April 4, 2005

Intel Previews Desktop Dual Core Extreme Chip - Josh

Intel has provided a handful of reviewers samples of their upcoming Pentium EE 840, which is the dual core Pentium 4 running at 3.2 GHz.  Alongside this release is the i955X chipset, which offers a significant upgrade from the previous i925X.  The first thing that I noticed from these reviews is that the chip runs incredibly hot.  Not only that, but it requires the use of a ATX 12v ver. 2.01 power supply (note the 2.01).  It is essentially a EPS 12v server level power supply.  HyperThreading is still enabled on each of the cores, so it can address four threads at once.  Intel reworked the memory controller on the i955X, and it turns out some impressive performances (as well as unofficially supporting DDR-2 800).

Not every application supports multi-threading, and so in many benchmarks and games the 840 performs lower than the top clocked 6x0 series of chips.  In applications that do incorporate multi-thread capabilities, then the 840 shows some significant gains.  This EE will not be for everyone, but it does appear to be the first dual core product to be released in significant quantities for the X86 market.  Apparently only a handful of sites were given these processors for testing, so the ones that do feature it are getting hammered.  Legit Reviews, Trusted Reviews, and Hexus are currently the three I know about.

Today was also supposed to be the day when AMD released their Venice core products, but so far only X-Bit has anything to say about it.  Perhaps a bit later this week?  I am not sure, but supposedly a couple of sites I regularly visit have some E3 rev. cores on the test bench that we should be very interested in.

My initial overall take of this is that Intel did their best to bolt two Prescott cores together, and in so doing created one hot processor.  The guys at Legit Reviews had to use watercooling to get it to work as it should, as the supplied heatsink and fan would often cause reboots in intensive applications.  Still, these are not considered production ready parts as of yet, but I am very curious to see what Intel will do to successfully bring this product to market.

April 1, 2005

State Comments and News - Josh

Thanks to the latest SO3D article, as well as some other articles and front page postings, March was the best month here for around 3 years.  Somewhat sad to consider, especially when I look back at some logs around 1999 and 2000 (a lot of traffic back then).  Oh well, life goes on, and I am working hard to keep the momentum going at this site.

It seems that I am well off on my speculation that R520 was taped out in February.  From what I am gathering, the tape out occurred near the end of 2004.  So, most likely there are quite a few working R520 chips, and with ATI's push to only release product when they can adequately supply the market, we are still on track for an early Summer release.  Something else that was pointed out to me is that TSMC only ran 130 nm Low-K out of one Fab on 200 mm wafers.  TSMC's 90 nm runs out of several fabs, some of which feature 300 mm wafers.  I don't have the exact specifics on how many Fabs and their wafer sizes, but overall it appears as though there will be far fewer production constraints for ATI with this product (as compared to the 130 nm Low-K parts).  I for one am really looking forward to seeing what ATI will produce with the R520.  If they get their yields and speed bins in order, then we could see a really fast and impressive part!  Now, just using a more advanced process does not automatically mean that it will be a better product (just look at NV30 vs. the R300 for proof).  However, I highly doubt that the R520 will be the disaster the NV30 was!  If the R520 does what it is supposed to do, and runs well on that process, then ATI could have a significant leg up on NVIDIA with this next refresh.  Time will tell though, and it looks as if ATI already has two large parts on TSMC's 90 nm (R500 and R520) that are probably being mass produced as we speak.

The G70 from NVIDIA is increasingly looking like a 110 nm product utilizing 24 pixel pipelines and 10 vertex shaders, but no word on how many ROPS are being used (though I would guess 16).  Something that I was really wrong about was that the GeForce Go 6800 Ultra is not in fact a G70 variant, but rather it is the NV42M.  With NVIDIA's seeming expertise at TSMC's 110 nm process (with the NV44 running at 500+ MHz, and the much larger mobile NV42M running at 450 MHz), I would not be surprised to see the G70 running at 500 MHz as well (if not a little more).  I think NVIDIA will still have a competitive part, but if ATI can get their R520 running at a significantly faster speed (like 100 MHz more) as well as tweak their design (and the R520 is supposed to be a very new and optimized architecture supporting SM 3.0), I think that NVIDIA has a good chance at having the 2nd fastest top end product.  Again, this is only speculation, and we won't know for sure until both ATI and NVIDIA release their products.  Still, looking at the situation from far above, ATI may have a good leg up on the competition this round.  We shall see!  If you want to comment on this extra bit of info, or on the SO3D article as a whole, please go here!

