AMD Sneaks in 90 nm Semprons - Josh

Some weeks ago AMD quietly announced that its new series of Socket 754 Semprons were released.  What really wasn't mentioned much was that these new Semprons were all 90 nm based Palermo cores.  This is the new generation of Sempron cores, and they consume a very small amount of power!  AMD expects around 67 watts max for these products, which is down from the 89 watts from the previous generation of Semprons.  This is a significant power savings, and it will be a very easy processor to cool.  There are several reasons for this excellent power consumption, but one of the largest reasons is the use of AMD's 90 nm SOI process.  Over the past year AMD has been working very hard to get its 90 nm process running at its fullest potential, and they have pretty much achieved that.  Intel doesn't utilize SOI on their 90 nm process, and from all indications SOI provides around a 30% decrease in leakage over a non-SOI 90 nm process.  Now, there are of course many differences between AMD's and Intel's 90 nm process, but it seems that SOI does what it is supposed to do very well.  Also, the design of the Athlon 64 requires transistors to switch much slower than competing Intel products, so less power needs to be applied to these transistors, and the less power applied also means less leakage.  So, AMD has a nice one-two punch when it comes to power consumption and heat production.

AMD is not sending out any review samples of these processors, but since the Palermo core appears to be slightly advanced from the now common Winchester core, we can expect the new Semprons to overclock to impressive speeds.  The Sempron 3100+ runs at 1.8 GHz with 256 K of L2 cache, all at 1.4 volts.  Looking at what the Winchester cores could do, it is not that big of a jump to say that the 3100+ should be able to easily clock above 2.3 GHz, and for those with motherboards and memory that can run at 300 MHz HTT and 300 MHz memory speed, 2.7 GHz should not be hard to achieve with good cooling and a tad more voltage applied.  This would be a very fast little system on the cheap!  For those with systems that don't clock nearly as high, 275 MHz bus/memory and 2.475 GHz for the core are still pretty snappy.

In speaking with AMD, it is very clear that their main goals for the first half of this year are to transition all their product lines to 90 nm, and to introduce dual core products.  While Intel has been making a lot of noise about dual core at their Intel Developers Forum, AMD has quietly been producing dual core products to be released around this Summer.  AMD has been showing off their dual core chips, and have shipped pre-production versions to their partners for testing and integration into new and upcoming products (such as the motherboard guys).  What review samples we will see in the near future will be entirely dual core and FX based products (so no new Sempron reviews... sorry).

March 2, 2005

Industry News - Josh

Things are again heating up in the computer world, now that the first warm breezes of Spring are upon us (at least in some parts of the world).  Intel is holding its IDF, and they are showing off their dual core products left and right.  Intel expects to ship its first dual core product by the end of Q2 this year, which may be well ahead of what AMD can do at this time.  The first product will essentially be two Prescott 1 MB processors on one die (with all the associated interconnects to accomplish this).  The maximum speed so far that this product will reach is 3.2 GHz., which is quite a bit lower than the top end Pentium 4 to date (the 3.8 GHz J model Prescott).  Heat and power are going to be pretty sensitive topics for this product, and we can expect them to run pretty toasty at 3.2 GHz.  Intel has been working on the heat and power issues, but I don't think that they were able to bring out the magic wand for this one and make it a cool running product.  The generation after that will be the Pressler processor, and it will actually be 2 x 2 MB L2 cache processor cores, but not on one die.  These will be independent, but connected to a virtual front side bus so they can communicate with each other, as well as with the system chipset.

Apparently the increased competition from AMD has lit a pretty impressive fire under Intel, as their upcoming products for the next two years look to be very impressive.  Intel is well on its way towards a solid performing 65 nm process, and at this Spring's IDF they are showcasing quite a few products based on 65 nm.  They even have their Pressler dual core processor running at 2 GHz on the 65 nm process (which is actually quite impressive considering this part will not officially be out for at least another 16 months).  Intel's Pentium M based products are also continuing to improve, but now I wonder how long it will be til we see the Pentium M products truly on the desktop (where I believe they are sorely needed).  Intel also was showcasing new chipsets, namely the 955 series and the Napa platform (next generation mobile).

