March 26, 2004

News About Town - Josh

Yet again, it has been ages since I have posted!  Lots of news though!

April 13th is the date which NVIDIA has set to release its latest and greatest chips.  A few weeks ago I was on a conference call with NVIDIA and we discussed some of the technology involved with this release.  While it was a good overview, nothing terribly specific was mentioned, so the information that I am relating here has been gathered from other places/rumors/sources.  First off it appears that there are actually two chips that can fall into the NV40 spec, one made by TSMC and the other by IBM.  If I were to hazard a guess, the 210 million transistor model is made by TSMC and the 175 million transistor product will be from IBM.  I have no idea how these two will stack up, or even if they will be released at the same time.  Rumors have mentioned that there will be availability in quantity of one of these products by May.  The big question is which one will be released when?  With ATI set to release the R420 at nearly any time now, which product from NVIDIA would be the best to set up against the R420?  I am not privy to that answer, but it appears as though the R420 will be a very fast card, but it will not support PS 3.0 functionality (and some question if it will support VS 3.0 also).  Both of the NVIDIA chips supposedly support PS/VS 3.0 functionality, and it appears that NVIDIA will push PS/VS 3.0 across the board this year (from the low end to the high end).  This is a good move, as their push to put true DX9 from top to bottom certainly saved their bacon (though we could argue how effectively the low end parts actually ran DX9 apps, the FX 5200 series has been an excellent seller for NVIDIA).  We do know that GDDR-3 will be the memory of choice for the next generation of high end products, and while NVIDIA did not help in the original specification of GDDR-3 (Joe Macri of ATI worked closely with Hynix to develop it) they are certainly the first to have a product out that supports it (GeForce FX 5700 Ultra- though apparently in very limited quantity).

There are also many rumors about how each architecture will be set up, and while nothing concrete has really been released, both the NV40 and R420 will be able to provide large amounts of traditional pixels, as well as incredible shading power.  With the DX9 shader 3.0 spec being firmly established for the past 3 years, both companies know exactly what to work towards.  Both cards will be fast, even though the ATI product will probably not feature PS 3.0 functionality.  This is not a bad thing as PS 3.0 will not be widely adopted until Q1 2005.  This gives ATI ample time to get their next gen product out the door that will support PS 3.0 functionality.  This does give NVIDIA quite a bit of mindshare though, as people willing to shell out $400+ for a video card may be more interested in being a bit more future proof rather than having the absolute fastest PS 2.0 card on the planet.  It now appears that memory bandwidth could very well be a severely limiting factor for the next gen of cards, as GDDR-3 has only just been released, and the theoretical specs of the upcoming GPU's show that they may already be bandwidth limited in certain situations.  There is no chance that the manufacturers would design a 512 bit memory bus, as it would be far too expensive, not to mention tough to design!  Still, the memory manufacturers will not stand still, and the people designing the actual boards themselves will enable faster speeds.

Needless to say, the next two months will be very interesting for the graphics market, as a whole slew of new cards and formats (PCI-E) will be released.  There will be countless reviews of the new products, as well as comparisons between ATI and NVIDIA, as well as the effectiveness of bridged vs. native PCI-E solutions.  I will be attending the April 13th unveiling, and I am quite excited about being there!  I haven't heard anything from ATI about a release, but there are some suggesting that there will be one shortly after NVIDIA shows their new hardware off.

March 9, 2004

NVIDIA First out with GDDR-3 - Josh

Last year ATI announced that it was working with memory manufacturers to develop GDDR-3.  DDR-2 had already been used in products from both NVIDIA and ATI (GeForce FX 5800 series and the Radeon 9800 Pro 256 MB respectively).  The problem with DDR-2 for graphics is that even though it runs at a slightly lower voltage, the first examples ran very, very hot.  ATI worked with memory manufacturers to develop the GDDR-3 spec, and I would imagine that NVIDIA probably had their say also.  Now we are in 2004 and NVIDIA is the first to bat with a GDDR-3 enabled solution.  The first product will be a GDDR-3 enhanced version of the GeForce FX 5700 Ultra.  Many were not too terribly excited about the prospect of this, but I believe that it is the first major step to adopting a significantly new memory standard that hasn't been seen since the introduction of DDR memory to graphics boards (beginning with the GeForce DDR).

Since little has been said about this development, I decided to shoot off a few questions to Brian Burke and friends at NVIDIA:  Questions are in bold, Brian's answers are in blue.

