GTX 270? More shader processors and rumors ahead

NVIDIA will be quietly beefing up the power of their GTX 260 cards later this month, suspects the website Fudzilla. Currently, the GTX 260 has a GT200 GPU core that has 8 banks of 24 shader processors enabled, leading to 192 shaders. The new GTX 260 is going to have one less bank disabled, so that means it'll have 9 banks of 24 shader processors, which adds up to 216 shaders altogether. 

At this point, it is a bit too early to tell what the new naming scheme may be. Perhaps it is possible that NVIDIA will do what they did with the 9800 GTX -- to better compete with ATI's HD 4850, they added a bit of horsepower to the 9800 GTX, and renamed it the 9800 GTX+. So maybe we will have a GTX 260+ ?  However, the GTX 260 has not been out all that long though, so NVIDIA might be hesitant to rename the new GTX 260's to prevent annoying customers that have already purchased one. GTX 270? Who knows -- NVIDIA's naming schemes have been all over the place during the last six months, making it fairly difficult to anticipate their next name.  

Whatever the new name will be (or whether there will even be a new name at all), the revised GTX 260 will be selling for about $50 more than it is now. If the new GTX 260 sells for $300 or $350, it will compete well against the HD 4870 in both price and performance. If these rumors turn out to be true, it also seems reasonable that NVIDIA would further cut the price of the GTX 280 with the release of the new GTX 260.

As many of you know, the ball is in NVIDIA's court right now. The GTX 260 and GTX 280 came out in June, and where soon after upstaged by the launch of ATI's HD 4850 and HD 4870. While the GTX 280 was much more expensive than the HD 4870, it did out-perform the ATI card in a reasonable number of benchmarks, making it an attractive purchase to some willing to pay top-bucks for the best. The GTX 260 on the other hand, was less attractive, and less competitive -- it was more expensive than the HD 4850 and HD 4870, and in most games, slower than either of those ATI cards.

What a surprise... not... Nvidia gets QPI license, X58 no longer needs nForce 200 for SLI

Looks like the dumb "war" between Intel and Nvidia over SLI on X58 has come to an amicable cease-fire.

Digitimes reports that Nvidia's director of Technical Marketing for MCP products has stated that Nvidia will authorize native SLI support for X58 without requiring motherboard makers to put an nForce 200 chip on the motherboard - however it will require an SLI license from motherboard manufacturers, and those that license SLI will get to put a magic "key" into their BIOS that Nvidia's drivers will require in order to run SLI.

The SLI license will cost motherboard makers some money, but presumably less than the $30 Nvidia was going to charge for putting an nForce 200 chip on the motherboard.

Peterson also apparently said that Nvidia "is not currently planning to launch any chipset supporting QPI" but will launch a "DMI" based socket 1160 chipset later.

Frankly, I expect Nvidia to crank out a QPI based chipset as soon as it can if it expects to remain in the chipset business.

On a nice note, this means that we will be able to benchmark SLI vs. CrossFire on exactly the same X58 platform in the future!!!!!

War of the Words over GPU's continued

PCPro is reporting that Nvidia was expending a fair bit of effort cautioning people attending NVISION to take the Larrabee hype with a grain of salt - they apparently pointed out that not much in the way of detailed specifications have been released yet, and that there is no magic tool to make a highly parallel x86 processor be a great GPU.

Interestingly enough, with massive numbers of stream processors, Nvidia is in a bit of a "pot calling kettle black" position, as it faces the same problems with factoring software over large numbers of processors.

Peter Glaskowsky, a GPU architect and blogger, predicted that the "large" Larrabee of 2010 will have roughly the same GPU performance as a 2006 GPU from ATI or Nvidia.

Thanks for the correction Peter, and thanks for the links - good information in them.

Peter's blog post #1 & blog post #2

Nvidia also admitted to having underestimated ATI, for which it is now paying due to the superior price/performance of the 48xx series.

Low cost video goodness

Fudzilla has an interesting rumour about the upcoming RV710.

Apparently the RV710 will have 120 stream processors and very low power draw, and as such will give signficantly better performance than the Radeon 3400 series that it will probably replace.

Fuad thinks it will be called a "Radeon 4450" and sell for <$60, going after the 8500GT/Geforce 9400GS

Post news here

Enter quotes using <blockquote>...</blockquote>

Overclocking industry wastes no time cranking up the potential.

The HD4870X2 and 4850X2 have made their debut now, and everyone is already pleased to see how far these things are going in the realm of performance, but for those who are struggling to accept the figures presented, CoolIT has gone ahead and launched their new liquid cooling system for the 4870X2 with a current report of a 13.5% increase in clock and a healthy 11% jump in the memory speed.

Right now the current figures report a stable 880MHz GPU and 1GHz GDDR5 report from their samples, of course enthusiasts have their ways of using alternative cooling solutions through these outside of basic water treatment and given that AMD helped co-develop the system it seems like they should have quite a nice starting point for where the sensitive spots are with the X2.

For 3DMark fans they upped their lab score from X7125 at stock to X8148 which is quite a nice leap by them. Personally I can't wait to see what people in the field start doing with these coolers since the initial reports are usually a decent baseline for what to expect without a huge investment into mixtures. As usual with many CoolIT systems, we don't have a pricing figure to give out but their website will have it available to the masses come September. Those looking to invest in it should keep in mind that the cooler is a dual drive bay solution though so some space will need to be available for it.