Something that hasn't really been talked about much is if these new dual core processors will continue to use the LGA 775 socket.  I am assuming that these new chips will be able to drop into the 775, but I am not entirely certain.  I know that none of the current Intel chipsets (865, 875, 915, 925 series) support the dual core products, and that the 955 series will be needed for that.  It will be interesting to see if Intel will require yet another socket type to support these processors.  This is an area where AMD has really excelled at in the past.  The Socket A infrastructure lasted years and years, and it appears as though the current 939 and 940 architectures will also last for many years before requiring a significant change (even the Socket 754 should have a long lifespan in the value market with the upcoming Semprons).  Admittedly, the Socket 478 was a long legged architecture, and it is still in use today with all the excellent i865/i875 motherboards still hanging around.

NVIDIA is also showing off its Pentium 4 based nForce 4 product, and they are going back to a two chip architecture.  While having a single chip do everything sounds good on paper, my gut instinct is that it adds way too much complexity to a board layout, as well as the inability to mix and match features and product SKU's easily with both a northbridge and a southbridge.  Since HyperTransport is such an efficient interconnect, not to mention high speed, it makes little sense to me to continue putting every feature on one piece of silicon that has every trace on the motherboard running to once place (a design nightmare if you ask me).  Perhaps with this generation of chipsets from NVIDIA we will see a return of a two chip motherboard design for NVIDIA's AMD products.  It would make a lot of sense for NVIDIA to have one northbridge for Intel and AMD products, yet be able to share three to four southbridges with a variety of features between the two architectures (which is what the rest of the industry does... for some logical reason).  This would give the motherboard manufacturers a lot more control over what features they decide to implement on their product SKU's.  Something that is of interest to me is that pretty much all of NVIDIA's latest chipsets are made on the 130 nm process at TSMC, which is pretty much an industry first as only now is Intel moving their chipset products to the 130 nm node.  SiS, VIA, and ULI are all sitting around the 180 nm node with their products.

PCI-E is now finally hitting the big time.  With all the support from Intel in the past year, and now the addition of a sizeable collection of AMD based PCI-E products, AGP is now finally taking a backseat.  This transition has been a long one, but with the introduction of the X850 series of boards from ATI, as well as the X800 and X800 XL boards, the midrange and high end are well represented in PCI-E.  NVIDIA is also stepping up its PCI-E production with the 6800 series of boards, but the 6600 GT has been the darling of the ball when it comes to price/performance in the $200 range.  With the nForce 4 for AMD now being very well represented in the market, we should see AGP fade even faster.

LanFEST 2.1 and the PenStarsys.com Hardware Workshop - Josh

This weekend was the LanFEST 2.1 which was put on by the UW PC Gamers group.  Over 100 people total showed up for the LAN and X-Box tournament, as well as a goodly group of folks for the hardware workshop that I put on.  We talked about overclocking, AGP vs. PCI-E, trends in the industry, memory, power supplies, and pretty much everything under the sun when it came to the technology behind PC's.  I gave away prizes from Corsair (XMS 3200 XL Pro DIMMS), Tyan (their i915G based desktop motherboard- which is quite honestly packed with features), Abit (their Fatal1ty AAX8 motherboard- another board packed with features), and a couple of heatsink/fans from Thermalright.  There was also a lot of swag being handed out during this time, and it was a great success overall.  The entire workshop lasted about an hour, and I am looking forward to doing one again for the next big LAN that the UW PC Gamers are planning to put on.  If you want to see a few pictures of the LAN, go here.

February 24, 2005

LANFest and Graphics News - Josh

Only one day left to register for LanFEST 2.1 held by the UW PC Gamers, as well as my workshop!  The LAN itself will be from 6pm on Friday til Midnight on Saturday (a full 30 hours of gaming, for those so inclined).  Being an old man, I will most likely take a couple of breaks from the gaming action!  We have Corsair, Tyan, NVIDIA, and Abit as sponsors, and at the workshop I will be giving away several pieces of hardware as well as a bunch of other bits of swag!  Everyone should come away with something from the event.  The actual workshop itself will start at 3pm Saturday, and anyone attending the LAN is eligible to attend the workshop.  So, even if you only want to attend the workshop, you will need to pay the $10 entrance fee (which goes mainly to charity, and to networking equipment for future LANs).

