As you may have already read in our news article this morning, AMD released six new CPUs today. For your viewing pleasure, we are graced with not one but two of these puppies to put under the microscope! This is a review of the new hex-core Phenom II x6 1075T. Don’t forget to check out splat’s review of the Phenom II x2 560 BE after you read this one!
For quite a few months now, there have been two choices of Thuban – the Phenom II x6 1055T and the Phenom II x6 1090T. This new hex-core drops in right between the two, in clock speed (3.0GHz stock / 3.5GHz with Turbo) and in price, coming in at a respectable $245. This is a heck of a lot of processing power for less than two hundred fifty bucks, that’s for sure. Compared to Intel’s cheapest six core offering (currently retailing for $900), it’s a steal!
The CPU we received came OEM, meaning it was just a CPU by itself with no accessories. Presumably it will come complete with the same package (albeit non-Black Edition) as its big brother, the 1090T. The stock cooler was the same for the Phenom II x4 965 BE & x6 1090T, so it will likely make its way in the 1075T box too.
For reference, here is a Black Edition box next to a non-Black Edition box.
One thing is for sure, the AMD stock cooler of recent years is a fair bit better than the non-extreme edition Intel stock cooler. If you’re not overclocking, this will keep your processor nice and cool. Now let’s take a look at today’s subject.
Of course, what’s a CPU review without a couple gratuitous die pictures?
Six true cores and a bunch of cache is oh so pretty!
Enough gandering, what does this thing have under the hood?
Model Number & Core Frequency: X6 1075T / 3.5GHz (Turbo) / 3.0GHz (Base)
L1 Cache Sizes: 64K of L1 instruction and 64K of L1 data cache per core (768KB total L1 per processor)
L2 Cache Sizes: 512KB of L2 data cache per core (3MB total L2 per processor)
L3 Cache Size: 6MB (shared)
Total Cache (L2+L3): 9MB
Memory Controller Type: Integrated 128-bit wide memory controller *
Memory Controller Speed: Up to 2.0GHz with Dual Dynamic Power Management
Types of Memory Supported: Unregistered DIMMs up to PC2-8500 (DDR2-1066MHz) -AND- PC3-10600 (DDR3-1333MHz)
HyperTransport 3.0 Specification: One 16-bit/16-bit link @ up to 4.0GHz full duplex (2.0GHz x2)
Total Processor-to-System Bandwidth: Up to 37.3GB/s total bandwidth [Up to 21.3 GB/s memory bandwidth (DDR3-1333) + 16.0GB/s (HT3)]
Up to 33.1GB/s total bandwidth [Up to 17.1 GB/s memory bandwidth (DDR2-1066) + 16.0GB/s (HT3)]
Packaging: Socket AM3 938-pin organic micro pin grid array (micro-PGA)
Fab location: GLOBALFOUNDARIES Fab 1 module 1 in Dresden, Germany (formerly AMD Fab 36)
Process Technology: 45-nanometer DSL SOI (silicon-on-insulator) technology
Approximate Die Size: 346mm2
Approximate Transistor count: Similar to Istanbul’s ~904 million
Max TDP: 125 Watts
AMD Codename: “Thuban”
*Note: configurable for dual 64-bit channels for simultaneous read/writes
Expected retail price: $245
Here we have a true six-core CPU that clocks in at 3.0GHz for $245. What’s not to like?
Test System, Opponents and Methodology
Happily, we get to see this processor tested in a top of the line current-generation AMD setup. Here’s the skinny:
|Motherboard:||ASUS Crosshair IV Formula|
|CPU:||AMD Phenom II x6 1075T|
|Cooling:||Thermalright Venomous X|
with Push-Pull Delta Screamers
|RAM:||G.Skill Pi DDR3-2400|
|GPU:||ASUS Matrix 5870 Platinum|
|PSU:||Cooler Master Silent Pro 600M|
|OS:||Windows 7 x64|
There is some stiff competition in today’s review. There are two CPUs from Intel and three variations from AMD:
|Intel’s Entries||AMD’s Entries|
|i5 655K||Athlon II x4 640|
|i7 870||Phenom II x4 965 BE|
|Phenom II x6 1090T BE|
Stock benches were all run three times and the average of those is what you’ll see in the results. Overclocked tests were run once.
