the gains elsewhere.

 

[Tom Sunday]

2021 should be a good year for CPU sales from both camps!

Your are 100% correct. Many choices and with Intel alone pushing their 11th and 12th generations CPU's on the market and all within months of each other. Motherboard producers will also be having a field-day or 'double-dip' sales opportunities. My question reading your message is: "Will Intel buyers purchase Rocket Lake or wait a few months for Alder Lake with its advertised PCIE 5.0 and DDR5?" Besides Intel's brand new and physically much larger (rectangular) and better performing Z590 CPU. The old addage prevails many must have the 'latest and the greatest no matter what.

 

[RandallFlagg]

Thinking about this a bit more, and looking at how I use my corporate laptop, AES encryption is way more important in that sector - more than geekbench is weighting it.

Typical workload for me would include VPN to corporate network, 4-6 browser tabs, 1-4 server RDP connections, one or more SSMS (SQL Server) connections, connection to SVN code repositories, and one or more SSH connections. All of these are encrypted, and while not all are in use at once (maybe one or two) many like RDP are exchanging data to do things like screen updates.

A typical user may have even more connections, albeit different types. Shared file systems - and all the different apps that can use that (many, many), web tabs, Oracle application UIs, SSH to unix applications, and so on. Most if not all are secured connections, hence encrypted.

So yeah, AES performance is critical.

There's a relatively new benchmark out for the 11700K. This one is on Geekbench 4.4

Gigabyte Technology Co., Ltd. Z490 AORUS MASTER - Geekbench Browser

Benchmark results for a Gigabyte Technology Co., Ltd. Z490 AORUS MASTER with a 11th Gen Intel(R) Core(TM) i7-11700K @ 3.60GHz processor.

browser.geekbench.com

 

[Tom Sunday]

Who really cares about Geekbench, Core Workloads, HTTP AES, Security or Realworld's! For the man on the street it's mostly to have the latest and the greatest, not talking things to death and taking prisoners later. If there are problems after the purchase most are sure there will be ready-made manufacturer fixes being made available. Its like the masses out there do not care about overclocking either and potentially wringing out a mere percentage point of more performance. For them its too complicated, nerve-racking, uncertain and takes too much time. As such the masses want to keep things simple, uncomplicated and turning on their PC and it simply works.

 

[AleXXX666]

Your are 100% correct. Many choices and with Intel alone pushing their 11th and 12th generations CPU's on the market and all within months of each other. Motherboard producers will also be having a field-day or 'double-dip' sales opportunities. My question reading your message is: "Will Intel buyers purchase Rocket Lake or wait a few months for Alder Lake with its advertised PCIE 5.0 and DDR5?" Besides Intel's brand new and physically much larger (rectangular) and better performing Z590 CPU. The old addage prevails many must have the 'latest and the greatest no matter what.

Lol where you see DDR5 in sale?
Or it will be, 1.5x the price DDR4 - no thank you, lmfao

 

[efikkan]

My question reading your message is: "Will Intel buyers purchase Rocket Lake or wait a few months for Alder Lake with its advertised PCIE 5.0 and DDR5?"

"A few months"?
At best, we're talking about the very end of 2021, much more likely mid-2022.
People should generally not wait unless the next big jump is imminent.
PCIE 5 and DDR5 will be much more relevant for HEDT and servers than the mainstream.

So yeah, AES performance is critical.

AES is important, yet do you know if all those applications are rewritten and recompiled to use these new AES instructions? (hint: they're not)
So unless someone rewrite that software and recompile it, those Geekbench scores are not that useful.

 

[Tom Sunday]

At best, we're talking about the very end of 2021, much more likely mid-2022.

Our friends at INTEL are having their "Earnings Call" on Thursday, January 28. Mr. Huang of NVIDIA will be on call by Thursday, February 11. Wall Street and investors want to see 'cash-flow' and especially with AMD for now having turned on the heat. Besides what progress will Huang be reporting on his US$40 billion purchase? Timing is everything.

 

[RandallFlagg]

AES is important, yet do you know if all those applications are rewritten and recompiled to use these new AES instructions? (hint: they're not)
So unless someone rewrite that software and recompile it, those Geekbench scores are not that useful.

You don't write your own encryption algorithms unless you work for a security company selling things like a 3rd party vpn client. You use the built-in libraries and functions of the OS if you need to do a custom secure transport, using things like TLS library calls. The TLS library is where the algorithm would be.

Most of the time, you don't even do that. An example of something standard would be, where I deploy a web application that is using HTTPS (handled by IIS) and possibly a SSO service like GetAccess. The web developer does not write any encryption algorithms for that, it's part of IIS.

VAES512 has been around since 2017. It is not the same thing as AVX-512 which you may be getting it confused with. It's already used on most encryption libraries, and by most operating systems, and major 3rd party encryption / security providers.

 

[AteXIleR]

It is rather impressive. I could not yet go over 1600 even at 5.6 with my 9900K.
The single core performance must have substantially been improved then.