AMD’s Spider platform. CPU + chipset + video card

Written by Ingeborg Schiffer. Posted in Motherboards. 12 views

Introduction

Today, AMD officially unveils the new “spider”platform. Properly read, what they knew only of the major competitors of Intel and there mainly only from the mobile sector, is AMD now in the desktop sector to: the connection between the processor, chipset and graphics card to a fixed network.

Let’s start with the CPU. With the new Phenom models AMD introduces the first native quad-core processors for the desktop area. The technology behind it is on the whole of Barcelona. Main difference is the socket corresponding to the requirements of the new AM2 +. Although the Phenom processor has to remain above 940 pins and the operation also in mainboards with the older socket AM2 AMD’s information to result should be possible. To develop its full capabilities, however, he needed the socket AM2 +.

The appropriate chipset forms the second part of the platform next to the CPU. Hence the term “7 series” begins with a seven, the first digit of the three-digit designation of the new chipset family. At the beginning are three versions on the market, all of which are paired with the old SB600 Southbridge. All three chipsets support the new PCIe 2.0 standard and are perfect companions for the new 3000s graphics cards from in-house.

The third of three parts of the platform we were able to illuminate already in detail, the new HD-3000-graphics cards series (computer base test).

In the following article, we want to illuminate the new platform from AMD more. On the “benchmark”Festival in Warsaw, in which we participated, we could also complete some benchmarks ready assembled systems to gain a first impression. Unfortunately no own benchmarks were installed and lack of comparable hardware also no comparison measurements could be carried out, why are the benchmark values listed in the article with extreme caution to enjoy. Due to lack of processor, we can do still not own values at this stage, because AMD so far has left no complete systems or individual CPUs for a test us. For these reasons, this article can venture only a very cautious view of the new Dreisamkeit from AMD.

CPU: Phenom

Already in November 2006 the first quad-core processor for the desktop area saw the light of the world – the chip had code-named Kentsfield and came from the House of Intel. Well one year AMD wants to confront the competition with a quad-core own creation called Phenom (codenamed Agena).

Fundamentally different however is the realization approach: while at Intel two dual-core processor this (core 2 Duo, codenamed Conroe) coupled together on a (multi-chip package, MCP) processor enclosure has, is it a step at AMD – from their own point of view – and offers a native quad-core processor. The Phenom in Austin, Texas has been developed. It is manufactured in the 65-nm process in Dresden. In the course of next year, we would like to then – then already commonplace – on the 45-nm lithography process like at Intel.

The TDP of Phenom is 89 Watt 9500 and 9600. Each individual core has 512 KB L2 cache. In addition, the entire processor has a shared L3 cache 2 MB, which is shared by the nuclei. However, AMD as far as concerned the memory hierarchy has taken hold in the bag of tricks. So, the L1 cache prefetcher can request data without the usual going through the L2 and L3 cache data from memory.

The processor’s internal data paths, such as the execution units for SSE and FPU doubled so far 64 bits to 128 bits (as in Intel’s Core microarchitecture). Thus, SSE instructions can now be edited within a measure. AMD’s architecture allows parallel execution of two SSE commands – Intel is ahead but still a upmanship with three execution ports for SSE (not all ports can perform all SSE instructions). The instruction fetch buffer of Phenoms was expanded to a width of 32 bytes. Thus, more x 86 commands that are between three and 15 bytes long, in the instruction Fetcher, the first part of the command pipeline fit. This is also when the 64-bit extended commands (on average about 4.5 bytes) make sure that the continue to triple superskalere processor is supplied with enough commands to process. Intel’s pipeline-frontend cache reads only 16 bytes of the instruction (since the Pentium Pro) and thus highly is in danger, especially for 64-bit code, not enough to work can provide the 4 x super skalaren.