For the i7 870, tests were only run at its 24/7 overclock of 4.0GHz. For all of the rest, they were run at stock and their respective 24/7 overclocked speeds.
|Processor||Stock Speed||Overclocked Speed|
|i5 655k||3.2 GHz||4.5 GHz|
|Athlon II x4 640||3.0 GHz||3.74 GHz and 4.0 GHz|
|Phenom II x4 965 BE||3.4 GHz||4.0 GHz|
|Phenom II x6 1090T||3.2 GHz||4.0 GHz, both 200 x 20|
and 267 x 15
Now you know the competition, let’s dig into the processor.
We’ll get testing under way with the Everest suite of CPU tests. There are four CPU tests and three FPU tests. The i7 870 didn’t make it for this round. All tests were run at each processor’s stock frequency. Note that none of the results’ graphed ranges start at zero. This is to help highlight the differences between the processors. Actual differences in score may not look as distant if graphed from zero. First, the CPU quartet.
As expected, the 1075T played a very close second fiddle to the 1090T in all of the CPU benches. The big surprise here is the AES test, where the i5 655K just eclipsed every other competitor. According to Everest, AES is an “integer benchmark [that] measures CPU performance using AES (a.k.a. Rijndael) data encryption.” While Photoworxx and Zlib both are strongly dependent on memory subsystem performance while Queen benefits from the shorter pipeline, it seems AES focuses solely on the CPU’s processing of integers.
I’d give the Thubans an overall win here, hands down considering the other three tests are more dependent on the CPU and its memory controller and not the CPU itself, giving it a strong real-world computing advantage. Up next, the FPU trio.
No real surprises here as far as the pair of Thubans go, with the 1090T showing off its 200MHz clock advantage and the 1075T right on its heels. What is surprising is the 965 BE’s ever so close race with the 655K, but that’s for another review.
Overclocking for Stability
Results in the remaining tests will show stock and overclocked speeds. So we don’t put the cart before the horse, we’ll see how it faired for the average every-day overclocker. It would be remiss not to mention Dolk’s superb Guide to the Phenom II on our forums and thanking the man himself. It and he have helped numerous people clock their Phenom II processors, myself included.
To start, it’s always good to see how far a chip will go on stock voltage. Mind you stock voltage on the Crosshair IV Formula is a bit less than some other boards. Since everything was left on auto (including no LLC – load line calibration), there is a pretty strong Vdroop of ~.05 V. So stock volts on another board may be able to handle more than this.
Anyway, here’s what we got with no changes what-so-ever except for bumping up the bus speed and seeing where we ended up.
3.3GHz, not bad at all with the aforementioned Vdroop. No one ever cried about a free 300 MHz! Had I turned LLC on (which turns the Vdroop around and makes it a roughly .05 V overshoot of set Vcore under load), this number would no doubt have increased.
Note that ten passes of LinX are a quick and dirty stability test. You will probably want to run more to ensure absolute stability, but these numbers give you a good jumping-off point.
Moving up to see what a solid, reasonable 24/7 overclock was, I set sights on 4.0Ghz. That’s ‘my’ 24/7 overclock. Has been since an E8400 of yore; it has just gotten faster each time because microarchitectures have improved. So that was the goal.
The goal was met, with gusto! For reference, the loaded Vcore (remember the Vdroop / LLC thing?) was 1.416 V as read in CPU-z. As you can see in CoreTemp, there is plenty of headroom left. 46° C max on quality air cooling in a not-so-cool 3rd floor office in North Carolina means there is room to go up, even more so in the winter. We’ll stick to 4.0Ghz for testing though. It’s completely stable, hassle-free and a nice round number.