Will memory speeds be increased with the GeForce FX 5700 Ultra cards utilizing GDDR-3 from the base 450 MHz (900 MHz effective)?

The clocks for the NVIDIA reference design are increased to 475 MHz.

How will GDDR-3 effect the pricepoint of the GeForce FX 5700 Ultra?  Is it the same price as DDR-2?

The 5700 Ultra/GDDR-3 will hit the same price point as the 5700 Ultra/DDR-2

Does this memory generate significantly less heat than DDR-2?

GDDR-3 consumes less power and therefore generates less heat compared to DDR-2 for a few reasons:

1) The GDDR-3 devices run at a lower core voltage compared to DDR-2.  This is where most of the real power savings comes from since reduced voltage has an exponential effect on power.

2) Both DDR-2 and GDDR-3 feature on-die termination in the memory.  The termination technique used in GDDR-3 consumes less average power which helps thermally as well.

3) It has a higher density than DDR-2, which allows us to use 4 pieces of memory on the board instead of 8.  Fewer pieces of memory mean less total heat to dissipate.  The reduction in the number of memory pieces has also allowed us to discard the backside heatsink that was used on the original 5700 Ultra.

Using the 5700 Ultra allows NVIDIA to quickly deliver products based on GDDR-3 technology, and also gives the memory manufacturers some time to ramp up production on the new memory standard.  The next generation of high end graphics cards will most likely use this type of memory.  We can extrapolate a few things from the above statements.  PCB design will be simpler across the board with GDDR-3 due to the on-die termination, as well as the ability to only use 4 chips to create a 128 MB card.  This in turn will allow lower end 128 MB cards to be made on fewer PCB layers, which then decreases the overall amount for the card.  256 MB cards will also benefit from the on-die termination, and also the lower power consumption.  On newer, midrange cards we may see the need for a secondary power source for the video card decrease (it will more adequately run off of the AGP port and not need a direct line to the power supply).  This is not the case with the FX 5700 Ultra GDDR-3 based board, as it still needs the auxiliary power plug to work.  512 MB boards are now in the range of possibility, and we will most likely be seeing these cards in Fall of 2004.  With DDR-2, there would have been 32 chips, and the power needed to push them would have been phenomenal (and not to mention the heat produced melting the PCB).  Current DDR-1 modules have also reached the same density as GDDR-3, but the limitations of DDR-1 are now very much holding back next generation cards.  While we will still see DDR-1 based video cards for quite some time, the high end (and some midrange) will now be dominated by GDDR-3.

You can discuss this topic here.

February 24, 2004

AMD and HP - Josh

AMD and HP have announced that HP will be providing Opteron based Proliant servers in the very near future.  This announcement is a watershed event for AMD, as it signals to the rest of the business world that their processors are now competing at the highest levels.  While many have known that the Opteron is a high performance processor, and its 32 bit/64 bit support opens up new worlds of operation, it has not achieved the success that many were hoping until this point.  Now that HP has announced plans to deliver products based upon the AMD Opteron, only one major OEM stands alone by not having an AMD Opteron based server product (Dell, of course).  We cannot deny that this is a huge event for the industry, and for AMD in particular.

HP and Intel have had a very long relationship, and HP essentially stopped all work on their PA-RISC processors to lend their support to Intel's Itanium and IA-64 initiative.  In the past decade billions of US dollars have been spent on the IA-64 architecture between HP and Intel.  Initially IA-64 was supposed to be delivered in 1998, and by 2004 IA-64 products were to have migrated down the ladder where they were to have been positioned to workstation class products.  Then a few more years down the road IA-64 would have headed for the desktop.  As we all know, that original plan has long been scrapped.  Itanium based servers are now considered niche in many aspects, and while there are areas that Itanium excels in, it has not had nearly the success that Intel and HP had hoped for.  The vast majority of servers out there run on Xeon based processors running X86 code (Windows, Unix, Linux, and other operating systems).  AMD's evolutionary approach to 64 bit computing has made the industry take notice.  When AMD announced its plans for 64 bit desktop and server processing based on an enhanced IA-32 architecture, many around the industry breathed a sigh of relief.  The idea of massive porting of code to IA-64 was not a pleasant thought for developers, and the further idea of implementing new software on IA-64 servers was not desirous for consumers and companies.  The evolutionary upgrade path that AMD was offering had very few pitfalls for early implementers, while it gave those more adventurous souls a chance to upgrade to 64 bit applications and operating systems.  The idea was good enough that Intel quietly started working on its Yamhill technology near the end of 2000 so it could expose this extra functionality if they saw fit.  As we know from this February's IDF, Intel saw the need.