One other interesting thought is that NVIDIA has publicly claimed that it won't have a 90 nm part until the second half of this year.  For once NVIDIA is being conservative with their migration to a new process, but from what I am hearing TSMC's 90 nm product is very clean and performs very well.  Now, ATI will definitely be the first to make a massive 300+ million transistor part on this process, and with anything new like this (not to mention complex), bad things can happen.  It is not a slam dunk to produce a part of this complexity on a relatively new process (again, just ask NVIDIA and its use of 130 nm for its NV30).  Still, from all indications that I am hearing, the R520 should be a very high performing part.  Plus, with the amount of capacity that TSMC's 90 nm process looks to give IHV's, I doubt we will see the availability problems that ATI experienced with its X800 series from their release until January of this year.

So does the R520 ring the death knell for NVIDIA if it is a smashing success?  Of course not.  While it is very important to gain mindshare in this market, it is not where the money is made.  Currently holding the mid-range and budget level crown is much more profitable than having the best top of the line card.  This is one area where NVIDIA does hold an advantage.  The GeForce 6200 TC is selling in very large quantities in both retail and OEM circles.  Recent benchmark reports show that the 6200 TC outperforms the competing X300 Hypermemory, sometimes by a large amount.  Now, one area where the X300 HM does excel at is the use of Anti-aliasing (it does beat up on the 6200 TC pretty good in that situation).  Still, in the majority of benchmarks the 32 MB 6200 TC outperforms the 128 MB X300 HM.  When we look a little higher we see much the same situation with the X600 and the 6600, as well as the X700 Pro and 6600 GT.  At each of those price points NVIDIA is showing a goodly amount of performance over its competitor.  Winning in this market is a lot more important than at the $600 price point.  So, NVIDIA does have an edge at this time over ATI in this very lucrative area.  Of course the situation will change when ATI releases their R5x0 value and mainstream based products.

In other news I should be receiving a SLI setup, and I will have some extras in my article that should really catch some attention.  I shan't be going into specifics now, but rest assured my "review" will have something that nobody else has exploited as of yet.  Should be some very good stuff!

March 30, 2005

The State of 3D- ATI and NVIDIA, 1H 2005 - Josh

March has certainly been a busy month here, and to cap it off I have published my latest State of 3D.  This covers most of last year, this current year, and what looks to be happening through June/July (and slightly beyond).  Both NVIDIA and ATI have been very busy as of late, and the amount of rhetoric going on between the PR people for the respective companies is reaching a fever pitch.  I try to cut through some of this and give an objective look at where the industry is, what trends we might be seeing, and how each company is reacting to the other.  Here is a quote:

            When placed head to head with the GeForce 6600 GT, the current high end X700 Pro typically performs slower in a wide variety of games and applications.  It is still a very solid performing part, and for the record it does surpass the 9800 series of cards.  It is a very solid midrange card, but due to the pressure from NVIDIA, it may not get the recognition that its predecessors, the 9600 Pro and XT, got when they were competing with the midrange GeForce FX products.  There have also been initial reports of more fabrication problems with these chips and their design.  This could be one of the largest factors as to why the X700 XT was dropped from the lineup, as the speed bins were not shaking out as they were supposed to.  The X700 cards in peoples’ hands do not appear to want to clock much more than 500 MHz at most, which is the base speed of the GeForce 6600 GT.  The one advantage the X700 Pro has over the GeForce 6600 GT is that many of the X700 models come with 256 MB of memory.  More is always better, and many feel that while the 6600 GT is an excellent performer, it doesn’t come with enough memory to let it stretch its legs on games such as Doom 3 (which in High Quality mode utilizes over 200 MB of texture data).  Other upcoming games will also feature a greater amount of texture data, and will more aptly utilize 256 MB cards.

