Athlon X4 845 Beats 7.4 GHz Bulldozer - Excavator, the Harbinger of Improved AMD Performance?

  • News, Articles
  • 13
  • HWBOT

Athlon X4 845 Beats 7.4 GHz Bulldozer – Excavator, the Harbinger of Improved AMD Performance?

In recent weeks we have been able to explore the potential of AMD’s redesigned CPU architecture, codenamed Excavator. However, before we get down to the deeper details let’s first consider just how good these new Excavator x86 cores really are. For the sake of comparison, consider these numbers: an AMD FX-8370 ‘Vishera’ core processor clocked at 7.273MHz (+81.82%) scores 10min 44sec 344ms in Super Pi 32M and an AMD X4 845 ‘Excavator’ core processor clocked at 4,597.57MHz (+31.36%) scores 10min 12sec 78ms in Super Pi 32M. Excavator wins, despite being clocked much lower.

These somewhat jaw dropping scores we are seeing from the X4 845 offer a glimpse of hope for AMD, adding fuel to the fire that maybe, just maybe AMD has finally managed to break out of the performance mud it has found itself mired in for the last three years or so. Let’s now consider the details.


A First Flirt with Excavator: The AMD Athlon X4 845

It’s been three months since AMD officially and rather underwhelmingly released the Athlon X4 845 CPU. The new quad core CPU used newer Excavator cores that had previously only surfaced in Carrizo APU mobile parts. The Excavator microarchitecture succeeds Steamroller, and represents the fourth generation of Bulldozer-derived x86 architectures. However, where Piledriver and Steamroller had actually moved the performance delta by very little, Excavator actually pushes AMD to a whole new level of performance. One that admittedly, we didn’t quite expect.

The Athlon X4 845 is a quad core CPU with stock settings of 3.5GHz, turboing to 3.8GHz. It features a 2x 1MB L2 cache, DDR3 -2133 memory support and a TDP of 65 watts. Currently here in Taipei the chip retails for a NT2,090 Taiwan Dollars, which works out at a snip over $60 USD. It’s currently the only Athlon available that uses Excavator cores, with all other current FM2+ CPUs and APUs instead using the previous Steamroller architecture. At such a low price point the X4 845 is not marketed as a performance part; it is in fact half the price of the A10-7860K, an APU which uses four Steamroller cores and a full compliment of 8 Graphics Core Next GPU units.

Despite not being built for glory or high-performance benchmarking, the performance we have seen thus far has been utterly outstanding.


Excavator: Fastest AMD Cores Ever

From what we have seen of Excavator so far however, there is certainly plenty of cause for optimism. If we take PiFast as a solid representation of single-threaded performance, we see that core to core, Excavator is actually in a whole new ballpark compared to previous generation cores. With reasonably little time and effort Massman had the X4 845 pushed to 4.3GHz with the North Bridge configured at 1.6 and DDR3 tuned to DDR3-2220 C7 (un-optimized). Very quickly it became evident that Excavator was not only outperforming previous generations, but doing so with better efficiency and on lower clocks.

n a PiFast shootout, the picture looks like this:

PiFast Score
(lower is better)
Gen CodenameCPU FrequencyIMC FrequencyMemory Configuration
27.92 secGen4 (Excavator)4.3 GHz1.6 GHzDDR3-2200 C7
29.52 secGen3 (Steamroller))4.5 GHz1.9 GHzDDR3-1660 C9
29.78 secGen2 (Piledriver)4.6 GHz2.4 GHzDDR3-2400 C10
31.39 secGen1 (Bulldozer)4.9 GHz2.4 GHzDDR3-1600 C6

The fact that Excavator is considerably faster than previous iterations is somewhat overshadowed by the manner in which in wins. The X4 845 is actually locked and can only be overclocked by upping the BCLK to a peak of around 123 MHz, this somewhat limits the potential of the chip in pure clock terms. Regardless, the Excavator cores are delivering in spades with lower clocks, a lower NB frequency and unoptimized slower memory configurations. To quote l0ud_sil3nc3, the current holder of the fastest SuperPI 32M AMD <5 GHz score in the world: “This is the fastest <5Ghz 32M on AMD ever done.... What's crazy is this is pretty close to the all out fastest AMD time iirc done by Roman or HiVizMan‘s at > 8Ghz.”

Note: The Super Pi 32M AMD <5GHz Challenge compares generations of AMD CPU cores according to their performance at or below 5GHz on the Super Pi 32M benchmark.

Looks like l0ud_sil3nc3 is impressed, and rightly so. The American used an Athlon X4 845 pushed to 4,597.57MHz (+31.36%), this resulted in a Super Pi 32M run of 10min 12sec 78ms, more than 2 minutes faster than the fastest Steamroller-based submission. This indicates that core-to-core, even at lower clocks, the X4 845 is way ahead.



Just to reiterate… this is with considerably lower clocks and inferior memory settings overall. This indicates enormous improvements in overall efficiency, and the potential for plenty of performance once the brakes are off. We can only imagine what an unlocked and memory optimized Excavator chip could be capable of.


So, Why No Fireworks?

To say that the arrival of AMD’s Excavator had the media reach of a proverbial fart in a spacesuit would not be a total exaggeration. The mainstream tech media were more focused on the two APUs that were released the same day (the A10-7860K and the A6-7470, both of which feature Steamroller cores) and the fact that AMD was bundling a new Wraith stock cooler. This is understandable however as AMD FM2+ Athlon CPUs get very little attention from any media, tech or otherwise as they do indeed occupy the lower reaches of the CPU food-chain.

