AMD Phenom Technology Demonstration November 18, 2007 Brandon Sandman Bell

AMD Phenom Technology Demonstration

Page:: ( 1 / 9 )

As most of you know by now, the Phenom launch was reportedly pushed back to the end of 2007 in order to ensure that Phenom would be able to launch at high clock speeds: while Barcelona launched in September at speeds as high as 2.0GHz, word on the street was that Phenom would launch at 2.6GHz. The appearance of RD790 motherboards at retail sites like Newegg in the past few weeks has fueled further excitement around Phenom’s launch.

Unfortunately, we’ve got bad news for you folks. Today’s Phenom launch isn’t as rosy as AMD’s official press release would have you believe. Instead of launching at 2.6GHz, Phenom is actually launching at speeds of just 2.2GHz and 2.3GHz. Availability is also extremely limited; in fact, for all intents and purposes, a more appropriate name for Phenom right now would be Phantom. To illustrate why we say this, have a look at our Phenom CPU sample:

[image]

Phenom CPU and 790FX Chipset

Page:: ( 2 / 9 )

So exactly what is launching today? As we mentioned earlier, Phenom is debuting at speeds of 2.2GHz and 2.3GHz. The 2.2GHz chip is named the Phenom 9500 and is priced at $251, while the 2.3GHz CPU is branded as the Phenom 9600 and carries a price tag of $283. Of course, keep in mind that these are the prices AMD charges distributors in quantities of 1,000 CPUs and not the actual street price of the CPUs. In case you aren’t familiar with the architecture behind Phenom, we’ll provide a very quick overview going over the highlights of the processor.

Phenom is AMD’s first quad-core design, and it’s a native quad-core solution, unlike Intel’s first generation quad-core which is simply two processor dies dropped into one package.

Each of Phenom’s four processing cores has been redesigned to deliver improved performance clock-for-clock over Athlon 64 X2. For instance, the floating point unit supports single-cycle SSE: all 128-bit SSE, SSE2, and SSE3 instructions will complete within one cycle, which effectively doubles the execution speed for these instructions. AMD has also added new extensions to SSE3, while branch prediction has also been improved.

Phenom also boasts a new L3 cache that can be shared across all four cores. The L3 cache size is 2MB. In addition to the L3 cache, each processing core also has its own 64KB L1 cache and 512KB L2 cache (2MB L2 cache and 512K L1 cache total per processor). The memory interface between L1 and L2 caches has also been widened to 256 bits.

To improve performance, the memory controller has been tweaked to take better advantage of the higher memory bandwidth offered by DDR2 memory. Phenom’s new memory controller officially supports DDR2 speeds up to 1066MHz. Phenom also supports HyperTransport 3.0, offering speeds of 3.6GHz full duplex.

With four CPU cores inside Phenom, power is obviously a huge concern. To reduce power consumption, the CPU’s four cores can run their clock speeds and voltages independently of each other: if you’re running a single-threaded app that’s only using one core for example, the other three processing cores can run at lower clocks and voltages to save power. The memory controller can also be powered down to conserve energy.

Phenom is AMD’s first processor to use the new AM2+ socket. AM2+ is fully-backward compatible with AM2 though so those of you with AM2 motherboards should be able to drop in a Phenom CPU when you’re ready to upgrade your existing processor, all you’ll need is a BIOS update. Obviously you won’t get HyperTransport 3.0 speeds though.

Phenom is built on AMD’s 65-nm manufacturing process with SOI, and is manufactured at AMD’s Fab 36 facility in Dresden, Germany. The CPU consists of approximately 450 million transistors and boasts a die size of 285 square millimeters. Both the Phenom 9500 and 9600 share a max TDP of 95W.

AMD 790FX Chipset

Launching alongside the new Phenom processor is AMD’s 790FX chipset. 790FX is designed to appeal to the high-end enthusiast crowd. With up to 42 PCI Express 2.0 lanes, 790FX supports features like triple and quad CrossFire, which can deliver a performance improvement of up to 3X over a single Radeon graphics card, or be used to drive up to eight monitors simultaneously.

HTPC users will love AMD’s new 7-series platform as well. Since it’s built around TSMC’s 65-nm manufacturing process, the chipset consumes very little power. We’ve been told that the chipset boasts a TDP of just 10-12W! That’s less than half the power consumption of high-end chipsets from Intel and NVIDIA, and as a result, the chipset can be cooled by a simple aluminum heatsink. Active cooling isn’t required, even for overclocking. The new chipset also supports AMD’s new Overdrive software.