For reference, other voltages used to accommodate the stable 4.0GHz overclock were 1.25 V each on the CPU-NB, HT Ref & NB. Less Vcore was not an option (tried it), but these three were safety voltages and may be able to be reduced. As always, YMMV and overclocking is never guaranteed. I will say this though – if you’re on the Crosshair IV Formula, get the newest 1105 BIOS and flash for using this chip. Previous BIOSes (0905 and 1005) gave me problems with higher HT and NB clocks, but this one gave 2400 MHz stable as you can see in the screenshot above.
Benchmarking the Beast
First up, let’s check out 3D Performance. We’ll be assisted today by one gaming benchmark – Stalker: Call of Pripyat – and two 3DMarks – 06 and Vantage. Enough dilly dallying, there’s a Stalker behind you!
Stalker: Call of Pripyat Benchmark
These tests were run at 1280 x 1024 with Ultra detail and MSAA both off and at 4x. The 1075T had a strong showing, barely edged out by the 1090T and the i5 655K. The i7 870 jumped a bit ahead here but with only a 4% advantage. With MSAA turned on, the results for the 1075T were stunning. Take these with a grain of salt though. The third test, “Rain” is what accounts for this anomaly. I ran this test at least six or seven times. The benchmark was uninstalled and all references to it (even registry keys) obliterated, then reinstalled again. It gave the same result every, single time. To prove I’m not crazy, here’s a screenshot.
Your guess is as good as mine. These numbers are simply reported as performed. I would expect actual results to mimic the 1090T, which are right on par with the Intel offerings.
3DMark06 and 3DMark Vantage
We’ll examine both the full benchmark and the scored CPU element of both of these. Oldest first!
In overall score, the i7 870 rules the roost with the 1075T coming in second. This is a surprise considering its CPU score was bested by the 1075T and the 1090T with the latter just barely edging out the 1075T. Surprisingly, it seems bus speed (which includes an increase in HT & NB clocks) has little to do with the outcome of the CPU test. Now for the new(er) kid on the block.
The 1075T comes in a very respectable second place here. While Vantage clearly likes its cores and HT, the hex-core AMD processors still show some spunk. So while 06 seems to prefer physical cores, Vantage is equally content with virtual ones.
Moving on, let’s check out 2D performance.
Cinebench R10 and Cinebench R11.5
These are the two most ‘real-world’ tests we’ll see in the review. These direcly correlate with how productive you can be if you are using your computer for productivity akin to rendering (i.e. audio / video / photo editing and, you know, rendering).
This was a very tight race. The i7 870 squeezed out a win in R10, but both Thubans pushed it aside in R11.5, with the 1075T even beating the 1090T by a nose. I’d call rendering a draw between the toughest competition. Considering the i7 870 carries a $45 premium over the 1075T, I’d be looking in the red direction.
SuperPi 1M and 32M
SuperPi has long been a favorite of mine. Unfortunately it is no favorite of AMD in general as most people know. Here’s how they ended up comparing.
Make no mistake, they’ve come a long way, but they still lag behind Intel when using this metric. SuperPi performance is still squarely on the side of the blue team. It’s still a blast to run to compete AMD vs AMD though, and versus its own kind, the 1075T was the top performer.
Hold on just a minute though – we have one more single core benchmark to go!
Now the red team has something to smile about. The 1075T was number one in the AMD camp, but not only that – it tied the i7 870 and only lost to the 500 MHz-higher-clocked i5 655K. Obviously these two benchmarks are coded differently, but it’s nice to see AMD keeping up with a single-core bench.
WPrime 32M and WPrime 1024M
Now we come to this processor’s bread and butter – multi-core benching. As far as my favorite benchmarks go, WPrime is second only to SuperPi.