There are a couple of bits of graphics news that caught my eye.  The first is that we will finally see 512 MB cards, with ATI being the first to show one (and giving it away to a lucky gamer attending the TXGF).  This looks to be a X850 XT PE with 1.2 GHz GDDR-3 memory, and a couple of sites are hinting that they have this little number on their test beds as we speak.  Expect to see some numbers from these products very soon (unfortunately I don't have one, so don't come looking here for numbers).  Gainward is also producing a 512 MB version of its GeForce 6800 Ultra product, and they expect these to hit around March.

That being said, the first solid proof of a new NVIDIA product has hit the web.  The latest leaked 75.90 drivers have a product called the "G70" inside.  Now, there has been a lot of speculation around about NVIDIA's next generation products, and I can share with you what I know (and then speculate afterwards).  There originally was going to be a basic refresh of the NV40 chip, and this was supposed to be about 10% faster mostly due to clock optimizations.  It would have also had a fixed video unit (as the one in the NV40 is only partially functional), and a couple of extra tweaks.  It was not going to be a huge change from the original, and NVIDIA probably wisely decided to drop that project and continue churning ahead with the NV40 and its NV45 variant.  This is where the speculation begins.  There also appeared to be plans for an "uber" NV40 which would take the architecture, tweak it a bit, fix the bad parts, and give it more than 4 quads of pixel pipelines (as well as a couple of extra Vertex units).  There was some confusion as to if this would be called the NV47 or NV48.  But, it looks as if this project was cancelled as well.  The logical presumption here would be that the NV50 is coming this spring.  I really don't think this is a case though...

Here are my thoughts.  The NV50 (or it has probably had a new code name by now) is still a future product for NVIDIA that will embrace the new technologies in DirectX Next (or whatever they are calling it now).  Since that technology will not be exposed for quite sometime, NVIDIA appears to have changed their roadmap around dramatically.  The G70 does appear to be their next high end part, and it also appears to be a very significant upgrade from the NV40 architecture.  This means higher performance, more functionality, and most likely one of the first chips to run on the 90 nm process (probably TSMC, as I am not sure if IBM has a whole lot of space on their 90 nm lines for a company like NVIDIA).  This will have more than 16 pixel pipelines, and probably around 8 vertex shaders.  I think we will finally see a new multi-sample unit, so we can expect to see a programmable AA unit, as well as gamma correction thrown in (and possibly a few new wrinkles to AA- though we have been hoping for a true stochastic setup for a LONG time).  The video encoder will be fixed of course, and I think we can expect to see a few more features thrown into that arena.  Now that HDTV is becoming very common, I think NVIDIA will probably spend a lot of time making sure that this functionality works flawlessly in this new chip.  Now, none of this is of course confirmed, but from keeping my ear to the ground, this certainly seems possible.

ATI is not far behind with the R520, and we most likely will see this product before NVIDIA's G70.  The R520 still appears to be the same classic architecture for separate pixel and vertex shaders.  While the R500- Xenon chip, which is going into the X-Box 2 looks to have the unified shaders, the R520 doesn't appear to follow that technology quite yet.  This will be ATI's SM 3.0 part, and there is a great debate going on whether it will still feature 16 pixel pipelines and 6 vertex shaders.  In my opinion, it could go either way.  Adding SM 3.0 as well as other features to keep up with NVIDIA, ATI will already be adding a lot of transistors to the mix.  It is hard to say if they can afford to add more than the 16/6 units, but since they are using TSMC's 90 nm process, they could possibly do just that.  If they expect TSMC's 90 nm Low-K process to be fast and efficient, perhaps ATI will simply concentrate on clockspeed to keep up in terms of performance.  It is very hard to say at the moment what exactly each company is doing, as it seems leaks are becoming scarcer and scarcer.  Still, this Spring we will have two new architectures to play around with from both NVIDIA and ATI.  Can't ask for much more than that!

February 21, 2005

Monday Observations - Josh

Well, I may be very wrong with the FX-57 being released next month.  Some people in the know simply asked what exactly was I smoking to come to that conclusion.  I am guessing that the latest cores that AMD has listed on their roadmap will start to make their way into product sku's without us knowing much about it, but I am sure that many will be keeping their eye out for them, and once these new cores are spotted, we all will know about them soon enough.  The Opteron 252 release appeared to be a success, but not much fuss was made about them.  I really wish that the folks who got their hands on these parts would have been so kind as to try to overclock one or two of them to see how high they would go.  There are many that still consider the 3.0 GHz mark well out of the question for the current 90 nm Athlon 64's, but I think that with heroic efforts 3 GHz will be readily attainable for most of the 90 nm 2.6 GHz products that will make it onto shelves by the end of the first half of this year.  I guess we shall see.