Up to this point Overdrive has been used solely for GPU overclocking, but with the 790FX chipset, Overdrive can be used to overclock the GPU, memory, and CPU.

[image]

The Core 2 Extreme QX9770

Page:: ( 3 / 9 )

At the end of our Core 2 Extreme QX9650 Performance Preview article, we acknowledged the superior performance of Intel’s latest Core 2 Extreme processor, but speculated that Intel could have an even faster part in the works. The Core 2 Extreme QX9770 is that part.

The QX9770 chip is based on Intel’s 45-nm Yorkfield core, just like the QX9650, only it’s an even faster processor. The CPU sports a 1.6GHz FSB and runs at 3.2GHz. With its 12MB L2 cache and quad-processing cores, this CPU makes the Core 2 Extreme QX9650 – already the world’s fastest desktop processor – look like child’s play! In addition, unlike Phenom, Intel actually sampled processors to the media; in fact we’ve had our processor for about two weeks now. The CPU runs fine with X38 motherboards that support 1600MHz FSB speeds like the ASUS P5E3 Deluxe and Gigabyte X38-DQ6. Intel will also be releasing their X48 chipset with the QX9770. The X48 chipset will officially support 1600MHz FSB speeds, while X38 will officially be limited to 1333MHz.

While we’ve got clearance from Intel to publish Core 2 Extreme QX9770 benchmarks today, Intel won’t officially release the processor until Q1’08. We’ve been told that the processor will be priced higher than traditional Extreme CPUs, so expect a price tag somewhere well north of $1,000, while the Extreme’s will continue to sell for $999. Due to its higher clock speed, the QX9770 sports a slightly higher TDP than the QX9650, 136W versus 130W.

The X48 chipset

Just as 790FX is launching with Phenom, the Intel X48 chipset is supposed to launch with the QX9770 in Q1’08. Again, X48’s most notable new feature over X38 is support for 1600MHz FSB operation. Officially according to Intel the X38 chipset is capped at 1333MHz FSB speeds, although obviously ASUS, Gigabyte, and others have developed X38 motherboards that are capable of hitting, and more importantly, fully supporting FSB speeds of up to 1600MHz without voiding the motherboard’s warranty (of course, if you go beyond 1600MHz, that’s considered overclocking).

The X48 chipset also supports DDR3 speeds up to 1600MHz. If you recall, the X38 chipset is technically limited to DDR3 speeds up to 1333MHz.

Intel is positioning the X48 chipset as a replacement for X38 in the high-end enthusiast segment. Intel expects the QX9770/1600MHz FSB/1600MHz DDR3 platform to outperform QX9650/1333MHz FSB/DDR3-1333 by around 3-7% on average, and of course the new X48 platform will continue to support Intel’s Extreme Memory Profiles, as well as the Intel Extreme Tuning Utility.

To see how the new QX9770 CPU performs, we’ve included benchmarks of this CPU in this article, and we’ll be taking a more extensive look at its performance at stock speeds as well as overclocked in a dedicated article shortly.

System Setup

Page:: ( 4 / 9 )

Test System submitted by AMD:

AMD Phenom 9900 (2.6GHz)
AMD Phenom 9700 (2.4GHz)
AMD Phenom 9500 (2.2GHz)
ASUS M3A32-MVP Deluxe
ATI Radeon HD 3850 w/Catalyst 7.10 Press Driver

Our testbed

Intel Core 2 Extreme QX9650 (3.0GHz)
Intel Core 2 Q6700 (2.67GHz)
Intel Core 2 Q6400 (2.4GHz)

ASUS P5E3 Deluxe Wi-Fi AP

2GB Corsair TWIN2X2048-6400C3
2GB OCZ DDR3-1800 Platinum Series

ATI Radeon HD 3850 w/Catalyst 7.10 Press Driver

Windows Vista Ultimate 32-bit

Benchmarks

Windows Media Encoder 9
Enemy Territor Quake Wars Demo
Crysis Demo CPU Benchmark
Call of Duty 4 Demo
Lost Planet DX9 Demo Cave Benchmark
Cinebench 10

Notes

Obviously since I was on the road I couldn’t truck all my game CDs I use for testing with me, so I simply transferred as many game demos as I could to the Phenom system AMD had setup for testing. I only had enough time to run 4 game benchmarks, as well as WME9, Valve’s particle simulation test, and Cinebench.

