AMD FX: The Wait Is Over
To say the FX processor, formerly known as Bulldozer, has been eagerly anticipated would be a gross understatement. This chip has been the holy grail that AMD has been striving toward when it comes to their “Enthusiast” line of CPUs. However grail has not had an easy road, it’s journey has been fraught with peril and delays. The reason there has been so much anticipation around this chip is that AMD has redesigned it’s high end CPU line from the ground up, everything is new. Well the wait is over and the prodigal chip has arrived but is new always better?
With the FX series AMD went back to the drawing board and made some adjustments to the way the basic function works. “Bulldozer” moves from the traditional single processor and multiples put on a single die, to creating a dual processor module that has a number of shared resources. This allows more processors to be put in the same physical space, something that had to be done to achieve an 8 core processor with current production methods.
The most significant change in this new design is the removal of a floating point unit from each processor and allowing them to share the FP. The result is an FP that is capable of splitting itself in half to allow dual 128 bit instruction or combine for a single 256 bit run.
I asked AMD about the reasoning behind this move, the explanation I was given is two fold. First the way most software works the need for a dedicated FP unit on each CPU was redundant. The actually usage was low enough in real world apps that moving to a single unit should have minimal impact. To reduce the impact this could possible have on some applications they claimed to have beefed up the FP performance and then have put two instruction sets into the hardware to allow programmers to actually speed up FP calculations even more. The second reason was the need to reduce the space taken up on the silicon so they could get the two extra cores.
The FX also brings a second version of the Turbo Core system introduced in the Phenom X6 lineup. The idea is simple, when the processor is using half it’s cores or less the chip overclocks. However this time around AMD introduced a second level that overclocks ALL of the chips cores if the work load leaves enough headroom.
Now at this point I could throw a hundred different benchmarks at you with a slew of numbers but that is not our style and frankly is not needed. Every hardware site on the planet will be posting numbers at the same time this review goes live. In fact you will be sick of reading all the numbers in a few days. Instead,as is our style, we will instead focus on the meaning of what our testing reveals and cut to the conclusions of what this chip offers. If you want to see some of the specific testing done I suggest checking out Overclockers Club for their in-depth look at the testing.
When I got the package in I must tell you I was impressed, AMD pulled the stops with the presentation. A huge black box with a large FX logo on it opened to reveal an Asus Crosshair V motherboard, a metal box that I presume is the actual box the FX chips will market in and a nice FX logoed belt buckle. The FX chip was already installed on the motherboard just waiting for a heat sink and RAM. For our initial testing I used the board provided to ensure stability for the first round of testing and then switched over to a 990FXA-GD80 that MSI graciously provided. I had not problem testing on the board provided but I hate following directions. Okay seriously the idea is that when a company provides a board and a chip they will sometimes tweak it out, so I like putting it on a different board to double check numbers.
For my initial testing I wanted to see how the new core stood up to the previous generation, the Phenom II. To do this I took an 1100T processor and ran it’s clock speed up to the 3.6Ghz base of the FX-8150 chip we where sent for review. This chip is the high-end of the release with a suggested price tag of $245. I set the two chips to the same clock speed to ensure a true core to core comparison. I also turned off Turbo Core on both chips to ensure this was a straight up comparison.
For testing I was looking at pure CPU performance so I used 7 Zip, Wprime, Super Pi and Cinebench. I made sure the test was set so it was running single core and then also made suire of this by setting the chip affinity in Windows to only Core 0. The results where NOT what I expected. In every test the FX core was coming slower than the Phenom II core. I had the systems set side by side using the same RAM speed, 1333 and the same amount, four gig.. I ran the tests about 5 times before I accepted the results. Averaged out the Phenom II was about 8% faster than the FX in a true Core to Core test.
Next I set the 1100T at stock speeds and the FX at stock speeds both with Turbocore on. I also changed the FX RAM over to Kingston ram set at a speed of 1866, the suggested RAM speed for the FX. In single core testing the FX took the lead but over by about 5% and that can be attributed to a higher clock speed. When I fired up the testing to use all cores the FX gained a slightly bigger lead but that can be attributed to the 2 extra cores, however the bump was not as much as I would have expected, the chips were still close in performance.
In our briefings we where told the 8150 was being aimed squarely at the i5 2500K. So that was the next round of testing. I did not worry about single core testing this time around except a little curiosity benchmarks but for our purposes they do not matter this is a whole chip vs. whole chip match.
I started with some testing to push the chips and used Cinebench and 7 Zip, the results where more along the lines of what I expected. When the extra cores had a chance to kick in the FX picked some solid speed and motored past the i5, test results showed the extra cores really being put to good use.. AMD in it’s briefing had pushed gaming pretty hard so that was next on the block.