You can read the entire article here.

Be sure to comment on this article in the forums!

March 29, 2005

News About Town - Josh

The inbox was stuffed to the gills this morning with news items, so I thought I would share a few articles I found interesting.

The Tech Report (note the lack of a dash) throws down the Gigabyte GA-K8VT890-9 motherboard and stomps all over it.  Literally.  I kid.  This motherboard is based on the VIA K8T890 and is definitely aimed at the budget sector (hence is suggested $83 US price).  It does come with a solid set of goodies, including the excellent Marvel PCI-E Gig-E networking chip.  One thing of interest is that apparently there is a new revision of the VT8237 Southbridge that no longer requires a driver disk to install an operating system on a SATA HD.

Ecoustics has put up a speaker position guide, so all those hoping to create a home theater of their own should run over here and take a look at what they have to say.

Those fellows from Madshrimps have taken the Corsair Cool water system and see how well it actually works.  So far everything I have heard about this system is quite good, and a significant step up from some of their previous attempts (anyone remember their first watercooling product that continually plugged up due to corrosion?).  This looks to be an excellent introduction to water cooling, and for not a huge price either.

The infamous British hack and lush, Dave Baumann (I have seen pictures) was sent 3 video cards (and some hush money) from Club3D.  These boards are based on the X800 XL and X850 Pro chips, but to make things confusing one of the X800 XL's has the Rialto bridge chip, making it a AGP card.  This is Dave's typical review, which means that everything is tested exhuastively.

Doug "I'm Really Not Short" Kronlund takes a look at a very unique product, the X-Woofer 3D Vibro (sounds kinda kinky) is a product worn around your neck.  Its entire purpose is to enhance the listening "feel", but Doug essentially feels like the user is in fact wearing a horse collar.  You decide.

As you may or may not know, the Forum is back up and running!  While not a huge amount of people are there yet, I would like to change that fact!  There are a couple of interesting posts so far, so any kind of feedback would be more than welcome!  One of particular interest is in how many readers here actually invest in tech stocks.  Sign up and join in the conversation!

March 28, 2005

AMD and its Upcoming Processors - Josh

It certainly appears as though AMD is very close to releasing its dual core Opterons, and the latest roadmap certainly shows that they plan to release in the 2nd quarter of this year (April, June, July).  Now, supposedly there are some meetings going on at AMD this week that not everyone was invited to, and this could be the showing off of the dual core Opterons, as well as a possible Venice core release.  I know that AMD is very keen on releasing the first true dual core X86 processor, no matter what Intel has to say to the contrary.  While Intel showed lots of dual core silicon at the last IDF, AMD first showed their dual core product in late summer of last year.  As some of you might remember, around the time that Intel released their LGA-775 chips and PCI-E, AMD said that they taped out their first dual core product.  Everything seems ready for an official dual core introduction.  The first dual core Opteron products will run at 1.8, 2.0, and 2.2 GHz and will consume a maximum of 95 watts.  These will be released as the standard Socket 940 chips, and with a BIOS update from the motherboard manufacturer, this can go into nearly any Socket 940 motherboard.

On the desktop side we could be looking at some rather subtle changes.  The Venice core is the codename for the 512 K L2 rev E Athlon 64.  This will replace essentially all of the current 130 nm and 90 nm Athlon 64's.  This is the revision with the improved memory controller, SSE3 support, tweaked SSE/3DNow! unit, and a host of other optimizations.  The first of these cores have already leaked out, and a few lucky people have picked up some Athlon 64 3500+'s and 3800+'sthat are based on the Venice core.  These buggers are clocking upwards of 2.8 GHz on air, and above 3.0 GHz with some more exotic cooling.

