Intel has released a chipset with wider capabilities compared to the previous top models of the i845PE/GE line; it's also an ideologic predecessor of the desktop series of the Springdale chipsets scheduled for the spring next year. Nevertheless, the newcomer is not a direct competitor for the existent chipsets, except maybe the i850E. Let's look into the matter. First of all, we will compare capabilities of the E7205 with the current Intel's solutions.
Intel E7205 chipset (aka Granite Bay)
What wasn't changed. As a south bridge (ICH by Intel) they use the same 82801DB (ICH4), which is connected with the north bridge (MCH) by a hub interface with the same bandwidth of 266 MB/s. (Although the speed characteristics of the protocol of data transfer via the interhub interface remain the same, the interface used electrically is compatible with the Hub Interface 2.0 which will be used together with the ICH5.) The ICH4 can't boast about its characteristics as it doesn't support either ATA133 or Serial ATA (it will be added in the ICH5) or FireWire. However, there is an integrated USB 2.0 controller, and the rest can be added by using external controllers on the board.
The north bridge was redesigned considerably. As the market offers only Pentium 4 processors with 400 and 533 MHz buses, only these frequencies of the processor bus are supported in the E7205 (it's quite obvious that the Hyper-Threading technology is detected and enabled automatically); the chipsets of the next year though will acquire support of an 800-MHz bus. The memory controller has already undergone severe changes. Look at the photo above showing the back of the MCH chip and you will see that the number of pins has increased markedly relative to the i845PE/GE (1005 against 593). It's not surprising as the E7205 is just a limited functionality variant of the server (or hi-end workstation) chipsets line which includes the E7505. Hence the high price of the chipset and a complicated layout of the boards meant for it (all the developers have to use a 6-layer PCB design though Intel promises to get a 4-layer one by launch of the spring series of the chipsets).
Coming back to the memory controller, I must say that additional 250 pins in the MCH E7205 are on account of the input/output of the second memory channel. So, for the launch of Pentium 4 processors with a faster bus whose throughput (6.4 GB/s) will be much higher than that of any DDR memory type having a chance to be standardized, Intel started switching the memory controllers in its chipsets to the dual-channel mode. Note that the memory channel are independent, i.e. usage of any one doesn't require usage of the other (like in the NVIDIA nForce/2, contrary to the i850/E) - in practice you can insert only one or two different memory modules getting an operable system, though of a little inferior efficiency. At the same time, usage of two identical modules will let using the dual-channel memory access mode, so the operating speed in such mode will...
...won't be very impressive. The DDR333 memory supported by the Intel chipsets is useless here, not to mention the DDR400. In this case Intel stakes on synchronized memory and FSB frequency, so using Pentium 4 processors with 400MHz bus the memory will run in DDR200 mode, and in case of the 533MHz bus - in DDR266 mode. This will ensure a high theoretical memory throughput of 3200/4200 MB/s that is not an absolute record though, so making very curious to compare in real conditions the tandem of E7205+2xDDR266 with the i850E+PC4200 RDRAM which has the same theoretical performance. Note that the chipset supports only synchronous operation even in the one-channel mode which makes impossible to estimate the E7205+1xDDR333 that looks very close to the i845PE. (The future chipsets of the Springdale line will possibly allow to change the FSB-to-memory bus ratio, and will require the dual-channel access to DDR400 for the maximum performance.) The memory controller also supports registered modules and modules with ECC (that's not surprising due to the chipset's targeting). The maximum allowable memory size is 4 GB.
It's the first time when an Intel's chipset supports the AGP 8x. This feature makes no troubles, but as far as advantages are concerned... The AGP 8x mode boosts the maximum data rate on this bus up to 2.1 GB/s (twice compared to the AGP 4x), but the need in it is arguable. On one hand, the latest video accelerators having (in the minimal configuration) 128 MB memory onboard don't need a higher rate of data transfer via AGP as the textures (which take the most part of the bandwidth) are stored in the local memory, and 128 MB is enough even for the newest games as our tests and many others tests show. On the other hand, junior cards of the budget accelerator families are often equipped with only 64 MB of memory, but this size is usually also sufficient as their functional limitations are a real bottleneck for the same modern games (that only may need more than 64 MB).
