January 18, 2005

Two ATI Tidbits - Josh

I have seen the X850 XT PE in the wild!  Yes, it is true.  One of my distributors has them in stock, and I know that I should probably get one and eBay it!  Of course, with my luck, most other online folks will have these cards and I will be left high and dry (and light some money).  Oh well.  Still, it is very good to see that these cards are now being made available to the public, and hopefully it bodes well for users wanting to buy this product (unlike the last Platinum Edition product from ATI).  My distro does not have any of the other models of the X850 or the newer X800 series.  Many believe (and rightly so) that the X800 XL will prove to be a very impressive part with excellent price/performance.  One thing to keep in mind though is that this particular distro also carried the X800 XT PE for about a week this last June, and never had it available again.  ATI will still have to prove they can provide solid amounts of chips to the OEM and retail market.  If they can't, it certainly appears as though NVIDIA finally has enough 6800 GT and 6800 Ultra products available to satisfy consumer demand in the high end space.

The second bit is that ATI has released their latest Catalyst 5.1 driver set.  These are the first issue of the new year, and they appear to solve a handful of problems and provide better performance in Half Life 2 (and thats about it).  Another solid release from ATI, and I applaud their dedication to providing new drivers with plenty of fixes on a monthly basis.  Personally I hope they continue to release two versions (though it is a lot more work for them), as I really don't feel like installing Microsoft's .Net 1.1.

January 16, 2005

Happy New Year and AMD News - Josh

Well, 2005 is here whether we like it or not!  2004 was a very interesting year around these parts, with lots of good, and enough bad to keep things interesting (and us on our toes).  We welcomed my son Tai into the family, and I have undergone a few career changes that left us a little light in the wallet (but we are getting by).  I started a new job as a sysadmin at a engineering firm here in Laramie, but I should still be able to spend quite a bit of time on the website.  Some really interesting technology will be coming out soon, and I want to finish up a bunch of testing on products that I have had laying around here for a while now.

Now for the AMD stuff...

AMD has taken a hit as of late in their stock price due to an apparent shortfall in their flash memory department.  When you stop to think about the flash market, this is not surprising.  Nearly all of the flash makers have high quality, high density flash products that are out in great numbers.  The supply looks to be outstripping demand by a significant margin, so flash is very very cheap at this time.  PDA's, digital cameras, cell phones, and nearly every other handheld and set top electric device utilizes flash, but even in the holiday season only so much product can be sold.  The manufacturing capabilities of all of these companies appears so efficient that any three of them together could probably supply enough flash to satisfy the market.  As flash densities grow and grow, I think that more companies will start to drop out of this market due to pricing pressures, and the cost of producing flash with diminishing margins.  Perhaps this is a market that AMD is actively addressing so that it will be one of the few left.  I would say that within the next 10 years we will see only a handful of flash manufacturers, even with the massive demand the consumer market has for cheap, high density flash memory.

If AMD wants to continue to address this market, they must consistently make money in their other business aspects, namely processors.  The next 9 months for AMD are going to be exceptionally important for the company in their processor division.  Intel currently doesn't have a flagship processor that really competes well with the top end Athlon 64 (and at each price point it seems that AMD does provide a performance advantage in many current applications).  This will change around August/September.  Last year Intel killed off development of future Pentium 4 based cores, and instead focused on dual core and Pentium M production.  Recent tests have shown the current Pentium M in a desktop environment to match the fastest processors from both AMD and Intel (with a slightly overclocked version of Dothan- usually around 2.1 GHz).  I think we can see where Intel will be heading when it comes to its next generation of desktop processors.  If AMD expects to survive against Intel after September of this year, they must continue to show improvement in their Athlon 64 line to match everything that Intel brings to market.  Once the desktop version of the Pentium M hits the market, it should redefine Intel's entire product line.  It is a very efficient, cool running, and potentially explosive product for Intel.