Ahh yes, that’s nice. The 1075T beat out all of the competition except the i7 870. Even then, it was within spitting distance of its nemesis. Definitely a solid showing by a very stout CPU.
Pushing the Envelope
Now we get to the fun part. Everyone knows that AMD processors love cold; they thrive on it, eating close to absolute zero temperatures (think liquid helium) for breakfast. Unfortunately, not every overclocker has that ability. In the future, I hope to bring something sub-zero to this CPU, but for right now we’ll satisfy ourselves with a little air-cooled goodness. Note stability wasn’t even an afterthought with any of these benches; if it ran them, it was stable enough!
After pushing some reasonable voltage, it turns out this CPU is quite happy giving up 300 extra MHz with a little pushing. It ran SuperPi1M, PiFast and WPrime 32M effortlessly.
Definitely throwing up some strong numbers, and on air no less!
AMD has also come a long way with their IMCs (Integrated Memory Controllers). After running these tests with RAM rated far in excess of what most IMCs are capable of, it felt like a good idea to see just how far it would go.
Again, there is nothing even remotely stable about this. It was just an experiment to see how far the IMC could be pushed and still run MaxxMem. That said, WOW, I’m thoroughly impressed with how far it got. Not only did the CPU make it to DDR3-2111, it did that at 7-9-7-24! It blew me away, far exceeding what I was expecting. Three cheers for AMD’s new IMCs!
One important note about memory clocking: There is a trade-off. It seems you have a choice with the Thuban IMC – a) high memory clocks combined with lower HT / NB frequencies or b) lower memory clocks (presumably with tighter timings) combined with higher HT / NB frequencies. Which you go with will be determined by what works best on your particular setup, but now you know of the decision you’ll have to make.
Of course, it’s never enough to just run SuperPi 1M and other benches at a certain overclock. That just means you can run SP1M higher. How high?
4.4 Ghz, that’s how high! Good show 1075T, you exceed my expectations here too.
Now, what’s a good overclocker without a goal that feels silly? Well, it’s not a very good overclocker, that’s what! After running SuperPi 1M at 4.4 GHz, I got it into my head that I’d get a higher validation out of this thing if it killed me. My mind was set on 4.5 Ghz. After more than an hour of booting, pulling up CPUz, trying to save a validation and take a screenshot, crash, rinse & repeat, it finally happened. Here it is, complete with the voltages used.
Final Thoughts & Conclusion
Whew, what a ride. This processor was a blast to benchmark and overclock. 4.5 GHz on air was just icing on the cake. Competitive through 3D testing, real-word rendering testing and 2D tests, the entire experience was a romp through a field of silicon goodness.
Are there negatives? Sure, there always are. It’s not the fastest six-core CPU on the market. Intel holds that crown and AMD isn’t trying to dispute it. Where AMD wins out though is price-for-performance. The Phenom II x6 1075T has an MSRP of $245. That’s less than a third of what Intel’s cheapest hex-core CPU costs. While there unfortunately isn’t one here to bench, I daresay this CPU performs more than one-third as well as its Intel competition.
Speaking of competition, as you have seen, this processor is in a dead heat with Intel’s i7 870. Considering the i7 is $45 more than this chip, I’d call the 1075T the clear winner.
It’s also not a 1090T (obviously). It’s not supposed to be either. It’s an additional Thuban to help mould AMD’s offerings to your specific budget. For the frugal overclocker, this is a great CPU for the money. You can save a bit of cash over the 1090T while losing very little, gain an ever important multiplier over the 1055T and put the extra savings toward another component.
With the 1075T, you get what you pay for and – as you can see – a whole lot more. I wouldn’t hesitate a moment to recommend this CPU to anyone in the market, especially if you run heavily multi-threaded applications. The Phenom II x6 1075T is undoubtedly, indubitably, 100% (all together now!) Overclockers Approved.
–Jeremy Vaughan (hokiealumnus)