NVIDIA looks to have hit another homerun with the nForce 4 series of chipsets.  NVIDIA has basically said that the huge demand for these products has helped their bottom line hit record heights.  Some market research indicates that NVIDIA has captured 48% of the Athlon 64 market, and this market continues to grow at a furious pace (now that AMD has essentially discontinued the Athlon XP series of chips).  The nForce 4 series of chips are very, very impressive and their performance appears to be top notch.  Added features such as the networking engine certainly don't hurt, but many are still bemoaning the death of the SoundStorm technology.  Still, there do appear to be some decent Audio products to be hitting the motherboard scene, with VIA's Envy series leading the charge for high quality sound.  The nForce 4's have really hit the market in force, and there is a very good selection of AMD/PCI-E motherboards featuring this chipset (from around $105 to $200+).  MSI, Asus, and Gigabyte all have SLI boards out in force, and DFI recently released their SLI product.  All in all, the selection is quite good (with the Chaintech VNF4/Ultra looking to be a very affordable and robust product).

ATI does not appear to be having the same problems with the X850 products as they did with the initial release of the X800.  As you all well know ATI had a very hard time getting enough product into OEM's hands, as well as address the retail market.  For around 9 months the Radeon X800 XT PE was simply not available, and those that were retailed at around $750.  Now that ATI has relieved that pressure with the new X850 based boards, we are now starting to see a very healthy supply of the X800 XT/PE's, as well as the X850 XT/PE's.  The average user can now buy a X800 XT PE for around $510, while the X850 XT PE can be had for around $620.  While the PCI-E parts still sport a nasty premium, they are at least available!  It certainly appears as though ATI has pushed through their issues with the fabrication of their high end AGP parts.  Unfortunately, the midrange ATI parts still have not really made a name for themselves.  While the X700 parts are solid, they are usually overshadowed by the GeForce 6600 GT's.  The X800 and X800 XL are supposed to offset this performance difference, but so far these cards are in low supply and their prices are each around $50 to $60 higher than MSRP (eg. the X800 XL is supposed to be a $299 card, but it often retails for $360 or more).  Once these boards get into better supply, we will see the price drop.  Until then, many are turning to the GeForce 6600 GT for their $200 to $250 video card fix.

That is all for now, but more should be up soon!

February 14, 2005

AMD Releases New Opterons - Josh

AMD officially released their 852 and 252 Opteron processors.  These are based on the new Athens and Troy cores (90 nm SOI, more tweaks, SSE3 support, improved SSE performance).  These processors run at an impressive 2.6 GHz, and while this is speculation, are probably the first examples of the full 90 nm strained process (previous 130 nm and 90 nm products have used bits and pieces of the strained design, but these new processors make use of both stretched and compacted silicon).  Someone with much more money than myself, and an older socket 940 motherboard, should buy one of these and see how far they can push it!  I fully expect them to run at stock voltage at 2.8 GHz without a problem, and I wouldn't be very surprised to see 2.9 to 3.0 GHz with some heroic cooling measures.  This type of release is very reminiscent of the Athlon MP, and several months later the release of the Athlon XP (introduce the lower volume, higher margin server part, and then work on yields and speed bins and eventually transition the technology to the desktop).  Now, I am not sure if the Athens/Troy/Venus cores are the same as the San Diego (desktop Athlon 64 90 nm SOI 1 MB L2), and will need to get some verification on it.  Obviously the Athens and Troy cores have the extra HyperTransport connections for their multiprocessing capabilities, but Venus and San Diego could be identical.