So, if AMD expects to hold up against this assault, they need to capture a much more significant percentage of the market before Intel brings its next generation products this coming Fall.  To do this they must do several things right.  The first is to continue to push the performance envelope with their second generation 90 nm Athlon 64 processors.  This will have to have slightly better performance than the current Winchester based cores, yet still be able to clock upwards of 3.0 to 3.2 GHz.  While AMD initially appeared to be set to release the San Diego and Venice cores at the end of January.  This no longer appears to be the case unfortunately.  There are many that question if AMD will release these products anytime in the first quarter of this year.  This is unfortunate for AMD, as a further performance processor release would have helped to continue to win mindshare for AMD, which in turn provides more impetus to consumers to buy AMD based products.

The good news is that the current Winchester core appears to really have it going for it in terms of yields and speed bins.  While the only 90 nm products at the moment are the 939 based 3000+, 3200+, and 3500+, many users are finding that these chips easily clock to 2.5 GHz and above (with some claiming 2.8 to 2.9 GHz with air cooling).  This means that AMD is well on their way to achieving profitability on their 90 nm line in a very short period of time.  Not only does 90 nm appear to help increase the clock potential of these products, but these run cooler and with less power.  The next positive aspect is that AMD can pack more chips on a 200 mm wafer with their 90 nm process, which should peak the interest of many major OEM's.  If AMD can show these manufacturers that they can expect a solid supply of these processors, then the major OEM's are more comfortable in providing high demand computers based on mid-range and high end AMD processors.  Supply is a tricky subject with major OEM's, and it is still one of the more important reasons why Dell has never offered an AMD product as of yet.  Intel can promise Dell a nearly unlimited supply of processors, while AMD cannot.  One of Dell's fears is that if they decide to make a major line of computers using AMD processors, then other OEM's will follow and this could really dry up the supply of Athlon 64 processors.  No OEM likes to see their lines at a standstill due to a deficiency of critical parts.

In Q3 we will also see some of the first test wafers being put through the new 300 mm Fab at Dresden for AMD.  If the results of these tests are seen as positive, then OEM's can adjust their future roadmaps to consider the initial 65 nm AMD parts made on the 300 mm wafers.  This should help ease supply fears for many major OEM's, because by the 3rd quarter of 2006 AMD will have two megafabs producing AMD processors.  The big question is if AMD can capture enough marketshare to keep both fabs busy, and to be able to do that AMD needs to weather the coming storm from Intel this Fall.  Again, I believe to successfully do that AMD needs to be VERY aggressive in the next nine months with product releases and feature enhancements.  The first step is getting the San Diego (1 MB L2 Athlon 64) and Venice (512 K L2 Athlon 64) out the door and into consumers' hands.

The final aspect is to be able to provide the new 90 nm Opterons (Athens, Troy, and Venus) into the server peoples' hands.  Reducing the core size, power requirements, while increasing the performance will help AMD capture more of the server market (something that they have been slowly gaining on).  The Opteron is honestly a very solid and high performing server product with an excellent track record so far.  The infrastructure that the Opteron sits on is very robust, and it is becoming more and more popular in light of Intel's latest server products that eat up a lot of power and generate so much heat that it can become a problem in dense applications.

Well, enough for now as I am planning on a new AMD Today article that will cover these points and others in a bit more detail.

December 24, 2004

Off for Christmas, and Other Thoughts - Josh

Tis the Season!  I am about to take off to the in-law's house for Christmas.  No in-law jokes here, as I do get along very well with my wife's family!  Should be plenty of good food, some fun games to watch, and perhaps a little quiet time sandwiched in here and there.  I would like to wish all our readers a safe and happy Holiday season (no matter what religion or creed they follow)!

The Athlon 64 review seems to be doing well in terms of traffic, and I have received a couple of interesting comments about it.  One thing that I will be trying to explore more is the memory speed divider issue I ran into.  I don't know if this is something that is a "feature" of the Gigabyte K8NS Ultra, or if all Athlon 64 939 boards do the same thing.  I have seen very little mentioned about it, so it definitely calls for more testing!  As soon as I can get in another board to find out for sure, I will let you all know.

