Author: William Henning
Editor: Howard Ha
Publish Date: Wednesday, February 6th, 2008
Originally Published on Neoseeker (http://www.neoseeker.com)
Article Link: http://www.neoseeker.com/Articles/Hardware/Reviews/phenom_9600_black_edition/
Copyright Neo Era Media, Inc. - please do not redistribute or use for commercial purposes.
AMD was kind enough to send us a Phenom 9600 Black Edition processor - the 2.3GHz Phenom with an unlocked multiplier that they are targeting computer enthusiasts with.
As we have already covered the Phenom 9600 in our earlier review and the Phenom 9900 launch articles, we wanted to take the opportunity in this article to explore the overclocking potential of this unlocked processor, and also investigate the performance effects of the fix for the TLB erratum.
Just how far can this overclocking enthusiast chip be pushed, and how much performance will we really lose by enabling the microcode fix?
A quick recap of the Phenom architecture:
All the currently available Phenom processors are B2 (or possibly even earlier) steppings, and suffer from what is known as the "TLB erratum" - more plainly put, a "bug" or "defect" in the chip, that can at times cause the chip to lock up and crash. AMD has released two different patches that allow working around the problem - one a microcode fix that is loaded into the processors by the BIOS, and another, which is a kernel patch to RedHat Linux.
The Linux patch causes very little performance loss - some say less than one percent - whereas the microcode fix reportedly causes a 10-20% drop in performance across the board. Unfortunately, as Windows is not open source, AMD has not been able to release a kernel patch for XP or Vista to lower the performance hit of the bug; and Microsoft has not seen fit to build such a patch to date either.
We thought we would take this opportinity to examine the impact of the BIOS fix on the benchmarks we normally use to evaluate processors, and as such, we will run our benchmarks with both the patch enabled and with the patch disabled.
In order to keep the testing as fair as possible, we will use the following test platforms:
Socket AM2:
Software used during testing consisted of the following:
Please note that we are showing overclocked results in all the charts - we are not holding you in suspense until the end of the article. The chart labels incorporate a lot of information about the test configuration. The first line shows the socket type and the model of the processor. Since all the processors shown are dual-core devices, we did not specify that on the charts.
The second line shows the "FSB/HT clock rate" x "CPU multiplier" followed by the effective memory speed. All DDR2 tests were run at 4-4-4-12 timings unless otherwise specified.
Business Winstone
Business Winstone tells an interesting story about the Phenom 9600 Black Edition - the microcode based TLB erratum "fix" comes with a 2.1 point - almost 7% - performance hit when running at the stock 11.5x200 MHz setting.
The AMD AOD utility also causes a 0.2 point performance hit, and the Asus M3A32-MVP board is 1.1 points faster than the MSI K9A2.
Content Creation
For Content Creation, the TLB fix causes a 2 point performance drop (about 5%); and having the AOD utility on the system causes a 0.8 point drop in performance. The Asus board is 2 points faster on this test.
Sandra CPU
For the raw CPU test, the TLB fix appears to have no impact on the Phenom 9600 Black Edition - as a matter of fact, the newer MSI K9A2 Platinum appears to be a hair faster than the MSI K9A2 CF motherboard the non-black Phenom 9600 was tested in.
Sandra Bandwidth
Neither the TLB erratum fix nor the AOD utility affected memory bandwidth much in Sandra tests; but note how much better the dual core Athlon 64 X2 6400+ memory performance is!
WinRAR
OUCH OUCH OUCH!
The TLB "fix" absolutely devastates single threaded WinRAR performance. The test without the fix is 2.3X faster!
WinRAR MT
Very interesting. Even disabling the "TLB fix" in the BIOS still does not restore the performance of WinRAR MT - this tells me that the "fix disable" does not disable all parts of the fix.
This benchmark has horrible performance with the fix enabled.
RightMark Read
The Phenom 9600 Black Edition takes about an 11% performance hit for RightMark Read results when the TLB fix is enabled; the other results are more or less in line with what we would expect.
RightMark Write
The TLB fix only appears to have a marginal 2% or so performance hit for the RightMark write benchmark.
RightMark Latency
The Phenom 9600 Black Edition takes an almost 20% performance hit with the TLB fix enabled; mind you this could partially be due to conservative hidden BIOS settings.
RightMark Bandwidth
The Phenom 9600 Black Edition did surprisingly well here, even with the TLB fix enabled - I guess in some ways the BIOS's have been tuned some since my previous Phenom reviews.
