Author: William Henning
Editor: Howard Ha
Publish Date: Thursday, November 29th, 2007
Originally Published on Neoseeker (http://www.neoseeker.com)
Article Link: http://www.neoseeker.com/Articles/Hardware/Reviews/phenom_9600/
Copyright Neo Era Media, Inc. - please do not redistribute or use for commercial purposes.
Phenom 9600: Good for Upgrades, but is it worth it for new systems?
Finally we are able to bring you some Phenom benchmarks! After Phenom launched a few weeks ago, we were able to get our hands on a Phenom 9600 2.3GHz Quad Core processor and I did extensive testing on this brand new chip. The Phenom is perhaps one of the most anticipated chips of 2007: it is indeed the first ever Quad Core CPU to have all 4 cores on a single die.
AMD has spent a lot of time and marketing effort building up the Phenom brand - obviously intending to associate the new chip with "phenomenal" performance. There were presentations and white papers galore as to how having a "native" (all on one die) quad core solution is better than Intel's multi-chip module approach... a lot of talk about how using a shared L3 cache was superior to having cache coherency traffic sharing the FSB with I/O and memory traffic. Will AMD deliver on its implied promises?
Initially slated for 2.8GHz to 3GHz launch speeds, after several delays Phenom was finally launched at 2.2GHz(the Phenom 9500) and 2.3GHz (our Phenom 9600). Several months late, and 600MHz slower, was it worth the wait?
Only you can decide.
There is no question that there is some amazing technology in the Phenom 9600. It is the first native quad core desktop part - native, meaning that all four cores are on the same physical die - and it has some worthwhile improvements to the Athlon microarchitecture.
Unfortunately for AMD, the market could not care less if a quad core processor was a "native" quad core part, or a dual die "multi-chip module" - after all, as far as the public is concerned, if it has four cores, and fits into a single socket, its a quad core part. Joe Public could care less about cache coherency traffic going across an FSB. The public does not care about the FSB vs. point to point interconnect issue or on-processor memory controllers either - the public cares about two things:
Intel also took the opportunity to take more wind out of AMD's sails by sampling the upcoming 3.2GHz QX9770 quad core Penryn processor (plug: read our review of it here) pretty much at the same time as AMD launched its Spider platform and Phenom.
Back in September, AMD "launched" Barcelona - and we covered the improvements in the architecture, one of which is 512KB L2 cache per core
Basically the following improvements were made going to Phenom:
Now let's take a quick peek at the motherboard we are running the Phenom 9600 in...
MSI K9A2 CF Quick Peek
Before we delve deeper into the Phenom 9600, I wanted to do a quick overview of the motherboard we used: the MSI K9A2 CF. Unfortunately we did not have time to do a separate motherboard review for this 790 chipset based board, but we can present you with the specifications - and since we were able to test the processory, you know the motherboard and also have an idea of its overclocking potential and limitations.
MSI's website tells us that the motherboard normally comes with an IDE cable, SATA cable, and SATA power adapter cable; presumably you would also get a manual and a driver CD.
The board appears to be a decent way to get into a Crossfire setup.
The back panel has four USB 2.0, one Gigabit ethernet, PS/2 mouse and keyboard, serial port and audio connectors. A bit sparse if you ask me, it really could use an SP/DIFF output and some more USB ports.
The BIOS was pretty decent, with a fair number of tweakable parameters in the Cell menu - however I could not adjust the CPU multiplier at all, and the layout and behaviour of the memory settings was confusing.
• AMD® Phenom/Athlon/Sempron CPU.
• HyperTransport 3.0 supporting speed up to 2600MHz
• AMD® 790X and SB600 Chipset
• 1 mazarine PCI Express x16 slot with x16 operation (PCI Express Bus SPEC V1.0a compliant)
• Chipset integrated by Realtek® ALC888
• Realtek PCI-E GbLAN controller 8111B
Internal I/O Connectors
- ATX 24-pin power connector
Back Panel I/O Ports
- 1 x PS/2 keyboard
• 30.5cm(L) x 22.0cm(W) ATX Form Factor
• 6 mounting holes.
Here are the DIMM sockets:
And the SATA ports:
In order to keep the testing as fair as possible, we will use the following test platforms:
LGA 775: (please read the processor review to see which exact board it was tested on)
ASUS' M2N32-SLI motherboard is an excellent performer, and allows us to try very high FSB speeds.
Naturally, we also used a high-end 975X-based motherboard for our LGA775 benchmarking. In this case it was the ASUS P5B-E Premium -- a favourite of ours here.
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.
The AMD Phenom 9600 does very well at Business Winstone, with the 2.3GHz Phenom beating the 2.66GHz QX6700.
And we have another win for the Phenom 9600, again beating the faster QX6700.
The Sandra CPU tests show the Phenom 9600 outperforming dual core processors, but being outperformed by the Intel QX6700.
Frankly, I was dissapointed at the poor showing of the Phenom 9600 on the Sandra memory benchmark. Perhaps a new BIOS will allow for better results.
The stock Phenom 9600 results are quite poor, and the overclocked results are not much better.
The multi-threaded version of the WinRAR results is basically the opposite of the single core results - the Phenom 9600 does quite well at stock, and even better when overclocked.