Some other comments I ran into deal with the roadmap I had mentioned in the article.  I am positive I have seen January as the time when AMD releases the next high end part, but other places such as Anandtech show that the FX-57 will not be released until Q3 2005.  I am not really sure who is right here, but when looking at the official AMD roadmap, we are supposed to see a whole slew of new 90 nm Athlon 64 parts hit the market.  While the Opteron x52 series are postponed until early next year, I find it hard to believe that AMD will not release another high end desktop part until Q3 of next year.  With Intel looking very close to releasing the 2 MB L2 cache Prescott's, I would imagine that AMD will release something new right around that time.  The code names for the 90 nm parts due in the first half of next year (and the way I am interpreting it, I still think it will be a late January release) are San Diego (1 MB L2, 90 nm SOI, strained features) and Venice (512 K L2, 90 nm SOI, strained features).

These new Athlons should have a couple of interesting features.  The first will be scalability.  I would imagine that the design has been tweaked to gain more clockspeed (much like the Thoroughbred and the revised Thoroughbred-B).  The fun doesn't stop there though, as it appears that the memory controller will be tweaked a bit more, and SSE3 functionality is added.  Everything else is a bit up in the air, but I wouldn't think that there would be any more major changes.  Perhaps with all the feature and performance gains, the FX-57 will still be a 2.6 GHz clocked part, but with everything else added to the mix it will perform better than the current FX-55.  Again, this last part is all speculation, as I still expect a 2.8 GHz clocked FX-57 to be introduced early next year.

nForce 4 boards are trickling into the market, much slower than originally expected.  NVIDIA originally thought that product would be on the shelves in good quantities by the end of November.  Only now are we starting to see these new boards hit retail.  VIA appears to be having a few problems with their K8T890 chipset, but the first samples are now finally hitting Akiba in Japan.  I would expect there to be only a handful of boards supporting the 890/8237 combination, and the chipset really won't take off until VIA can deliver their new 8252 southbridge.  SiS appears to be making a strong push in the Athlon 64/PCI-E market with their latest creation, the SiS 756.  SiS has a good history of making very solid chipsets, but for one reason or another their products have always been relegated to the value market.  In the past few years, I have felt that their chipsets performed better and didn't have as many issues as competing products from NVIDIA and VIA.  They have also always been solid performers in each market they address.  Perhaps 2005 will be the year that SiS gets the recognition they deserve, as their chipsets are seriously top notch products.  Now that SiS has divested themselves of their Fab and their graphics division, they can concentrate on providing the chipset market with top performing parts.  I am very excited to see what they hope to deliver this year.

GeForce 6800 PCI-E parts are also starting to appear in retail, and none too soon!  So far ATI has had a stranglehold on any retail PCI-E parts for the high end, but this should level the playing field a bit more.  January looks to be the time when we see a lot of 6800 PCI-E parts hit the shelves (just in time for AMD based PCI-E parts to be matched with them).  The NV41 (12 pixel pipeline, 5 vertex shader, PCI-E native) is now fairly widespread in the OEM market, and we will start to see these parts in greater availability very shortly.  From my understanding, this part is made on TSMC's 110 nm process, so we should start to see them go down in price til they hit the $250 point.  Anything under that will be addressed by the NV43 based GeForce 6600 series.  With ATI's recent introduction of the R430 chips, we should very shortly have a good selection of cards from a variety of manufacturers to choose from in the important $150 to $250 range.

That is all for now!  Have a great Holiday!

December 22, 2004

AMD Athlon 64 3200+ 939 90 nm Review - Josh

Several months ago AMD announced that they were doing volume shipments of 90 nm parts.  These first parts were mobile Athlon 64's, but it wasn't very long until the desktop 90 nm versions started to appear.  While Intel has come under fire with its Prescott based Pentium 4 (no pun intended), AMD has hoped to sidestep the heat and power question with its own 90 nm process.  SOI and transistor design have certainly paved the way for users to get a cooler running Athlon 64.  Here is a quote:

            This is not to say that the jump to the 90 nm node was not without its fair share of problems.  Many remember the birthing pains with the jump to 130 nm, but the transition to 90 nm was even harder.  Time and money are the necessary cures for any process jump, and throwing enough engineers at the problem tends to make things work better.  Intel had a lot of engineers working on the 90 nm node, and they were the first out of the gate with a mass produced product (though IBM had 90 nm PowerPC parts at this time, they were not produced in such large quantities as what Intel did with Prescott).  Intel engineers still do a lot of work to refine their process, but changes with the Prescott core have also been made to lower the power and heat problems.

