AES-NI Performance Analyzed; Limited To 32nm Core i5 CPUs

07:10 - Tuesday 2 February 2010 by Patrick Schmid and Achim Roos

Security is an important topic these days. However, it's typically only recognized as important by professionals. If security were to suddenly turn into a mainstream selling point, though, then perhaps it'd make more sense for companies like Intel to promote it.

The Advanced Encryption Standard (AES) has already been adopted by the United States government—including the NSA—along with many other institutions. Intel’s 32nm Clarkdale-based CPUs (only the Core i5-600-series, so far) now promise significant performance benefits for AES encryption and decryption via new instructions. Today we're looking at the real-world benefits of Intel's AES-NI functionality, comparing a dual-core Core i5-661 with AES New Instructions (AES-NI) to a quad-core Core i7-870, which lacks the new encryption acceleration capability.

Encryption is used much more intensively than you might suspect. Consider Internet sites that hold you sensitive personal information, or utilize sensitive data for transactions. They all use protocols like Transport Layer Security (TLS) or Secure Sockets Layer (SSL). VoIP, instant messaging, and email may also be protected with these protocols. Virtual Private Networks (VPNs) and electronic payments are other popular encryption applications.

However, TLS and SSL are cryptographic protocols for secure communication, while AES is a general-purpose encryption standard. It can be used to encrypt individual files, data containers, archive files, entire drives (including thumb drives), and even multi-drive volumes. AES can be implemented in software, and there are products based on hardware acceleration as well, since encryption/decryption represent a rather significant workload. Solutions like TrueCrypt or Microsoft’s BitLocker, which is part of Windows Vista and Windows 7 Ultimate, are capable of encrypting entire partitions on the fly.

Whether or not you have "sensitive" data on your system naturally depends on your definition of sensitive, as well as your personal comfort level. In addition, security always depends on the right strategy and diligence in handling important information. Vital bits of data, like your social security number or credit card data (including the ol' expiration date and security code), should never be left on a system in unencrypted plain text.

One thing is certain: it makes sense to be overly-cautious than the other way around, especially if the amount of effort required to protect yourself is minimal. Intel’s approach of adding AES acceleration to its hardware doesn't automatically mean AES-enabled apps suddenly see a speed-up (optimizations are still required). However, it is a solid first step in addressing the most popular encryption standard in hardware, and it will be, and it will be a notable value-add in that company's future 32nm desktop processors, though you might be surprised to learn that the feature isn't included in all of the Clarkdale-based chips launched last month.

Does AES-NI provide a significant performance gain in typical encryption scenarios, or is this mainly a marketing push? Let's put it to the test.

Comments
Read the comments on the forums
mi1ez 02/02/2010 10:03
Hide
-1+

Good grief. About 3 mistakes on the first page!

mi1ez 02/02/2010 10:22
Hide
-0+

I may be being a bit skeptical, butputting on the highest i5 chips that include a GPU? Does this not sound like a money spinner?

aje21 02/02/2010 14:13
Hide
-1+

Nice to see that Intel have finally caught up with Via...
Shame we can't see any benchmarks to compare the performance of the AES engines.

wifiwolf 03/02/2010 19:14
Hide
-1+

I'd think it's not all good things coming from this ability.
Malware programmers can benefit from it as it should accelerate decrypting passwords and alike.

psiboy 04/02/2010 10:15
Hide
-0+

Gee lets compare a quad core to a dual core? WTF! No balance or objectivity here at all! This got past the editors how?

Anonymous 04/02/2010 15:39
Hide
-0+

How does the CPU knows about to use the ASE instructions? Is there a special library comming with the Benchmarks?

Anonymous 04/02/2010 23:04
Hide
-0+

Please do some Linux tests! IMHO the support for the new AES-NI has been in the kernel for quite some time (done by Intel long before those CPUs even came to the market!) and dm-crypt is a very nice way to test REAL WORLD speeds.

Anonymous 05/02/2010 17:48
Hide
--2+

Imagine new i5 without AES-NI! Why would you buy it anyway when it is always inferior compared to i7? Well - there comes Intel marketing guys and say: We will put AES-NI just in i5 (in the beginning) hoping that the product will attract some buyers. If they put now AES-NI in i7, i5 will be doomed processor.

roots 03/03/2010 02:21
Hide
-2+

This would be very nice in a firewall. VPN thoughput on one of these CPU's would be awsome.

My Guess is that where this CPU will end up. The next gen of Cisco ASA series and the like.

Anonymous 02/06/2010 12:55
Hide
-0+

Still kinda sucks... as the AES-NI is only for the 1156 socket. Unless I feel like forking out 1K for the 980x (1366)

Comments are closed on this page.
Ads

Best offers

Newsletters


  • Ask your question about IT issues
  • Post