The Winchester core (that which powers the 90 nm socket 939 processors) has been a very large success for AMD.  The use of SOI has helped to squash the power leakage problem that Intel is facing with their 90 nm process without SOI.  The 90 nm Winchesters have been shown to run cooler, and more power efficient, than the 130 nm Newcastle chips.  This is a very good step for AMD, as these cores have been shown to run upwards of 2.7 GHz with very few problems (my 3200+ that is at 2.0 GHz can comfortably run at 2.5 GHz with a bit more voltage and cooling- all this on not a great motherboard).  This all speaks highly of AMD's 90 nm process and overall Athlon 64 design.  When the new San Diego/Venice/Palermo cores come out (the same basic tweaked 90 nm core that the new Opteron x52's share), then we will probably see clockspeeds grow to 2.8 GHz, and the total discontinuation of all 130 nm Athlon 64's (though they will still be available in the channel).  AMD has said that it plans to have the majority of production for all Athlon 64 products to be 90 nm by the beginning of 2Q.  From all indications, this appears to be the case.  The Opteron x52 series is the first shot in what is going to be essentially an entirely new lineup of Athlon 64 cores (from the Sempron to the Athlon FX).  AMD should be able to produce at least 25% more processor cores once they are switched to 90 nm production.  In fact, I would say that dice production will not be the limiting factor, but rather the packaging process that AMD uses for its processors.  Once AMD switches to 90 nm, their products will become a lot more attractive to major OEM's because there should be greater availability of processors up and down the line.

From what I am gathering, there do appear to be more tweaks and optimizations in these new 90 nm cores.  The addition of SSE3 and the performance boost that the entire SSE sets receive should be welcome news to many people that utilize applications optimized for SSE/SSE2/SSE3.  The memory controller has received more work as well, and it is rumored that it can run 4 dual sided DIMMS at full speed (often when more than 2 DIMMS were used in dual channel Athlon 64's it dropped the speed down to PC2700 or PC2100, depending on the motherboard and processor being used).  There are also a few performance increases with this revised memory controller, and it should have a small, but noticeable, performance increase overall as compared to a similarly clocked 130 nm or Winchester based Athlon 64.

Under the radar AMD has also released several more Socket 754 based Semprons.  The 2600+, 2800+, and 3000+ all flesh out the Sempron line on the 754 platform.  This is sorely needed, as AMD is desperately trying to ramp down the Athlon XP/Sempron line.  AMD wants to try to move on from the Socket A architecture, and focus on its 754/939/940 platforms.  754 looks to be very robust for the low end and midrange for the time being, and the 3700+ is still a very fast processor.  It will receive PCI-E treatment in the near future, and will eventually become the exclusive low end platform for AMD products (namely Semprons).  For the time being, there are many happy customers for AMD's Socket 754 products (me being one of them).

I fully expect to see a new launch from AMD this March, mainly due to the increased pressure Intel will put on with their 6x0 series of Pentium 4's.  These revised Prescotts feature 2 MB of L2 cache, and are supposed to give a healthy increase in overall performance.  We shall see.  In the meantime, expect to see the FX-57 based on the San Diego core hit the web in force next month.

February 9, 2005

Official LanFEST 2.1 Hardware Workshop Page - Josh

I have finally been able to nail down enough information to put up a page describing what will be happening at the hardware workshop that we here at Penstarsys.com are putting on at the LanFEST 2.1 event.  The UW PC Gamers group are putting on the LAN, and they were kind enough to ask me to participate by putting on this workshop.  We have several major sponsors already, and are expecting a few more to possibly come out of the woodwork as the event gets closer.  For those of you that expect to be around the Laramie area on February 25th and 26th are more than welcome to sign up for the LAN and attend the workshop (the LAN costs $10, and the workshop is free for any attendee).

Hope to see you all there!

February 8, 2005

LanFEST 2.1 Hardware Workshop - Josh

I was finally able to hook up with the local UW PC Gamers club here in Laramie, and they are holding a pretty significant LAN on Feb. 25 and 26th.  I donated some networking gear to them to use as the backbone, and we have agreed that I would be able to hold a computer hardware workshop on Saturday afternoon.  So far I have gathered up Corsair and Tyan as sponsors for the workshop, and I am working on a couple of more companies to hopefully step up to the plate with fun things for the attendees.  There will be hardware to give away, as well as other prizes, so if you are in the Laramie area (or anywhere close), please sign up for the LanFEST 2.1 and come to the workshop!  A full list of the prizes to be given away will be posted sometime before the event.  Stay tuned, it should be a lot of fun!