You can read the entire article here.

I would also like to welcome Abit to the website, as they have purchased advertising here.  I have worked with Abit for years, and while their reputation may have been sullied in the years past, I feel that their current lineup of products is very solid.  Gone are the days of multiple RMA's and caps falling off the board!

December 15, 2004

NVIDIA Releases the "Real" GeForce 6200 with Turbo Cache - Josh

You may remember some weeks ago that NVIDIA released the GeForce 6200, a sub $129 card that features 4 pixel pipelines.  The initial 6200 cards were all based on NV43 chips though (the same chips that power the faster GeForce 6600 series).  In the 6200 cards NVIDIA disabled 4 of the pipelines.  This was an impressive value performer, and it featured nearly all of the advances that its larger (and more expensive) brothers brought to the market.  Things like Z and color compression, as well as the FP 16 buffer, were removed to save on transistor counts, but other wise it is a fully functional SM 3.0 part.  It seems that the cut down NV43 6200's are destined to never truly be, as NVIDIA has finally released the true 6200 based on the NV44 chip.

The NV44 is based on TSMC's 110 nm process, and the core is very small compared to nearly anything out there.  This should make the 6200 very inexpensive to manufacture.  The part of the NV44 that really sticks out is how it handles memory.  Turbo Cache is NVIDIA's way of accessing main memory in conjunction with onboard memory to give excellent performance for a very low price.  The first several versions of the 6200 will only feature 16 MB and 32 MB of onboard memory.  The 16 MB card will only feature a 32 bit wide memory bus, while the two memory chip 32 MB version will have a 64 bit wide memory bus.  Now, 16 MB and 32 MB are not enough to play any modern game, but NVIDIA dynamically allocates space from the main memory to do both texture reads AND writes.  Now, texture reads are old hat and is supported in the AGP specification.  Texture writes are something else entirely.  NVIDIA leverages the power of the PCI-E bus to create a card with very solid performance for its price range, but allows board makers to create a very inexpensive card.  PCB costs are less because there is no need for the wider memory bus widths, and memory costs are significantly lowered due to the card using main memory for much of its needs.

In reviews that I have seen the 64 bit/ 32 MB DDR 6200 performs on average better than the X300 series from ATI.  The X300 features 128 MB of onboard DDR memory on a 128 bit memory bus.  At first glance one would assume that the X300 would have a significant advantage in bandwidth and local memory.  Once the benchmarks start to roll though, the strengths of NVIDIA's design quickly come to the fore.  Eventually 64 MB and above versions may eventually hit the market, but it will stay with a 64 bit memory bus at its highest.  When aggregate bandwidth is figured out for the 64 bit card, it comes out to be around 13 GB/sec (counting the speed of the main memory).  This is more than enough bandwidth for a value level card, and it has been shown to play even the latest games at decent resolutions and framerates.

NVIDIA is redefining the low end with this product, and I can foresee that other manufacturers will follow suite with products of their own that share this philosophy.  The 6200 will never make it to AGP land though, so the value products for legacy AGP parts will continue to be the GeForce 4 MX and FX 5200 series of cards.  AGP just doesn't have the flexibility nor bandwidth required to pull off the performance that the 6200 requires.

The 6200 with Turbo Cache looks to be a great little product for the value market.  The great strength of the Turbo Cache design is that it can be carried over into other future products from NVIDIA.  Do not expect Turbo Cache to be a feature used only in the value market.  The ability to read and write to main memory at the speeds that PCI-E can provide will allow NVIDIA to leverage this technology into the mainstream and high end market.  When titles such as Doom 3 can utilize over 500 MB of texture and art data, then that data will fill up in main memory.  With NVIDIA providing a relatively low latency mechanism for reading and writing that texture data to main memory, then the local video card memory limitations become less and less.  While this technology may not make a world of difference in performance, it does add a lot of flexibility to NVIDIA's offerings.  With future 512 MB cards and computers with 2 GB of main memory, the possibility of a video card being able to effectively use 1 GB of space for texture and geometry data opens up a whole new world of possibilities for immersive games.

For the time being though, enthusiasts should probably look elsewhere for a video card, but for the very casual gamer the 6200 is a very solid and well performing product.  Did I mention that retail cards will be under $99?  A pretty good deal for everyone involved.