There does not appear to be a change at the high end, but the FX-55 will remain the top core for AMD.  Now, whether they substitute the 130 nm SOI/Strained Clawhammer core for a San Diego core (90 nm SOI/Strained SSE3) is not clear to me.  One would expect this to be the case, especially considering the recent scarcity of the current FX-55 in the marketplace.  Remember, AMD wants to get away from 130 nm Athlon 64 production by the end of 2Q.  Apparently April 4 is the date when AMD will release the Athlon 64 4200+, which looks to be a Venice core clocked at 2.6 GHz.  The FX-55 will stay at 2.6 GHz as well, and it will have the 1 MB of L2 to distance itself from the 4200+ (which of course only has 512 K L2).  Not a huge difference if you ask me.  Still, the idea of a San Diego core in the FX-55 is quite nice, and I hope that we will in fact see it.  The strained Clawhammer was a good product, and a good stepping stone for AMD and its implementation of its strained technology (combined with SOI).  Overall, it appears that the effectiveness of AMD's 90 nm SOI/Strained process is edging out Intel and its 90 nm Strained process.

If you want to see how the Venice based 3500+ performs, go here.

March 25, 2005

512 MB GeForce 6800 Ultras Announced - Josh

Both BFG and EVGA have officially announced that they will be bringing forward 512 MB GeForce 6800 Ultra's by the end of April.  Apparently supplies of the necessary GDDR-3 modules are coming off the line at a goodly pace, and it is now possible to provide solutions based on this memory.  BFG will package the 512 MB card with a bundle worth around $60, but I imagine the initial price will be around $600 US.

Now, with NVIDIA and its partners releasing a new high end product like this in the Spring, what does this mean about a possible Spring refresh for NVIDIA?  It doesn't seem to make much sense to release a refresh right at the same time as a new SKU of the 6800 Ultra is available.  Now, the only way I can see this really happening is if the 6800 GT now hits the $299 mark, the 6800 Ultra is $399, the 512 MB Ultra is $499, and the refresh part is $549 and above.  But what are the chances of this happening?  I have no idea.  AGP 6800 GT's are getting close to the $300 mark, and Ultras are starting to go down in price as well.  Where does this leave the G70 parts that are supposed to be the new refresh?

Well, my personal feeling is that the G70 part has already been seen in the wild with the GeForce 6800 Go Ultra.  Here is a NV4x part that runs in a mobile platform at 450 MHz.  Sure, it only has 12 pipelines, but 450 MHz!!!  And it doesn't require heroic cooling?  I would bet that the desktop version of this chip goes at least 500 MHz if not above that, and feature a full 16 pipelines.  So is this the refresh part we have been waiting for?  I have no idea, as details are very scarce.  One thing that is somewhat telling here is that in every preview of the GF Go 6800 Ultra, nothing was spoken about what core this was based on, what process it was using, or anything else that could possibly give away the parentage of this latest NVIDIA product.

March 24, 2005

Thursday Ponderings and Links - Josh

After reading HardOCP's SLI article, I believe that we are now finally seeing SLI as it was meant to be.  I think that there will still be a lot of polishing going on, and we will see definite improvements over the coming months.  One thing that I believe NVIDIA will be doing now is probably release a few more drivers than they previously have been doing.  While it will not be the monthly Catalyst release that ATI offers, they will probably release more official WHQL and non-WHQL drivers for public consumption.  Also, NVIDIA is working with developers to more adequately support SLI in their titles right out of the box.  We really won't be seeing the fruits of this labor for a while, but the push is there.

Now, how does NVIDIA's SLI solution compare to what ATI is supposed to bring?  I really can't make a final say on that until we see ATI's technology, but the information that has been leaked so far looks very promising.  Currently NVIDIA supports AFR (alternate frame rendering) AFR-2, and SFR (split frame rendering), and supposedly there is dedicated silicon in the nForce 4 SLI chipset that helps to speed up the process by shuffling data to where it needs to go (unconfirmed, and no way to really check to see if this is true).  As it is, there is a fair amount of driver overhead on the CPU side to get everything to work as it should.  Still, that overhead is usually overlooked when the player is able to set high resolutions, AA, and AF in their favorite game.  NVIDIA supports 70+ titles right out of the box, but they also have the new Coolbits hack that will allow the user to set profiles for games that are not currently supported.