Also the branch prediction was improved by AMD and equipped with a new logic. An accurate branch prediction is important, because modern processors, multiple instructions in parallel process and a jump (branch), which not as predicted acts makes obsolete the other instructions in parallel running on suspicion. The virtualization technology has been improved so that the hypervisor faster can make its memory pages the guest systems available. The speed of the memory controller for AMD continue to integrated rose of DDR2-800 on DDR2-1066 so far. The function block contained in the Silicon supports also DDR3-1600 MHz (effectively). This AMD makes however still no use.

With the Phenom, AMD makes the step to industry-standard HyperTransport 3.0 (HT 3.0), which offers significantly higher bandwidth. This high speed connection can theoretically transfer up to 20.8 GB / s. The Phenom can use them according to AMD currently up to about 18 GB/s and these values will rise with higher clock rates. For all this, the Phenom backward compatible with the older HyperTransport 1.0 only about 6.4 GB / s remains bandwidth.

AMD has made also to the energy requirements of Phenom thoughts. As a “Split-Power-Lane”, AMD is the ability to provide the memory controller and the cores with two independent tensions. The C1E power state, which now also supports the Phenom, AMD meet the valid since the middle of the year energy star 4.0 requirements.

Now, the Phenom for many users of price comparison services is not a newbie. For days the new CPUs are listed by different traders, of course without any availability. However, these early offers pricing is much more interesting. While currently (18.11.) the Phenom 9500 219,-euros is offered and the Phenom 9600 will cost proud 245,-euro, today introduced Phenom 9700 is (see next paragraph) presented for 259,-euros. Without many words we want to show presented pricing by AMD simply that the press at this point.

In the end we want readers not conceal, why the planned and just mentioned Phenom 9700 according to AMD will probably only come in January 2008 on the market. Blame is an errata, an error in the processor. Such mistakes are nothing new to such a complex object such as a microprocessor. Also the old Pentium was plagued by such an error, and many bugs that mean a problem only under very rare constellations are slumbering in core 2 and Athlon 64. According to AMD, the error only in extremely rare cases under full load and at high clock frequencies of processor occurs. On request, Dave Everitt told us that there should be an error in the TLB of the L3 cache.

The TLB provides high-performance conversion – implemented in hardware as a cache – the virtual memory addresses used by applications in physical memory addresses. AMD wants to provide a tool and/or BIOS update with the TLB of the L3 cache can be turned off. Disabling the TLB would lead to a loss of performance of approximately ten percent.

The chip set: 7-series

Let’s take a look at the new chipset family. It supports the new PCIe 2.0 standard and HyperTransport 3.0 through the Bank. Only for use with PCIe lanes, there are important differences: while the 770 only a graphics card supports, can the 790X of those two and up to four graphics cards raise the 790FX.

The 790FX (codenamed RD790) has a total of 42 PCIe lanes PCIe 2.0 standard – AMD speaks here of engines – administered by eleven controllers. Thus, plenty of opportunities to distribute the PCIe lanes available are the motherboard manufacturers. Compared to the previous solution with PCIe 1.0 and HT 1.0 performance with the new HT is to increase 3.0 and PCIe 2.0 as an example of Call of Juarez by about 20 percent, so AMD. As a Southbridge still the SB600 come somewhat in the years is used at the moment. Probably from January 2008, in each case in the first quarter of 2008 the new SB700 to the market will come then. This offers almost no new features, but only “more” of the previous connections.

< div >

Also the 790FX is made like the Phenom processor in the 65-nm process. He is the first desktop chipset, which is manufactured in this procedure. According to AMD, the TDP should be at a meager 10 Watts. Even if you’d add the need of the memory controller integrated into the processor for AMD added, you would be well below the needs of the Intel X 38 chipset, which is manufactured in 90nm and approved are happy times well over 30 watts. The BIOS was more divided into individual modules, so that motherboard manufacturers can dispense easily on the overclocking functionality of the BIOS, without compromising other functions. With memory management, there was no dual-channel setting more at least in the BIOS for the Gigabyte motherboards installed in the test system. Instead there are now two modes “Ganged” (connected) and “Unganged” (unconnected). According to AMD dual-channel should provide in the field of games currently corresponding “Ganged” setting more performance. “Unganged” is, however, useful for applications with multiple threads. Multitasking applications, the new memory controller in the Phenom can then appeal to the existing storage bar individually and thus for every single active thread faster data write or read. At the “advance benchmark results” on page 5, you can see the effects of the two modes in the SYSmark.