February 1, 2005

Corsair News - Josh

Quite a few things have come across my desk as of late regarding Corsair, so I figured I would pass them onto you.  The first is that they are now offering a 2 GB Flash Voyager for not a whole lot of scratch.  I have a 512 MB version on the table as we speak, so I will be commenting on this at a later date.  Needless to say, this huge amount of flash storage has some very attractive applications.  The case appears fairly rugged, and it is still only finger sized.

The next release is of the Cool system.  Corsair has had a few entries into the water cooling market, though with some limited success.  Previous versions have suffered a few major problems that made it unattractive to many users (eg. the fine finned copper cooling that kept corroding and clogging itself).  This release looks to be quite different from everything else they have done in the past.  This small, enclosed system hooks onto the back of nearly any case, and gives better cooling performance than most forced air based coolers.  This doesn't appear to have the cooling capacity of other, larger systems, but it should do very well and not cause your system to weigh a metric ton.

The final story here is about the official release of the Xpert series of DIMMS.  I am sure you have seen previews of this product, with the very special LED readouts that can show all kinds of information about the memory.  These are quite a step up from the XMS XL Pro parts in terms of functionality, and the user can interact with these DIMMS through a handy Windows application called Memory Dashboard.  First shipments are out the door, and more review samples should be sent out soon (hopefully to my front door).  Expect to see these on sale within the next two weeks (yes, the infamous two weeks).  You can read more about these modules here.

Corsair certainly seems like a company on the move, offering a wide variety of products (and learning from past mistakes).  I am quite happy to have them as an advertiser!

NVIDIA nForce Professional Chipset - Josh

NVIDIA quietly snuck the nForce Professional series of chips to the market, but evidence of its existance has been around since August.  If any of you remember the first showing of the Tyan dual PCI-E 16X boards, these were the first samples of the 2200 and 2050 chips.  I was initially quite confused about the setup, as I did notice two of these chips on the Tyan board, while none of the other SLI boards had anything remotely like it.  Being silly, I didn't question these images and just assumed that they were two nForce 4 PCI-E chips each attached to the individual Opteron CPU's.  Well, I was obviously quite wrong.

These two very distinct products look to fill a gap in the Athlon 64/Opteron workstation market, and they seem to address it quite well.  The two different chips provide a huge amount of features (both single and together).  It is still a single chip architecture, but up to three 2050 chips can be used with a single 2200 chip, which can give an amazing amount of bandwidth to the entire system (think 80 PCI-E lanes).  Each chip also supports the 1 GHz Hypertransport bus (though Opterons still use the 800 MHz version).

ATI and the X850 - Josh

Well, the X850 XT and X850 XT PE's are now out in decent quantity.  I can now order them from one of my distributors, and they are starting to hit the web pretty hard.  This seems like a very positive move from ATI, especially in light of the X800 shortages since its introduction.  The X850's are based on the new R480 core, which is essentially a respun R423 (supposedly a metal layer change to improve yields and bins).  This bodes well for future availability (though prices are still sky high).  Unfortunately, we haven't seen many of the new X800 parts based on the R430 core (110 nm version).  The X800 XL is supposed to be hitting the $299 price point, but so far the few that are out there are closer to $450.  Not exactly a high-midrange card price.  Still, things look much better in terms of availability for ATI this first half, and I will be going over more of this story in a future article.

Misc. News - Josh

The fellows at 3D Chipset have been very busy as of late with new files that should fulfill all your driver update needs.  The latest NVIDIA 71.80 drivers are there to download, and more interestingly this home-made SLI program enabler.  It is well known that only games with supported profiles in the NVIDIA drivers are SLI enabled, but this little tool should allow the average user to pick and choose what non-supported titles they have can be played with SLI.

The Tech Report has done a small GeForce 6600 GT AGP roundup that should answer quite a few questions for those looking to purchase a AGP 6600 GT.  They have products from BFG, XFX, and Albatron on the table, and the results may actually surprise you.  Definitely a good read!

MS from Lost Circuits takes a look at the aforementioned Radeon X800 XL (this one from Sapphire).  The price/performance standpoint of this product looks to be very solid, but availability is now the deciding factor.  Perhaps in a couple of months we will have adequate supplies of this product, but only time will tell.  Still, a good review by MS, well worth the time spent reading.

Thats all for today, hopefully I can finish the article I am working on in a short period of time!

Copyright 1999-2005 PenStar Systems, LLC.