ATI is planning a multi-GPU solution of their own, but they look to be using a technique called "Supertiling".  All Radeon chips since the 9700 Pro support this kind of rendering, and what it basically does is take the scene and split it up into many different tiles, and each card renders alternate tiles.  This should be a lot more effective in terms of rendering efficiency and workload balance.  Their setup also supposedly allows the use of two different cards, as long as they are "somewhat" alike.  That means that if the user has a X800 XT, and then decides several months down the road buy a inexpensive X800 XL to sit right beside the first one, the XT will downclock itself to XL speeds, and both will work in tandem.  This of course won't work with a X800 combined with a X700, as the overall differences between the two products is too large.  The idea behind this is great, the only issue we see right now is that nobody has seen it in action.  I am unsure when ATI is planning to release their multi-GPU product, but I would say that it will be at least Fall of this year before we see anything concrete from them.  If they do show it in the Fall, then it will be close to the end of the year by the time initial products hit the shelves.  This means that NVIDIA will have a year of product sales and experience under their belts before ATI sells a single product.  Still, the idea of supertiling is very attractive, and if it works as advertised and comes out before the end of this year, ATI has a good chance to capture a good portion of the high end market that SLI currently resides in.

I will have a much better idea about how all this works once I get the SLI setup into my hands.  Currently I am still working on the GeForce 6200 TC review, and that is turning out to be a very impressive product considering its price point.  I have found some very interesting tweaks that can really affect the overall performance of this board.  Also, it is very capable of playing Far Cry at 800x600 with Ultra quality enabled (SM 3.0 rendering), which really is saying a LOT.

Now for links around the web...

Ryan over at PC Perspective takes a look at the Pentium M 755 combined with the DFI 855GME-MGF, which is a more than adequate performer for the desktop.  Though Ryan may be a bit "over enthusiastic" about this combination's high price, it is still an impressive product nonetheless.

Chris and the gang at AMDZone take six of the latest ATX 12v V 2.0 power supplies and flail them with a SLI setup to see who can provide the best power.  All of them perform very well with the dual 6600 GT's, so it essentially comes down to features.  Still an interesting read if you are in the market for a new power supply.

Anand has released his latest opus with "World of Warcraft Performance Guide".  I am not sure where Anand had time to actually play WoW, but he certainly goes through the tweaking.  He takes a variety of video cards and systems and puts them through his own personal WoW torture chamber.

Doug over at ShortMedia has the Sapphire Radeon X800 PCI-E on the table, and he gives his $0.02 on how it performs.  The X800 is of course based on the R430 chip with one pixel quad disabled.  Is it the next $250 and below deal?  Read on!

I am 1/2 way done with my updated State of 3D, so if things go my way today, you will see that posted.

March 23, 2005

X800 XL Comment and News of Interest - Josh

For the past few days since I released the X800 XL review, I have received some very interesting comments.  I think that overall the $299 price point complaints are now moot, as users can get a variety of these products for that price.  While other manufacturers may be charging more, the bottom line is that any user can buy a X800 XL from a variety of manufacturers (including built by ATI) for that price.  Sure, sure, they were not delivered at the promised time, but how many of us are actually living in the past now?  If you are reading this, you are definitely in the present, and you can buy a X800 XL for $299 (and slighty cheaper at some places).  I guess that buying from that ATI store there is a coupon that will actually take the price down to around $269.  That is a smoking deal.  Anyway, the X800 XL is a great card for the price, and Sapphire makes a quality product.