Crysis/Lost Planet

Page:: ( 5 / 9 )

Lost Planet

CoD4/Quake Wars

Page:: ( 6 / 9 )

Call of Duty 4

Quake Wars

Media Encoding/Rendering

Page:: ( 7 / 9 )

Cinebench 10

Valve Particle Simulation benchmark

Windows Media Encoder 9

CrossFire Testing

Page:: ( 8 / 9 )

Conclusion

Page:: ( 9 / 9 )

Performance = IPC x Clock speed

With Phenom topping out at just 2.3GHz, clearly Phenom is playing catch up to Core 2 in terms of clock speeds. This limits Phenom’s performance potential. But based on our gaming benchmarks today, it also looks like they’re behind in IPC as well: at equal clock speeds Core 2 generally ran faster than Phenom. The only exception to this was our Windows Media Encoder 9 test, where the processor’s clock speed played a greater role than IPC in performance.

We’re going to reserve final judgment on this IPC topic until we can run more extensive tests on a Phenom CPU on our own testbeds, but if the benchmarks we presented today turn out to be true and AMD is indeed playing catch up to Intel in IPC, then Phenom has no chance of catching up to Core 2 in performance. They’re just behind too much in clock speed, and apparently IPC as well.

As a result, AMD is going to have to compete with Intel on price if Phenom is going to have any chance of succeeding in the marketplace. That means cheap Phenom CPUs for the general public.

Here the situation looks equally gloomy for AMD though. Measuring in at 285 square millimeters, Phenom is actually slightly larger than Intel’s quad-core Yorkfield CPU, which sports a 214mm2 die size. Assuming equal yields, this means that Phenom is more expensive for AMD to produce than Yorkfield is for Intel.

When you combine that with the low prices AMD is going to have to charge for Phenom, the end result is that AMD is going to be making very little profit margin off their Phenom line of CPUs. Or at least, their margins will obviously be lower than Intel’s. There’s only a couple of ways for AMD to fix this; they can either get the performance up to a level befitting of a $300-$500 CPU, or they’re going to need a smaller manufacturing process.

AMD won’t begin to transition to 45-nm until next year, and we’ve been told that faster processors won’t really be hitting until Q1’08. By the end of the year AMD plans to offer a 2.3GHz Phenom Black Edition with an unlocked clock multiplier for the enthusiast crowd priced “comparably” to the Phenom 9600, which carries a $283 list price. Then, in early Q1’08, we’ll see the debut of a 2.4GHz Phenom 9700 priced below $300. And finally, towards the middle to end of Q1’08 AMD will unveil a 2.6GHz Phenom 9900 priced below $350.

Of course, given AMD’s current track record when it comes to launches you obviously should take all this with a grain of salt. Today’s Phenom “launch” is clearly a paper launch designed to appeal to the financial community. In actuality AMD doesn’t even have enough parts available to seed the press with samples. That’s about as bad a sign as it gets when it comes to availability: if we can’t get our hands on CPUs, it’s doubtful that the general public will be able too either. We’re also not aware of a single Tier One system vendor that will be shipping Phenom PCs on launch day.

In our opinion, today’s Phenom “launch” should have been pushed back until AMD was actually closer to delivering Phenom processors in volume. Flying a bunch of press to Tahoe so they can run benchmarks for a few hours isn’t enough to justify this launch. AMD’s current Phenom prices are a bit out of line with reality as well: Intel’s Core 2 Q6600 officially lists for $266 in quantities of 1,000 CPUs. That’s $17 less than the Phenom 9600, and the Core 2 Q6600 is clearly the faster CPU overall.

The star of today’s show is AMD’s 790FX chipset, which by all indications is a tremendous product with lots of potential. 790FX should have had the spotlight to itself today.

Instead everyone will probably be buzzing about Phenom’s disappointing showing in benchmarks. As far as we can tell AMD’s woes with Phenom aren’t manufacturing-related either. In other words, the CPU isn’t scaling because of an issue dealing with manufacturing (i.e. leakage, poor yields, etc) rather it’s the basic design of the CPU itself. The architecture just doesn’t seem to scale well to speeds of 3.0GHz and beyond. AMD needs to get this issue figured out, and they need a fix ASAP. If they don’t figure something out soon, they’ll never get their margins up and they’ll continue to lose money each quarter. Meanwhile Intel is basically toying with AMD, they could release faster Yorkfield and Kentsfield processors today if they wanted, but as long as they continue to dominate in performance, they simply haven’t felt the need to do anything.

This is the exact same position AMD was in roughly two years ago – boy how quickly things have changed…