Also on the software side, AMD has done a lot. The RAID management tool, which so far, image was based on a relatively antiquated version of silicon, has been revised and is designed to provide more convenience especially for inexperienced users. More important however AMD Overdrive, a Übertaktungs-and monitoring tool, is to this quite as counterpart is nTune from NVIDIA. On motherboards with overdrive support, you can raise system performance via slider. In addition to a AutoClock function, the savvy user of all kinds of voltages and clock rates can adjust and check with the built-in stability test. The engineering samples of the Phenom processors installed in the test systems were able to be overclocked by OverDrive while, however, Dave Everitt asked not to increase the VCore. Supposedly stable clock rates should have been reached yet with increased multiplier, processor clock and slightly higher VCore from about 2,700 to 2,800 MHz.

&Lt; div > Photo Gallery “AMD OverDrive” (20 pictures)

Let’s take another look at the placement of the single chipset versions in the respective market segments. AMD pushes for mainboards – similar as with the graphics cards and processors – with aggressive prices in the market, to to make the major competitors of Intel. You may be curious to see whether I will counter Intel own price reductions. Interestingly AMD system price compares a 2.3 GHz Phenom with an Intel Core 2 Quad 2.4 GHz. Also, the price difference between (AMD system) DDR2 and DDR3 (Intel system) is to keep in mind.

The video card: HD 3000

We want to take this article not more deeply on the part of graphics cards, has this been treated deep having already in our review of the ATi Radeon HD 3870. All test systems with a CrossFire group were provided from two HD 3850 in Warsaw. Systems with four HD 3850 graphics cards were shown additionally. This quad CrossFire, or also CrossFire X, will be supported by the then available catalyst from January. Interestingly, then also a such CrossFire X composite with DirectX 9.0 is possible, as was the current demo of 3DMark06. At the end of the event, AMD showed even a running system with two HD 3870-X 2 cards, so CrossFire X with only two cards.

&Lt; div > “ATi Radeon HD 3870 X 2″ Photo Gallery (5 pictures)

Energy requirements

To demonstrate the significantly lower energy requirements of their new platform in contrast to its predecessor, AMD had presented two special systems, on which a VC1-encoded HD video was shown. A power meter was already activated, both systems, as it is also used in our tests. While the “old” system with an AMD Athlon 2900 XT graphics cards was 64 X 2 6400 + and two HD, was the spider system equipped with a 2.4 GHz of fast Phenom, the 790FX-Chipsatz and two HD 3850. Especially the UVD engine of the new HD 3850 cards comes to good, which drastically reduces the CPU load when playing a HD video and thus power consumption in this discipline the Spider platform.

“Old” system

< div >

Spider platform

< div >

Preliminary benchmarks

As already mentioned, the assembled journalists in Warsaw on the approximately 40 test systems were allowed to run only pre-installed benchmarks. In teamwork with other German press colleagues we could complete some benchmarks, because the time to free measure was limited.

However, it is installed special caution in the interpretation of the values. So, the results are not comparable with other previously measured values at the time, because it has acted to completely different platforms and equipment. The measured values give a rough orientation of the HD 3870, which you can read in our graphics cards article however. However, there is still on a PCIe-1.1 based system was measured. Here be able to show only later test exactly how big is the influence of the new PCIe 2.0 interface.

AMD Spider platform test system

Processor: 2.4 GHz Phenom, engineering sample
Motherboard: Gigabyte MA790FX-DQ6
Video card: 2 x HD 3850 in CrossFire together
Memory: 2 x 1 GB Corsair Dominator CM2X1024 8500C5D
Hard drive: 150 GB Raptor, Western Digital
Power supply: Akasa 850 Watt
Operating system: Vista 32-bit
Monitor: Mirai 22 “TFT

3DMark06