Ouch. The AMD Phenom 9600 gets the distinction of getting the worst result here.
The Phenom 9600 did better for writes, but the results are still nothing to write home about.
I think there must be something wrong with how the BIOS configured the on-chip memory controller; while the read latency numbers are good, earlier dual core AMD chips get a better result than the new Phenom 9600.
Ouch. The Phenom 9600 had the worst bandwidth of the listed results.
Ouch. Granted this is a single core test, but the 1.8GHz E4300 beats the 2.3GHz Phenom 9600.
The Phenom 9600 fares much better at TMPGEnc, with the 2.3GHz Phenom 9600 beating the 2.13GHz quad core Xeon X3210
The 2.3GHz Phenom 9600 is outperformed by the 2.13GHz X3210.
And the Phenom is outperformed even more significantly here.
Call of Duty
Ouch. The 2.1GHz X3210 is 22% faster than the 2.3GHz Phenom 9600 here.
Or almost 32% faster if you adjust for clock speeds.
Interesting. The Phenom 9600 beats the Xeon here.
The Phenom 9600 loses to the slower Xeon again.
Ouch. That pesky 2.13GHz Xeon beats the Phenom 9600 by about 12%. Running at a lower clock speed.
By this time none of us are surprised that the Phenom loses.
I'll just say it.
Overclocking the Phenom 9600 was a pain in the posterior.
Surprisingly enough, the memory controller. Heck, the reported TLB errata may have had something else to do with it as well.
When I initially set the board up, I tried to use our trusty Corsair PC2-8888 modules that I know work at up to 1111MHz at default timings, and up to about 1080MHz at 4-4-4-12 timing. The idea was not to be limited by the memory.
I spent over an hour trying to get them to work, and gave up in disgust. Since some motherboards I've tested in the past have had problems with DIMM's that require a default voltage over 1.9V, I put in a pair of Corsair PC2-5400's that I knew work at up 10 1000-4-4-4-12 - and finally I was able to get reliable operation at stock settings. CPU-Z reported some insane 119-1-3-3-5 setting, but I wrote that off to the utility not being Phenom aware.
The memory benchmark results were... sad.
Eventually I managed to get somewhat better memory results, but I have a nagging suspicion that the BIOS on the board I was using is not using the modules in dual channel mode - as previous Athlon's I tested were able to obtain significantly higher memory performance.
To get the system stable when overclocked, I had to boost Vcore to 1.4375, raise Vdimm to 2.1V, and set the memory to 766-4-4-4-12 2T
I persisted, and eventually I was able to get a stable system with settings of 11.5x230 with memory running at 766-4-4-4-12 2T
Mind you, the AMD Overdrive utility reported this as 240.5x11 800-4-4-4-12 2T.
But CPU-Z and the BIOS said it was what I set it for... 11.5x230 766-4-4-4-12.
It was at least a 15% overclock; possibly 20%, and the system was stable at 2.65GHz (or is it 2.75?)
The performance was quite decent, but it did not blow me away - remember, I just finished overclocking the highly overclockable Penryn QX9650 and QX9770's, so the numbers I was seeing did not impress me - and I am a staunch AMD fan.
I guess I have become spoiled by Penryn's and other lower powered Intel parts.
Unfortunately I don't have power consumption figures from my previous QX6700 review, so the closest figures I had were from my recent Xeon X3210 review - and the Xeon wins hands down.
We wish to thank NCIX for lending us the Phenom 9600 and motherboard we used in this review.
Testing the Phenom 9600 has been interesting.
The BIOS and CPU-Z agreed on the speed of the processor, but AMD's Overdrive utility showed the "actual" speed as being roughly 5% faster than what the processor was set for. Regardless, at either 2.65GHz nor 2.75GHz the Phenom turned in some decent numbers, but frankly it could not touch the overclocked Intel results.
The office results for Phenom were outstanding, however Phenom is not at this point capable of matching Intel parts clock-per-clock in performance.
There is no doubt that it is an interesting piece of technology; and AMD's initial pricing for it does give it a chance for some market penetration.
Personally, I would not hesitate to recommend a Phenom 9600 to someone who could use it to upgrade an existing Socket AM2 system. Getting a quad core processor for around $250-$280 for an existing single or dual core Socket AM2 system - without having to change anything else except updating the BIOS - is an excellent deal (provided such a BIOS update were available).
However, I cannot recommend getting a Phenom at this time for new system builds, nor for enthusiasts and overclockers; at least not at the current prices. A Socket 775 Xeon X3210 will easily outrun the Phenom 9600 the vast majority of the time, even at stock speeds, and it has far greater overclocking headroom for enthusiasts - and it costs about $15 less.
In case you are wondering, I did not compare the Phenom to the new QX9650 quad core Penryn parts as they are at vastly different price points (the latter being MUCH more expensive), and comparing apples to pumpkins is not a fair comparison.
Now before you think that this review is too negative, there are a couple of silver linings to all the clouds:
I guess I am disappointed in Phenom after all the hype; but considering it on its own merits, its a decent upgrade.
Please do not redistribute or use this article in whole, or in part, for commercial purposes.