The only real issue I have with Sapphire is their seeming lack of support.  After looking around their site, I can't seem to find any kind of warranty information, such as how long is the product under warranty.  I think this is a rather important piece of information, but one that I can't easily find.  The next issue is that they lack a 1-800 number in the US to call to get support.  This is the same around the world, in each of their offices.  To their credit they do offer a user support forum, but after reading through quite a few posts I rarely saw an official Sapphire tech person post on there to assist with any problems.  Typically it was users helping users.  While there is nothing terribly wrong with this, it would be nice to see a full time tech support person there to supply answers and give the average users some help.  Now, compare this with a company like BFG who has a true lifetime warranty (well, for as long as the company sticks around, which looks like quite some time) and a toll free tech support hotline (only toll free in the US though) that runs at an impressive 24 hours a day, 7 days a week.  On BFG's support page they clearly list their warranty standards, and how to contact them.  I think more manufacturers should adopt this strategy, but it takes time and money to implement (and sometimes both are in short supply for many manufacturers in this cutthroat market).

Now for some interesting things around the web...

The [H] takes a second look at NVIDIA's SLI with the latest 71.84 drivers.  These drivers support a much larger number of titles in SLI use, and it also has a Coolbits registry hack that allows users to create their own SLI profile.  This is going to be very handy for those with games not officially supported by NVIDIA.  There is a bit of a learning curve getting it to work, but nothing that a solid 10 minutes of trying won't cover.  Supposedly I am getting in a pair of 6800 GT's for testing in SLI, so keep your fingers crossed for me!  The boys over at [H]ard did a nice job with this one, and even uncovered a few interesting issues about the setup.

Scott over at The Tech Report covers the initial X86-64 implementations of both AMD and Intel, using WinXP 64 Professional RC-2.  He takes a variety of programs and measures their performance in both 32 bit and 64 bit modes.  While there are still not all that many programs that support 64 bit on the X86 platform, they are starting to grow.  This is an excellent look at what 64 bits can get you, and what it will cost you.

Legit Reviews grabs the Thermaltake Silent Purepower 680 watt power supply and loads it down.  It is amazing to think that there are 680 watt power supplies on the market for desktop use.  Crazy.

Ron from Hard Tecs 4U has posted their Athlon 64 cooler roundup, and they cover 16 competing products.  If you are looking for a nice, new cooler for that Athlon 64, this is the place to start.

Dave A. from Hot Hardware has posted a nice review on the HIS Radeon X850 XT and X800 XL graphics cards.  If you want his thoughts on the X800 XL, I suggest you read here.

And in the final bit of news, John from BFG has let us know that there is a nice little bundle of products available at Chumbo.  This includes 2 x 6800 GT OC's, a premium BFG backpack, BFG branded Ratpadz GS, and some BFG Teflon Slick Pads for the mouse feet.  All this for $899.  An interesting deal to say the least.

March 22, 2005

Intel Dual Core Thoughts and Motherboard Issues - Josh

While reading this article at The Inq this morning, a very peculiar thought crossed my mind.  The basic's of The Inq's article is that the Intel dual core program looks essentially from the outside to be two single cores bolted together, with very little pre-planning for true dual core implementations.  If you remember, the Smithfield processor is essentially 2 x Xeon DP's connected by a component that essentially acts like a separate front side bus on what is typically the Northbridge chipset.  So basically each Xeon acts as if it is actually connected to the true Northbridge, when in fact they aren't.  A clever idea, yes.  Terribly efficient?  I am unsure.  When AMD designed its original K7 core, in the back of their minds they wanted to create a dual core architecture.  When they started to work on K8, having a dual core architecture was very much in the front of their mind.  The AMD dual core products were designed at the outset to be integrated into a dual core package, and as such they have a very fast interconnect between each core that is HTT based, as well as an intelligent way to share the memory controller on the die, as well as share the system HTT link, all the while maintaining cache coherency.  Intel really only has one point of connection between the two cores, while the AMD product essentially has three.

