1st REVieW: Intel PCI Express and 775 CPU get going over

[apoppin]

Review Grantsdale I915 board in the house

GUESS THAT EVERYONE IS wondering how fast the new Intel families will perform, with a new chipset, an LGA 775 CPU and the PCI Express interface.
In the most complex operation, we managed to get access to a working 775 motherboard, CPU and PCI Express card and we managed to do this quite some time before Intel is actually ready to announce it. Beware that this is our first experiences with this platform. We're still unclear about DDR-2 performance because this board we've reviewed uses DDR-1 memory.

We were lucky to have Intel's new platform in our hands for quite some time and we have no obligations to the firm to keep our information quiet. You have to bear in mind that this is a demo board, an early sample that we managed to secure, but with a sample board, a final CPU, and with a Gainward Gainward PCX5900 card based on Nvidia's reference design.

We won't get into many details on motherboard as we might compromise the people that helped us to see the future technology today but we will give you much more details about its performance and what you can expect from it.

For this platform you have to use a 24 pin power supply unit that costs quite a lot and right now is a very rare bird. It's something that workstation PCs and servers are using but now Intel wants to make us all these new PSUs. Someone even told me that it is possible to make a 20 pin PSU work on the motherboard, but this involves a high risk of roasting the board.

The second interesting detail is that you have to change your cooler as none of the coolers that we had for the current P4 generation will fit on the motherboards. We tried a water cooling kit that we have but nothing that you currently have will work with that socket. The dimensions are a little bit different from socket 478 and you need a new cooler. A great move for Intel and its cooling partners will love it. They will sell more units. The Intel reference cooler is a scary device, a huge rounded tower with a big propeller on top but it works. We can't say it's not hot under the hood, but it works fine.

We modified a Titan Athlon cooler and it actually was able to do the job, but we cannot say that this looked very nice. The machine was stable most of the time or should I say for all the testing, apart from one hang up in the Halo game in 1600x1200 resolutions. Again, note you won't be able to use any existing water cooling or phase changed systems as it simply won't fit on to the motherboard holes. You will have to buy new water blocks once they are available to place on these CPUs.

The graphics card is 16xPCI Express and theoretically should speed up things compared to AGP. PCI Wxpress motherboards will end up with small PCI express 1X slots for future connectivity devices. We heard that there will be quite a few devices that might benefit from more then the 133MB/S that a normal PCI can offer. We see some opportunities for TV tuner cards, Gigabit LAN cards and possible even sound cards that will use this bus.

Chipset
Grantsdale P and G will come with some new and some old features. The only real big difference when we are talking about Grantsdale P from I865 or even 875 boards is PCI Wxpress. Grantsdale G will come under the commercial name i915P and i915G for the discrete and integrated solution respectively. Most of the 915P/Grantsdale boards will use only DDR-1 400/333 MHz memory even though the chipset can support DDR-2. As with previous chipsets such as the 865 and 875, they use a dual channel memory controller something that definitely gives a turn of speed.

It makes little sense to have a Grantsdale P chipset with DDR-2 support since that would be much closer to the Alderwood i925P chipset that will have some extra features.

The system we looked at only has 533/800 MHz FSB support, although future revision will support the 1066MHz CPUs that will emerge later this year. Boards will support PCI-E 1X slots as well. As for the south bridge - the IC6R/W - that now features as many as four native S-ATA and at the same time supports good old IDE drives.

For audio, Intel will introduce its Azalia audio. This chipset uses LGA 775 CPUs that come in a special socket. The i915-G comes with additional integrated graphics and something called stable image quality but apart from that is the same as 915P. The 915P will be able to support 1066MHz front side bus speeds.

The i925-X will also come with FSB 1066 support and ECC memory, while the rest of the features will be similar to the i915-P. This board, like Intel's PAT stuff last year, is aimed at enthusiasts and for workstations.

CPU
Intel's Land Grid Array 775 socket doesn't use pins, is much cheaper than PGA and BGA (ball grid array) but might cause some future problems.

The 775 CPU has 775 connections and the socket itself has 775 pins. The pins from the socket just touch the CPU connection which is secured with a special socket retention mechanism. It's easy to use but also easy to break your CPU or the socket. There have been some problems and the board that we tested had some bent pins, so we had to straighten them first before we inserted the CPU. Then it worked fine.

The CPU is based on the ES Prescott core clocked at 3.2 GHz.

Test
We used a Gainward PCX 5900 card clocked at 350MHz core and a modest 550MHz memory. In our motherboard we used 3.2 GHz LGA 775 CPU and plugged two 512MB modules of TwinX Corsair DDR 550 PC4400 memory and used a Maxtor 80GB 2MB hard drive.