For gaming we tested with Champions Online, Civilization V, F1 2010, Batman AA and two games that I am beta testing so cannot reveal the names of. Now I mention these because one of the points AMD made was that this chip will excel with new software, so how much newer can you get? All games where tested at 1920×1080 with the in game detail levels set on high using a Sapphire 6950 for the GPU.
Across the board the games playability was unchanged using the FX or the i5. I never noticed any lag or hiccups in any of the games, all delivering an outstanding play experience. When I took the time to look at the frame rates within the games while playing I did noticed that in most cases the i5 managed to hold better rates, but both chips easily kept play rates anyone would be happy with.
Looking at the performance I have seen with the FX I have to say I am a bit disappointed. Now let me be clear this is not a bad chip, it has been designed looking forward and when it comes to an app that makes use of it’s full core capacity it really can carry some heavy load. However after all the wait and even the hype AMD has given this chip it is lackluster to say the least.
The FX core is actually slower in a direct comparison to the older Phenom II design. It has some interesting features to be sure but it just did not deliver the goods. In fact I would go so far as to say that an 1100T is more competition to the FX 8150 than any chip from Intel. Priced at $50 less than the FX, the 1100T can deliver performance very comparable to the FX. When it comes to comparisons to the i5 this is a wash. While it is faster than the i5 when you hit it with a ton of thread, in most everyday apps it is slower and costs a little more.
AMD made a big deal during our briefing to talk about gaming but the truth is, with few exceptions, that when you game at resolutions most people use the processor begins to become less of a factor in the equation. The FX can game no doubt about it but with the games I tested the performance was nothing to make this a must buy. In fact AMDs own material shows that the chip is just running with the pack when it comes to gaming.
The last area AMD has hyped is overclocking. The entire FX line is unlocked and this mean there is some great overclocking options, however be aware that no matter how much it is hyped any overclocking voids your warranty with AMD. I am in fact a bit concerned with how much attention AMD has put on overclocking. They made a big deal and had a press conference about reaching a new world record in clock speed, that system BTW was not usable for everyday computing. In other words it was a record on paper that holds no other meaning. They then put some spin on a video showing some game play with a heavily overclocked FX chip using liquid nitrogen, something not every, or really anyone is likely to do.
There are even plans to release an FX chip with a liquid cooling solution. The cooler is designed by Asetek and is basically the same cooler as the Antec 920. I put this in my system real quick for testing and found that it does a decent job of cooling but was not really any kind of huge improvement over using a 620 cooler with a push pull configurations. However I will give kudos to Asetek on coming up with a twist to their standard mounting solution. They make use of the existing AMD back plate and use thumbscrews to mount to cooler, super easy.
At the end of the day, after all the hype we are left with a chip that feels more like a lateral move from the Phenom II design than an upgrade. The FP system has been hurt by the new design and is actually less effective than the old method. The use of a proprietary new code set to get a boost for FP operations is just a plain bad idea and history shows this. Unless Intel adopts the same options they will not make it into the mainstream and thus will not show up in software soon.
If you already have a solid Pheom II system, especially a 1090T or 1100T then I would not make the move to the FX. There are only some very specific programs and circumstances where the FX is actually giving a real boost over the Phenom II and the people that need that boost know who they are and will get the chip. For the rest of us this is not an upgrade worth the cost. As for competition with the i5 it is more of a wash than a competition. For most people the i5 is a better solution and actually costs less right now. However the FX can deliver better if the software gets serious about using threads.
As we went to press with this article AMD made some changes to the release info. The biggest news was the addition to a quad core at a suggest price of $115. This puts it in direct competition with the Phenom II 955 and since it comes in with a higher clock speed, 3.6 Ghz, it is a solid choice for those looking to build a budget system. They also dropped the price of the 6 core version of the FX, which has a stock speed of 3.3 Ghz at a suggest retail price of $165. This puts it in direct competition with the 1090T and again makes it an attractive choice.
The Hi-end FX chip, the 8150 is an okay chip but the fact it does not stand out in any way that matters to the majority of the people that would be using it, I have a hard time suggesting anyone buy it. The six core FX-6100 and the quad core 4100 are more interesting releases, they are targeted right at AMDs current lineup and actually make solid choices.
After all the anticipation the final release is a bit anti-climactic. This chip is aimed squarely at the future and is counting on the apps of tomorrow to become more heavily threaded. However in the reality of today the chip is just not something to get excited about.
We will be discussion my findings on the show live this Sunday as well as reviewing the MSI 990FXA-GD80 we used in our final round of testing. Special thanks to MSI for rushing that board out to us so we could test on an independent platform.
For more pictures be sure to head over to Facebook.com/ComputerEd and check out our FX album.
AMD FX Processor discussion segment as aired live 16 October 2011