Now, how much of a real world difference will this make?  I am unsure, but probably more than a few percentage points.  The AMD product will probably be more efficient per clock, but then again Athlon 64's already are (as we can see the 2.6 GHz Athlon 64 wins a majority of benchmarks over the 3.8 GHz Pentium 4).  With the top end Smithfield core running at 3.2 GHz, I am unsure what kind of factor overall internal latency will play.  Intel is already dealing with a lot more latency than AMD, due to the factors of the Prescott based cores having the massive 30 stage pipeline and the memory controller away from the processor core.  I am very curious how memory arbitration will work in the Smithfield, and I wonder how effective the connection between the two cores will be at handling each core's individual memory accesses.  Again, from the beginning AMD has planned for this with its three connections to each core.

Now, do I hate Intel or have anything against them?  No, I think competition is great, and I am happy that both AMD and Intel are doing well.  It is great to have choices as a consumer, and I believe both parties make outstanding products.  Currently I have gravitated all of my test machines to the Athlon 64 mainly because AMD actually provides me with processors, and the Athlon 64 really is the premier gaming chip at the moment.  Now for the peculiar thought that was inspired by The Inq.

It almost seems like Intel hates its motherboard partners.  The way to connect these new dual cores to the Northbridge is now slightly more complex, and it requires a new set of chipsets to accomplish.  So, once dual cores come out, then the motherboard manufacturers need to transition their products away the i915/G/P and the i925/X.  It doesn't really seem all that long ago that these products were released, but they did start to hit the market in August of last year, and really hit their stride by December.  Now, almost four months into its main production, these motherboard partners are being asked to utilize the new 945 and 955 based chipsets.  Now, how much of a redesign is needed?  I am not sure.  Some at the very least, a total redesign of the north part of the board at the very worst.  From all indications from many of the 3rd party motherboard manufacturers, the Intel based socket 775 motherboards have not been selling all that well.  So, any change for them is going to be a lot more expensive.

For users wanting to upgrade to a dual core Intel based processor, they will have to go through the extra hassle of upgrading their motherboard as well.  One thing that AMD has been pretty good about is the ability to keep a socket infrastructure going for a long time.  While the Socket 754 was short lived as the desktop Athlon 64 platform of choice, the Socket 939 will go on for quite a few years.  Users who want to eventually upgrade their systems with dual core Athlon 64 products will be able to drop one into their existing Socket 940 and 939 motherboards (usually with a BIOS update).  When looking at Intel's future plans for its parts, it really looks like motherboard design is going to get a LOT more complex for these manufacturers.  Here is a quote from The Inq's article:

When you look at the Dempsey/Bensley platform, which would be the Presler in XeonDP flavour, instead of adding the extra bus load of a second socket on the same FSB Hey, it would be a 5-load nightmare! Intel instead adds another FSB path in its north bridge, so that each dual-CPU DCM has its own 800 or 1,066 MHz FSB, feeding then on a wide multi-channel FB DIMM memory array. Well, we could also call this a crossbar - remember the six year old Alpha 21264 and its AlphaPC 264DP board? They had exactly this - separate paths for each CPU, then double-wide 512-bit memory bus to feed these two CPUs in parallel, six years ago. And yes, it was a 64-bit PC running Windows NT. A bit biiiiig-size PC with a 600W power supply, but guess what, we're about there now anyway.

To me, that sounds very complex.  When the Alpha was doing this, it was a very expensive machine to buy.  The motherboards themselves were typically around the price of a high end server board.  This sounds like a nightmare to design, much less get it to work at full speed.  These boards will definitely be at least 6 layers, and I tremble to think that it may be required to have 8 layers (essentially unheard of!).  Shoving a 512 bit wide memory bus, as well as the two individual connections to the processors will make the north part of the board very crowded indeed.  Then again, this product won't be for the desktop.  But consider the prices for the Athlon MP boards with their individual FSB's to each processor.  They were not cheap boards, and they only utilized a 64 bit DDR interface for the memory.  Oh yes, and don't forget the power draw for each of the dual core processors that will be fitting in this DP board.  Also, think of the BGA count on the Northbridge!  That will be a lot of balls!