Certainly, games are not the only use of modern functional systems, but, for example, in the review of Professional 3D Accelerators in 3ds max 5 we revealed no gain from the AGP 8x, and I can say that the tendency to equip more efficient cards with more memory makes acceleration of the AGP bus useless. However, it doesn't mean that there is no effect of the AGP 8x or that this mode will never come in handy. The matter is that it doesn't make sense to pay through the nose for a technology the benefits of which you don't realize.
Here are the brief characteristics of the newcomers:
As the ASUS P4G8X Deluxe performed better than the other E7205 based models (only a bit higher, but through all tests), we chose it for comparison with the competitive chipsets as E7205 reference (so the E7205 scores on the diagrams demonstrate the scores of this model). One-channel mode was tested only to estimate the performance drop - that's why we don't compare these scores with any others (because I can't imagine that an owner of an expensive E7205 would use only one DDR266 memory module).
The low-level memory tests will let us know what to expect further.
All the E7205 based models go on a par, the difference is 2% at most. It's interesting that the read speed of the E7205 in the one-channel mode is 2/3 of the speed in the dual-channel one, and this is also the maximum performance gap of the modes. As for comparison of the main solutions, there are two characteristics of the dual-channel DDR266 which can be marked: a very high data read speed (it bests all the competitive solutions including those chipsets from other manufacturers which support DDR400) and a modest write speed (it outperforms other DDR solutions but much worse compared to the dual-channel PC1066/PC4200 RDRAM). The stream execution of typical operations in the Wstream test proves this fact.
The video encoding into MPEG4 and the archiving with large dictionary size are our traditional tests for the memory speed in real conditions. The scores are very similar, and we will comment them together. First of all, the E7205 outscores the i845PE with DDR333 by 8-18% depending on the character and algorithm of the test. Secondly, the E7205 also turns to be better even in the one-channel mode demonstrating advantages of the synchronous chipset in some specific conditions. But the crown still belongs to the old speed king, and in the WinRAR the advantage of the i850E (we mean Rambus memory) reaches weighty 8%.
3ds max is an application appropriate for workstations but the final rendering speed is determined there only by a processor, that is why all the participants are on the same level.
SPECviewperf is a test suite that emulates operation of professional 3D applications, and this is the right benchmark for workstations. We are showing only one test results because the others has the same situation or the difference between the runners was unnoticeable. Besides, this test is oriented toward professional 3D accelerators, that is why it's not entirely correct to judge by a system with a powerful but still game-level video accelerator. The i850E and E7205 make operation of the 3D accelerator harder, with the i850E being a couple of percents ahead.
In the synthetic gaming 3DMark benchmark in 1024x768x32 the layout reminds the SPECviewperf; well, the whole load is on the shoulders of the video accelerator, and only the considerable speed gap allows for a marginal (3-4%) advantage for the two dual-channel chipsets.
In the games nothing unexpected is noticed: in low resolutions the difference between the boards reaches 10% with one of the swiftest 3D accelerators. But in more real game conditions the gap falls down to 5%, and it disappears in 1280x1024 with all capabilities of the modern accelerators used. Nevertheless, the i850E and E7205 take the lead where the difference is still tangible.
In closing I must say that the dual-channel E7205 is really fast enough, so high prices for the E7205-systems earn that money (almost equivalent to that of the i850E based systems), because these two chipsets take the lead in all our tests outstripping the tandem of i845PE+DDR333 by 5-20%. At the same time, their scores can differ only by a couple of percents or even less than one. However, the price is still quite high; that is why you have to think hard before ordering such systems. The situation is the same as with the AGP 8x: buy it only if you realize what benefits and where you are going to obtain. Also remember that Intel is not the only company offering high-performance solutions - in particular, the SiS648 coupled with the DDR400 runs very close to the i850E (sometimes even outrunning), and the system costs much less. However, Intel guarantees operability of the certified components in the tandem.
It's really very difficult to compare the newcomers
based on the new chipset. The speed difference doesn't exceed 2%. By the way,
the results of the Gigabyte's model are not maximum that can be reached without
overclocking: the board was tested with "Top Performance" disabled because enabling
this option makes the FSB frequency lift up by 5% (apart from minimized memory timings
and other things) and we considered that it could have been an unfair play. If
your choice is based on functions, it would be also very difficult to choose.
One one hand, the additional components increases the price much. On the other
hand, there are some users who would prefer "All-in-One" instead of saving on
everything. In the long run it's for you to decide, that is why you'd better go
to the table with look through the summary parameters.
Write a comment below. No registration needed!