This is not to say that IA-64 was a failure, but its market appeal has been limited at best.  Large database customers are very happy with the performance of the Itanium and its software, but the EPIC architecture has not lent itself well to the majority of applications out there (mostly due to the extreme compiling time to get good EPIC based codes).  It is very probable that IA-64 will never reach the desktop, and it will remain a high end server product until Intel phases it out for a future technology.

AMD64 (or as Intel calls it, IA-32e) fits the needs for a broad spectrum of markets, from the high end server to the lowly desktop and laptop market.  That HP has adopted this technology in the face of Intel's upcoming Xeon 64 products, as well as the current Itanium lineup, is nothing short of revolutionary for AMD.  While HP has adopted AMD products in the past, these were only in the consumer markets, or the low-end business class market.  Now that HP is releasing single, dual, and quad Opteron products, it is covering a large portion of its server market with these products.  AMD has done an amazing job with its Athlon 64 architecture.  The longer it is out, the more apparent it becomes that it is turning out to be a revolutionary product.  AMD made a good start with the original Athlon MP, but the Opteron class processors are a huge step up.

Please help discuss this news in our forum!

February 13, 2004

VIA Releases Dual Memory Channel Athlon XP Chipset - Josh

The much awaited KT880 from VIA has finally been announced.  While quite a few manufacturers have jumped on the wagon, no boards are currently available for this chipset.  This should change in about a month and a half, but until then the nForce 2 appears to be the best chipset for the Athlon XP.  While the Athlon XP is destined for the EOL pile, it is still a very cost effective processor with a very mature platform.  For those on a budget, an Athlon XP is still a very good choice.  One thing that is a bit disappointing is that there will probably never be a PCI-Express based chipset for the Athlon XP.  It looks to have only AGP for the rest of its lifespan (unless VIA decides to do a KT880 Pro or something in that line- I doubt NVIDIA or SiS want to dedicate the engineering talent to make that possible).  One thing is for sure, the latest Athlon XP chips from AMD run very cool, and many people are showing some very impressive overclocks.  From all appearances, AMD is having outstanding yields and speed bins on the Athlon XP.  I had one of the first Athlon XP 3000+'s and it ran hot.  I have a brand new one sitting in a test bed, and even with an average cooler, it is sitting around 95 F.  AMD could have easily done a 2300 MHz chip (or beyoned), but I can see that they did not really want to cannibalize their sales of the new Athlon 64 3000+ and above.

Still, competition from VIA in the high end Athlon XP chipset market is a good thing, and hopefully some good reviews of this product will come out shortly.  This will definitely keep the Athlon XP market interesting and competitive.

Forums Sorta Hopping - Josh

Only a handful of members have signed up for the forum, and I would love to see more!  So far we have had some very interesting discussions, and I hope to keep that ball rolling.  I am finishing up the benchmarks for the 5700 Ultra (comparing it to the 5900 Ultra at the high end, and the 9600 XT for its true competition).  So far my sample overclocked quite nicely to 576 MHz core and a 1.01 GHz memory.  That increase in clock speed sure gave the board a kick in the pants.  Keith has an article into me already, I just need to go over it one more time and actually post it!  Hopefully this weekend I can get many things done!  Anyway, sign up for the forums and join in the discussions!

February 5, 2004

BFG Wants Your Help! - Josh

John Malley of BFG contacted me about a little contest they are holding.  It seems that BFG really takes gamers' opinions to heart, and are holding a contest that will provide 15 lucky winners a chance to hang out at the BFG headquarters, play in the Cuckoo's Nest LAN room, and possibly acquire some BFG swag (not sure what the prizes will be, but I imagine a few vid cards will be offered as well as shirts, hats, etc.).  They unfortunately will not pay for transportation costs, so this might be easier for people closer to BFG headquarters than for all us folks a little farther out (like here in Wyoming).  So if you are interesting in giving BFG your opinion about their products, packaging, or their overall marketing strategy, sign up for the chance to tell them how you feel here.  BTW, BFG headquarters are located in Lake Bluff, Illinois.