Here are a couple of good, full previews of this product:

Tech Report

Hot Hardware

ATI X800 XL Takes the Spotlight - Josh

In a seeming "coincidence" we are finally able to see how the recently announce X800 XL performs.  This R430 based card is now priced at a very reasonable $299 and features 16 full pixel pipelines running at 400 MHz.  In terms of overall performance, this card slightly overshadows its much more expensive counterpart, the X800 Pro.  ATI is trying their best to cut the 6800 GT out of the picture for this holiday season, and for the price it does a very good job.  The only real issues with the X800 XL is that it is not available anywhere (and won't be for some time).  Still, ATI looks to capture the $299 space for itself with this very aggressive move.  In benchmarks the X800 XL matches the $100 more expensive 6800 GT.  NVIDIA will not take this sitting down of course, and when X800 XL's do finally become available I would expect the 6800 GT to fall closer in price to the $300 mark.  This will force the regular 6800's down to $250 (and possibly slightly below).  Again, availability will be key for this release, and there is no word when we will see a street date for this product.  So, for the Holiday buying season, the GeForce 6800 GT (in both AGP and PCI-E flavors) appears to be the solid buy for the enthusiast market.

Here are a couple of reviews of the X800 XL:

HardOCP

Tech Report

Hot Hardware

December 13, 2004

AMD/IBM Strained Silicon Announcement - Josh

Late last night AMD and IBM sent out a press release detailing their new, jointly developed, strained silicon process.  This process is termed DSL (Dual Stress Liners).  Now, AMD had previously done a simplified version of strained silicon with their products on the 130 nm and 90 nm lines this past August.  This strained silicon only affected N-channels, and was used mainly to stretch the silicon.  Well, stretched silicon doesn't work well with the other kind of channels used in transistor design, the P-channel.  To achieve better results with this type of transistor, the silicon has to be compressed.  Intel has been using both n-mos and p-mos straining since the introduction of the Prescott in January of 2004.  From the limited information that AMD and IBM has given about this process, it appears that their approach is a lot simpler to implement than Intel's way of doing it.

In August AMD implemented limited strain technology to their 130 nm line, which allowed the production of the 2.6 GHz Athlon FX-55.  Since that time nearly all 130 nm parts utilized strained silicon.  Just a little bit later AMD also integrated strained into their 90 nm products.  From hints given, I would say that the new DSL technology is now being implemented into AMD's 90 nm fabrication, and we should be seeing the first products of this change in January.  There is also another change in the air, as a new Athlon 64 design will make its appearance also at that time.  This new design is supposed to allow better clock scaling with AMD's 90 nm process, as well as add features such as SSE3.  I would imagine much of the core was reworked, so we can expect to see a more optimized product.  Many expect to see the first iteration of this processor as the FX-57, which should run at 2.8 GHz.  Of course, nothing is set in stone, so I will be very interested to see how everything pans out this coming January.

AMD and IBM have said that their strained process increases transistor performance by up to 24%.  This number does not mean that an Athlon 64 that runs at 2.4 GHz on a non-strained process will run at 3.0 GHz on strained, but rather that a single transistor in testing will run 24% faster using the same amount of power as the non-strained transistor.  A CPU is obviously the sum of its parts, so while we will see definite performance and power optimizations, it will not reach 24% without a major redesign.  Well, that is exactly what the product hitting the market in January is.  So, the first products will probably hit 2.8 GHz, while future releases will hit the 3.0 and 3.2 GHz marks.  After that we will most likely start to see the dual core processors hit the desktop market (as 3.2 GHz is really about as high as the 90 nm strained Athlon 64 can go).  Still, Intel has no product on the horizon that will compete with a 2.8 GHz Athlon 64, much less a 3.0 and 3.2 GHz version.

AMD continues to slowly trudge forward with their Athlon 64 design, making it better and better as the days go on.  I know that I am very impressed with the 90 nm 3200+ that I am currently testing, and the products that will be hitting shelves this next year will show yet again another marked improvement.