It was very hard to compare these results with a 478 Pentium 4 system due to the difference in the graphics card but we noticed that there was no epochal increase or decrease in performance. It's just a new Pentium 4 with a new socket and we didn't notice any significant performance increases with Sandra or in any other system performance test

Conclusion

We have to admit that we were very impressed with the stability of this early demo board. This platform looks good and it will sure have some benefits but it will take some time to introduce a new BIOS that will take advantage of its features. We are sure that Intel can squeeze more from this platform when they are ready to introduce it. This should happen after Computex so they still have more then a few weeks to improve things. As for PCI Express graphics, it works fine and I don't think it's any faster than a Geforce card clocked to the same speed. It can actually end up slower because of the HIS, the Nvidia chip who translates instructions from native AGP chip to PCI Express.

It's very encouraging to know that all is ready now so it's up to Intel just to press the GO button. Intel has had some criticism from mobo manufacturers because of the fragility of the socket, and they'll be expected to handle returns, not the chip giant. Let's hope that this new 775 Prescott ES will actually give some performance once when the chipset is ready and when they are ready to ship them on June 20th.

The platform looks good but it's not something that you would kill for. Right now I don't see the benefit to this platform apart from the fact it will enable you to use new 775 CPUs and PCI Express graphic cards, and possibly the audio. PCI Express is the future, whether we like it or not as the industry has to dance mostly to Intel's tune.

 

[remagavon]

Hm, they say it isn't faster/slower, which is odd considering IIRC the initial performance penalty of the new chipset was going to be due to the added latency of DDR-2 and this board uses DDR-1.

Either way I like the smaller pins that are hopefully less susceptible to bending.

 

[apoppin]

Originally posted by: remagavon
Hm, they say it isn't faster/slower, which is odd considering IIRC the initial performance penalty of the new chipset was going to be due to the added latency of DDR-2 and this board uses DDR-1.

Either way I like the smaller pins that are hopefully less susceptible to bending.

i
thought the pins were MORE problematic -

The 775 CPU has 775 connections and the socket itself has 775 pins. The pins from the socket just touch the CPU connection which is secured with a special socket retention mechanism. It's easy to use but also easy to break your CPU or the socket. There have been some problems and the board that we tested had some bent pins, so we had to straighten them first before we inserted the CPU. Then it worked fine.

Of course they 'only' had 800FSB, but this is something i would NOT have waited to UPgrade for.

i'd still be WAITing . . . it's (still) gonna be months before this is mature.

 

[LTC8K6]

I thought LGA didn't use pins at all?

 

[ForceCalibur]

I didn't understand half of that poorly written/organized article.

 

[apoppin]

Originally posted by: LTC8K6
I thought LGA didn't use pins at all?

Here's what confused me:

CPU
Intel's Land Grid Array 775 socket doesn't use pins, is much cheaper than PGA and BGA (ball grid array) but might cause some future problems.

The 775 CPU has 775 connections and the socket itself has 775 pins.

huh?

semantics?

someone please elaborate (so i don't have to look it up).



:roll:





EDIT (i think) i get it . . . the SOCKET has the pins.

 

[LTC8K6]

Okay, the chip is pinless.

Here is another look at the arrangement.

LGA775

 

[apoppin]

Originally posted by: LTC8K6
Okay, the chip is pinless.

Here is another look at the arrangement.

LGA775

Thanks for the link . . . i figured it out in my edit just as you posted . . .

LGA stands for Land Grid Array and refers to the flat contacts at the processor. The pins are now part of the socket.

So, What's the difference in practical terms?:

Thanks to the absence of pins, the processor is less delicate. However, the socket is now susceptible to twisted or broken pins.

 

[erikistired]

the mobo manufacturers foot the bill for broken/bent pins instead of intel.

 

[apoppin]

Originally posted by: fisher
the mobo manufacturers foot the bill for broken/bent pins instead of intel.

That IS a practical difference.

Clever of Intel.

:roll:

edit: in the other camp (related news):
Via close to sampling PCI-Express Athlon 64 chipsets

CHIPSET ROADMAPS seen by the INQUIRER show that the firm is close to sampling its K8T890 and its K8M890 chipsets, designed for the Athlon 64 and Opteron microprocessors.
The chipsets are PCI Express ready and the K8T890 will sample in June with mass production slated for Q3. The K8M890, which includes Deltachrome IGP with DX9 support will sample in the third quarter, with mass production in the fourth quarter of this year.

The K8T890 uses 1GHz Hypertransport, 2GT/second, and includes PCI-Express (16+4) and AGP 3.0, as well as the Ultra V Link. The Deltachromatic version has 16+2 PCI Express, and a l DVI/TV out interface.

On the Intel chipset front, Via is already sampling the PT890, which supports 400MHz, 533MHz and 800MHz system buses, has dual 64-bit channels, will support DDR-2 and and DDR-1, and PCI Express 16+2 and AGP 3.0. Support for 1066MHz CPUs has not yet arrived on Via's roadmap plans.

 

[LTC8K6]

LGA keeps the electrical path shorter and allows higher pin counts in a smaller space, or so I've read.

The "pins" in the socket are slightly springy to facilitate good contact.

I guess we are going to have to stick something through the hole in the cover, like the eraser end of a pencil, to hold the processor in place correctly when we latch that thing.