The truth is however that with Excavator, there really is a substantial story worth telling. AMD has been stuck at basically the same level of performance for several years, spanning several technology generations. Bulldozer was introduced more than five years ago, with successive generational updates that failed to deliver anything of significance. The X4 845 indicates that Excavator is the biggest performance bump the company has achieved for a decade or more. Oddly enough it’s not far off the performance and efficiency improvements that the company promised almost a year ago.



So why introduce the X4 845 chip at all? It would seem perhaps a little odd that AMD chose to utilize its latest, and clearly greatest CPU core architecture on a budget FM2+ CPU, especially when you consider how far off the pace their current flagship FX Series CPUs are performing. Why not give us a refreshed octa-core FX series CPU based on Excavator? It might just give us all hope that AMD has what it takes to get back in the performance race with Intel. The main reason is actually that the manufacturing process has been optimized to produce lower power consumption, mobile CPUs, and AMD is having problems getting the clocks high enough.

We have no definitive answers at this moment in time but we do know that the X4 845 is in fact a repurposed Carrizo die with GPU and FCH parts hacked off. Has AMD effectively found a way to monetize defective mobile Carrizo dies? Perhaps so, after all, the company is not in a position to turn its back on any way of maximizing revenue. Thomas Ryan, writing for SemiAccurate.com, perhaps harshly describes the X4 845 as a ‘dumping ground of a SKU’. Nevertheless, his examination of Excavator leads him to conclude that “Carrizo [is] ahead of Godavari by 6 percent overall and 14 percent in multit-hreaded performance.” Not bad for a dumped product.


Bristol Ridge, then Zen

On the other hand, from a strictly business perspective the release of the X4 845 is a great way for AMD to prove to OEM customers that they are back on the right track. If there are engineers and decision makers who need convincing that the company now has a product that resides within a whole new realm of efficiency and performance, AMD can politely refer them to the X4 845. A solid proof of concept if you will.

Recent rumors that AMD’s 7th generation Bristol Ridge will be 50% faster than Kaveri can be reinforced by the performance we are seeing today on the X4 845. Bristol Ridge is rumoured to use four x86 Excavator cores flanked by up to eight GCN3.0 GPU cores. When you factor in other design improvements including support for DDR4 and move to a more SoC-like implementation, Bristol Ridge might actually be something to get truly excited about when it launches in just a few weeks.


And then we have Zen, AMD’s real stab at getting back in the race, with CPUs that we all can truly compete with Intel’s Extreme Edition processors. Our findings with the X4 845 do nothing to dispel a fantasy where AMD actually gets back in the big time and the x86 processor market becomes livelier than it has been for more than a decade.


– by Sdougal


13

Noxinite says:

Nice article! Now just hoping zen lives upto our expectations.

Australia newlife says:

With my best kaveri time @ 5.588ghz the 845 has a good minute on it while 1ghz lower

United Kingdom GeorgeStorm says:

Gives me hope for Zen being decent :D

Ukraine StingerYar says:

Zen promises to be pretty powerful one, nice :)

K404 says:

So, some of us quietly hope to achieve Zen in the near future, but accept that, one day, even that will not be enough. I do hope AMD pull a well-fed rabbit out the hat here :)

United States l0ud_sil3nc3 says:

newlife said: With my best kaveri time @ 5.588ghz the 845 has a good minute on it while 1ghz lower


Did you end up picking up an 845? If so looking forward to your results :)

As for Zen I will be happy if it at least matches Haswell level performance, only a few more months and we will all know.

FlanK3r says:

I think, The Zen could be in single performance around the SB or so, but with LN2 could eb fun (hope 7 GHz or so and no CB)

Australia newlife says:

l0ud_sil3nc3 said: Did you end up picking up an 845? If so looking forward to your results :)

As for Zen I will be happy if it at least matches Haswell level performance, only a few more months and we will all know.


Not as of yet because of a lack of it being launched in Australia although I had a certain someone ask me if I had a fm2 board plus they asked a friend of mine my postage address so I can only guess but if I do get one I wouldn't hold my breath in me beating your time unless I bin CPU's because of the bclk and NB being so important

As for Zen on OCAU we had a guy do some testing with a 845 and it looked like Zen was going to be around ivy and haswell based on AMDs 40% increase

United States Schmuckley says:

I like this article!

France Taloken says:

It will be a SoC, NB and SB fully integrated in the package, don't you think it will cause high CB issues, like Kabini ?

Slovenia tiborrr says:

Full integration doesn't say nothing about CB issues. It's just the general design of maintaining correct path distance in order to prevent signal shorting to adjacent transistor.

Australia newlife says:

Taloken said: It will be a SoC, NB and SB fully integrated in the package, don't you think it will cause high CB issues, like Kabini ?


From my experience kabini actually has a better cb/cbb over kaveri in that the cb is only a few c different from cbb at around -60 assuming that board is insulated properly such as eraser not vas (otherwise could cbb at -10) while with kaveri I've had cpus cbb from anywhere around -40 to -80 while cb is around 20c higher

United States Redwoodz says:

Zen should be around i7 5930k stock speed IPC based on 40% increase.

Please log in or register to comment.

Leave a Reply: (BBCODE allowed: [B], [QUOTE], [I], [URL], [IMG],...)