Now, on the other hand, it seems like AMD is trying to take the workload off of the motherboard manufacturers.  In a single Athlon 64 setup, the user can essentially draw a line from the memory to the CPU to the Northbridge and to the Southbridge (if the chipset has one that is).  In dual processor setups the same line can be followed (memory -> CPU -> CPU -> Northbridge -> Southbridge).  If Intel hadn't already coined the term "Hub Architecture" I am sure AMD would have used it.  Dual processor boards for AMD are also a lot easier to design, and because the Opteron processors can easily share one memory controller via the HTT and NUMA setup, inexpensive dual processor boards like the Tyan Tiger K8W can be produced and sold for slightly under $200.  Not bad for a dual processor motherboard.  Even boards that feature DIMM slots for each processor still are simpler to design than the Intel counterpart.  The use of HTT and its 16 bit connection makes the design a whole lot simpler.

I think that the success of the Athlon 64 and Opteron took Intel totally by surprise.  And by the time that Intel started to make plans to counter AMD, it was already too late.  Right now it appears as though Intel is running at overdrive to get new products out to compete with AMD, and their recent IDF really showed how hard they are pushing it.  They are planning dual core for pretty much everything by the end of this year (Celeron's excepted of course).  From the outside it looks like a brute force affair, and it will continue on through 2006.  Then in 2007 I think things will make a significant change for Intel.  That is when their own HyperTransport technology will be released, and their processors will start to feature their own integrated memory controller.  It is unfortunate for Intel though that AMD will have had that technology a full four years ahead of them.  By 2007, what will AMD have introduced?  I don't think the tough times are over for Intel, and I have a feeling that their engineers are going to be pushed flogged for a long time to come to keep their products on a level basis with AMD.  The original idea behind the Pentium 4 architecture was good, and it was the dominant processor for a long time.  It just seems that Intel was stuck in a rut for about two years too long with the "clockspeed is king" mentality.  AMD brought an elegant and effective approach to server/desktop/notebook processing in 2003 with the initial launch of the Opteron and Athlon 64, and they continue to build on it.  Until Intel's new technologies hit, I think we will see a lot of overly complex affairs that will try to make up the difference between these two companies' products.  The recent release of the Turion products from AMD was a definite shot across the bow for the Centrino brand, and the interconnect advantages the Athlon 64 architecture has over the Centrino (integrated memory controller, simplified HTT connections, etc.) may play a large factor in the popularity of these products.

We definitely live in interesting times!

March 21, 2005

Sapphire Radeon X800 XL Review - Josh

For the past few weeks I have been playing around with the X800 XL from Sapphire, and overall I have been quite impressed by this piece of technology that ATI has delivered.  Though the X800 XL has been the target of some controversy in the past few weeks, we are now starting to see a much greater availability of these parts near the MSRP set by ATI.  While it is not the new technological marvel on the block, it does look to redefine the performance table at the $299 price point.  Here is a quote:

            In December ATI released the specifications on a whole gaggle of products, which were the eventual refresh of the X800 Pro and X800 XT series of products.  The new chip that the X800 XL is based on is the R430, which is essentially a R420 (that which powered the X800 Pro and XT series), but built on TSMC’s 110 nm process.  The 110 nm process is a size optimized process, and does not appear to be as fast as TSMC’s 130 nm Low-K process.  Apparently yields and speed bins for this process are excellent, which is why we are seeing the $199 to $299 price points for products based on this chip.  Other than the switch in the process used, the chip is essentially the same as the R420.  No major features have been added or changed.  This means that Pixel Shader 2.0b and Vertex Shader 2.0 are supported by the hardware.

You can read the entire review here.

Be sure to sign up for the forums and comment on this article/card!

March 20, 2005

Forums Have Returned - Josh

Yes, you have read that correctly!  I finally decided to go ahead and put those puppies back up, as it seems that the redesign that was promised is probably not going to happen.  Big bummer, I know.  Hopefully I can make enough scratch to have someone help me redesign it and put in a lot of nifty features.  So far, I am a ways away from doing any such thing.

So, sign up for the forums, and give them a spin!  Suggestions are welcome.

Copyright 1999-2005 PenStar Systems, LLC.