LAME MP3
As LAME encoding is pretty much processor bound, I did not expect the TLB fix to have a huge impact - and it did not. The Phenom 9600 Black Edition showed barely more than 1% performance hit due to the fix for LAME.
LAME MP3 MT
LAME MT is more memory intensive, and the performance loss due to the TLB fix is a bit over 4%.
TMPGEnc
Ouch. The single-threaded test shows a roughly 10% performance hit due to the TLB fix, but when multi-threaded, the difference becomes negligable.
CineBench
Once again, the Phenom 9600 Black Edition takes a pretty big (roughly 10%) hit on the single threaded test due to the TLB fix, but suffers far less when the test can take advantage of the multiple cores.
POV-Ray
Interesting. Here the TLB fix does not really affect the single core results, but causes a 30% performance hit when all cores are used!
Call of Duty
The Phenom 9600 Black Edition takes about a 3% performance hit when the TLB fix is enabled.
Commance 4
OUCH! The TLB fix costs 20% of Comanche's performance!
Doom 3
The TLB fix costs the Phenom 9600 Black Edition 11% of performance for Doom 3.
Halo
Halo only takes a 2% hit for the TLB fix.
Jedi Knight
The Phenom 9600 Black Edition takes a pretty bad 13% performance hit in Jedi Knight for the TLB fix.
UT 2004
And UT suffers a 16% performance loss due to the TLB fix. OUCH.
World In Conflict
YIKES!
The TLB fix causes an incredible 42FPS performance loss in the Minimum FPS of World In Conflict with the Phenom 9600 Black Edition - you read it right, the TLB fix drops the minimum performance by greater than 50%.
The Average performance "only" drops by 20%, and the Maximum performance drops by 35%.
If you like World In Conflict, DISABLE THE TLB FIX!
Ouch.
Overclocking
Unfortunately the Phenom 9600 Black Edition is not a great overclocker. How far was I able to exceed its stock 2.3GHz?
On the MSI K9A2 Platinum I was unable to get it stable even at 2.6GHz (13x200) - possibly due to not being able to use our Noctua 12 cooler and having to use an AMD stock heatsink.
On the Asus M3A32-MVP Deluxe I was able to run at 2.6GHz without any problems with the Noctua-12 - as long as I increased Vcore to 1.4V; but I simply could not get the system stable at 2.7GHz.
Basically, the unlocked Phenom 9600 Black Edition let me run it at the Phenom 9900's stock speed, but no higher. I am looking forward to getting my hands on a B3 stepping of the Phenom to see if that will overclock better - in the past, new steppings often ran at higher clocks than previous iterations of a chip as the manufacturers keep tweaking their processes.
Conclusion
I was not expecting miracles from the Phenom 9600 Black Edition - I did not expect it to be as great an overclocker as the Athlon X2 5000+ Black Edition as that chip is a mature product.
I was very interested in finding out how badly the microcode based "fix" for the TLB erratom impacted the performance of the processor, and the results look pretty bleak for enthusiasts.
For business use, the performance hit ranges from 5-7% for office applications, to a whopping 66% for WinRAR.
MP3 encoding takes a 1%-4% hit; video conversion 10% or so, and 3D rendering 10%-30%.
Synthetic memory and processor benchmarks are not affected greatly by the TLB fix.
In the games that we tested, performance takes a 2%-53% (ok, 52.5% to be more accurate) hit, with the average speed loss being around 14%.
If I were you, I'd simply disable the fix. For most personal use, the low risk of occasionally hanging Windows due to the TLB erratum is not worth the significant performance hit you will incurr by enabling the microcode based fix. I really doubt that the TLB issue would significantly increase the number of crashes you would otherwise experience anyway due to software errors. I personally did not experience any crashes in testing on stock speeds with the fix turned off.
AMD is apparently currently testing the B3 stepping of the Phenom, and it is hoped that this stepping will fix the TLB erratum so that the microcode fix will no longer be needed. Rumours are that this fix could be available this March or April.
For server use, I'd simply use the Linux kernel patches that AMD has made that work around the issue with a very small performance loss; frankly, Microsoft should really release hotfixes for their operating systems with a similar patch; then the slow microcode based fix would be history.
Is the Phenom 9600 Black Edition worth it?
Well, its the same price as the regular Phenom 9600, so if you were going to get a Phenom 9600, you may as well get the Black Edition for some relatively painless overclocking to Phenom 9900 levels without having to increase FSB speeds.
Please do not redistribute or use this article in whole, or in part, for commercial purposes.