December 10, 2004

NVIDIA/Sony Announcement - Josh

It seems the biggest news from this week was the announcement that NVIDIA is in cahoots with Sony to design the GPU for the PlayStation 3 console.  Many were speculating that the design would be in-house at Sony, but this announcement seems to clear up quite a few misconceptions (as well as raise a few questions).

The X-Box GPU (NV2X) was a good moneymaker for NVIDIA.  They initially received $200 million for the development, and averaged around $40 million a quarter in X-Box GPU sales.  Many were wondering why NVIDIA didn't fight harder for the X-Box 2 project.  Even though NVIDIA and Microsoft have had several tiffs over the project, it was still a very good source of income for NVIDIA.  My gut feeling is that after the X-Box was unveiled, Sony came calling to NVIDIA.  The X-Box is a great challenger for the PlayStation 2, and in many ways it is far superior.  Graphically it is a big step forward, and the developer tools that Microsoft provides are second to none (as compared to the painfully awkward tools that Sony provides for the PS2).  While Microsoft's share of the console market is far behind that of Sony, they have taken up a serious chunk, and in many circles the X-Box is the console of choice (better graphics, better sound, X-Box Live, etc.).  Sony obviously thinks the X-Box is a big threat, and looking many quarters down the road they can only imagine what the X-Box 2 will be like.  Sony has had a front row seat in watching how 3D graphics have evolved over a short period of time.  The original PS1 had a pretty basic 3D graphics chip, and only a little while later did 3Dfx release the original Voodoo.  The PS2 had a better 3D engine, but while it could push the triangles, it was no match in visual splendor to the GeForce 3 that NVIDIA was poised to release mere months later.  Sony knew they probably could not compete without getting help from ATI or NVIDIA, and of course the best way to compete with Microsoft is to cut their legs out from under them.  I have a feeling that Sony approached NVIDIA with "an offer they couldn't refuse".  In one fell swoop Sony took the GPU/Northbridge/Sound/Southbridge provider from Microsoft.  I think this speculation may hold up well, as NVIDIA really didn't fight very hard for the X-Box 2 contract.

While it is unknown if NVIDIA will do the sound for the PS3, we do know that Microsoft handed off the sound capabilities of the X-Box 2 to SiS.  Now, SiS isn't a sound chip company, and they don't have a whole slew of engineers from Aureal to fall back on (like NVIDIA did).  Also, since the original X-Box deal with Microsoft was that they buy the chips from NVIDIA, the technology within still belongs to NVIDIA (as compared to the royalty deal that MS has with ATI).  So, the sound for the X-Box 2 cannot use any "SoundStorm" technology unless NVIDIA strikes a royalty deal with Microsoft for the X-Box 2.

Something like 76 million PS2's have been sold worldwide, and we can expect the PS3 to be as popular (if not more).  So, NVIDIA has the chance to rake in a lot of bucks from Sony, though their terms are based on a royalty agreement instead of NVIDIA selling Sony the finished chips.  Apparently Sony will be producing these GPU's at one of their own Fabs in Japan.  While this type of deal will decrease revenues a smidge, NVIDIA no longer has to worry about details such as; TSMC has issues with their process and a whole batch of PS3 GPU wafers are rendered useless, the substrate producer can't deliver the parts they were supposed to nor on time, finished GPU's that were either lost or damaged in transit, etc.  So, while revenues may not be as nice looking as when finished chips are sold at a good price, the positive aspects for NVIDIA far outweigh the negatives (nobody likes to deal with liability).  Details are obviously scarce so far about what the chip will be like, but we really need only look at the foundation made by the current NV40 to get a good idea what the PS3 graphics will be like.  Needless to say it will be a far cry above that of the PS2, and running with all the bells and whistles (most likely with AA at HDTV resolutions) will make for a breathtaking experience.

Another area where NVIDIA will help Sony out a great deal is with the development tools they are also going to help create.  With their PC Developer relations experience, as well as their Cg experience, NVIDIA is in a good position to provide far more powerful tools that are easier to implement than what Sony has provided in the past.  This could make porting titles to the PS3 a lot easier than what developers previously had to do with the PS2.  Sony is honestly trying to "out-Microsoft" its main competitor with this one simple defection.  Interesting stuff.

Copyright 1999-2005 PenStar Systems, LLC.