I am currently finishing my BFG 5700 Ultra review, and this is a very solid (and somewhat exciting) midrange card from BFG.  Some of the high points are its overall performance compared to its competition, the nifty cooling system (which looks nice and is actually quite effective), and the lovely blue color that it shows up with.  I hope to have this review out soon, as the card has been a pleasure to work with..

Prescott Availability - Josh

I just checked with a major distributor, and they currently have thousands of Prescott based P4 processors in stock.  The only real problem is that their entire selection is comprised of 2.8 GHz and 3.0 GHz chips.  No sign so far of the 3.2 GHz and 3.4 GHz P4 "E" processors so far.  So, expect to see quite a few online stores start to sell 2.8 and 3.0 P4 "E" processors very shortly.  The good news is that the availability of the 3.4 GHz P4 "C" processor is outstanding.  If I were to buy an Intel based processor today, this would probably be the best choice in terms of overall price/performance.  While you will miss the SSE-3 bandwagon this round, the overall performance of the 3.4 GHz "C" model will more than make up for it in most instances.

February 2, 2004

Intel Releases Prescott - Josh

Here I was expecting Prescott reviews to be released today!  That was not the case of course, as yesterday afternoon the NDA's expired and we were able to see how well the new processor did.  Prescott (or the P4 "E" model) showed performance for the most part slightly slower than a Northwood based P4 ("C" model).  Intel did a complete core redesign for the Prescott, and it does include many new features that either help its overall performance now, or will be exposed by future software platform upgrades.  SSE-3 will be the big selling point once software gets compiled to take advantage of this unit, and Johan over at Ace's has commented that the overall transistor count doesn't match what should be expected, and he is wondering if there isn't a 64 bit portion of Prescott that is not exposed as of yet (the fabled Yamhill technology).  We know that Intel will be showing off a Xeon this month with 64 bit capabilities, so it would not be a huge stretch to assume that Prescott also shares this same 64 bit functionality.  Of course, the question is if the 64 bit part is nearly identical to what AMD has, or if Intel has made a totally different unit that is only compatible with AMD-64 but with extended features.

Prescott will not be widely available to the general public until March, but many major OEM's will be offering systems based on this processor shortly.  Intel also released the 3.4 GHz Extreme Edition, which will retail for around $1,100 US.  This is a very fast processor, but many question its price/performance for the average enthusiast.  Many reviews think that buying an Athlon 64 3400+ is a much better investment, as the enthusiast could essentially buy the entire system for that much money (sans monitor).

At the same time Intel has quietly released the 3.4 GHz version of the Northwood P4.  This is the processor that will sell the most for the time being.  Intel is in fact flooding the market with these as we speak.  I checked with a major distributor, and they have over 60,000 of these processors in stock and ready to ship.  The price is well under $500 US also.  This makes the 3.4 GHz P4 one of the better deals around at the moment!  Reviewers that have these processors in their hands report overclocks at default voltages of up to 3.8 GHz.  Major OEM's are already offering systems using this processor.

January 30, 2004

Intel and 64 Bits - Josh

It is now being widely reported that Intel will be showing off its 64 bit Xeon processor at next month's IDF.  Apparently this will be the fabled "Yamhill" technology for Intel, and will most likely be based on AMD-64 (since MS says it refuses to develop another operating system for the x86 market).  So, Intel looks to have knuckled down and is using its main competitors overall 64 bit design and instruction set.  This is quite a change considering past innovations.  If you remember, while Intel was the first to use an integer based SIMD unit (MMX), AMD had the first floating point SIMD unit (3DNow!).  Intel followed the introduction of 3DNow! with SSE about 9 months after AMD.  AMD of course introduced an advanced version of 3DNow! in later processors, but all development for that instruction set has now stopped.  Instead, AMD is utilizing SSE-2 in its latest Athlon 64 processors.  Most likely, sometime after Prescott is released, AMD will again integrate SSE-3 functionality into its processor line.

Still, the fact that Intel looks to have adopted AMD-64 (or at least their own compatible version) is huge.  This is a company who has a track record of rejected anything NIH (not invented here).  To have to adopt the basic technology of its main competitor is a blow to Intel's ego, but something that is very good for the industry.  After the multi-billion sollar disaster that is Itanium, the world needs a solid 64 bit set of processors for the future.  And that future is X86-64.