Most of the faulty chips found in laptop graphics chips

For about two months now, a few tech news websites have been writing about increasing rates of failure for NVIDIA GPUs. Up until yesterday, there has been little official confirmation or visible action on the part of NVIDIA to address this issue. But yesterday, NVIDIA made a move to deal with the situation -- they have earmarked about $200M against their second quarter revenue to go towards warranty charges for people sending back faulty GPUs.

The most affected chips seem to be in laptops GPUs. It also appears as if the following models from HP might have NVIDIA GPUs that are more likely to die an early death:  HP Pavilion dv2000, HP Pavilion dv6000, HP Pavilion dv9000  / Compaq Presario V3000, and Compaq Presario V6000 series computer. And if you have one of these models from Dell, you might also be concerned -- especially if your warranty is running out.

The problem purporedtly is 'weak silicon' -- that the quality of materials used in the making of the GPUs was not up to snuff, and breaks down under steady use. The GPUs found in the 8400M and 8600M seem to be citied as the most prone to failure, while some websites such as the Inquirer say that the problem is not limited to laptops, and that many more GPUs are liable to fail (G92 and G94 GPUs, found in the 8800 GT / 8800 GTS 512 / 9xxx series, and more.)   

NVIDIA has been quoted by saying (on PC World) that this money set aside should be enough to cover the problem. Hopefully, for all parties involved, more NVIDIA GPUs don't punch out earlier than they should.

Q2 sees decline

CEO and President Jen-Hsun Huang revealed during NVIDIA's second quarter financial earnings call last evening the company's GPU shipments dropped 20 percent within the last quarter. Moreover, the average selling prices of the GPUs dropped by 25 percent. The loss incurred amounts to $120.9 million, contrasting the  $172.7 million of profit recorded the year previous.

All in all this hasn't boded well for NVIDIA's financial standings: company revenue decreased by five percent to $892.7 million (this time last year they stood at $935.3m).

This is particularly striking news because it's the first quarter in nearly six years they've recorded a loss. Reportedly the 'mood on the call was understandably sombre.' What goes up must come down, though, so it appears.

The losses weren't exactly unexpected, however: Huang pins it on a variety of factors: economic slowdown, large inventory of 65nm chips, over-pricing, underestimation of the competition (AMD), delays to integrated graphics chipsets for Intel processors, and more:

“The desktop PC market around the world weakened during the quarter. And our miscalculation of competitive price position further pressured our desktop GPU business. We have a great product line-up and, having taken the necessary pricing actions, we are strongly positioned again. Our focus now is to drive cost improvements and to further enhance our competitiveness through the many exciting initiatives we have planned for the rest of the year.”

This also follows the news of its higher-than-expected failure rates on some notebook GPUs in July (you can read the newest details on that situation here).

It seems it would be best for the company to look at this as an opportunity to learn from mistakes and move forward, however, this will take some time:

[W]hen asked about when he expects the market to recover and for desktop GPU shipments to grow again, [Huang] said “it’s hard to say when it will recover.” Huang mentioned that the company is ready to ship 55nm chips in volume, but because of a large inventory of 65nm products, 55nm products will not carpet bomb the market until later in the year – as soon as 65nm inventory has been worked through.

It's not all bad though: he said, “The notebook GPU, MCP, and Professional Solutions groups grew a combined 27 percent year-over year."

Free compilation of games, demos, and mod packs brings PhysX tech to PC

NVIDIA is pretty proud of its PhysX technology, it seems. "Interactive entertainment's most pervasive physics engine," it's used in more then 140 shipping titles for all current-gen video game consoles. Anyway, they want you to download the GeForce Power Pack, a free compilation of games, demos and mod packs for the PC which also brings PhysX to the world's most popular platform:

On the PC, PhysX technology harnesses the power of any CUDA-enabled general-purpose parallel computing processor, including any NVIDIA GeForce® 8 Series or higher GPU, to handle 10-20 times more visual complexity than what’s possible on today’s traditional PC platforms. All of the 80 million plus GeForce 8 Series and higher GPUs in the field are CUDA-enabled, the largest installed base of general-purpose, parallel-computing processors ever created.

NVIDIA claims "competitive solutions" do not offer hardware scaling capability; only PhysX can "leverage the best of both CPU and GPU architectures to deliver the ultimate, immersive, end user experience." Hmm.

Upcoming PC titles that will incorporate the technology include Cryostasis, Backbreaker, and Aliens: Colonial Marines, with nearly 20 more expected before the holiday season.

“Game physics is essential in enabling deeper interactivity and real-world effects in any game. Epic is pleased to offer PhysX as a standard feature within Unreal Engine 3 to enable such effects,” said Mark Rein, Vice President of Epic Games. “The introduction of GPU acceleration for PhysX promises both additional potential effects and faster performance. You can get a glimpse of the possibilities of what PhysX is able to do with the special levels for Unreal Tournament 3 where damage effects greatly enhance the gameplay.”

So, as of today, anyone with a GeForce 8 Series or higher GPU can benefit from this, and many power packs of this kind are planned for the future, too.

The PhysX-enabled content from this first GeForce Power Pack includes the following:

• Warmonger—Full free game! Destroy walls, floors, and whole buildings to open up new paths or close existing ones. Destructive power is more than eye candy here—it’s a tactical weapon in this ground-breaking action game.
• Unreal Tournament 3 PhysX Mod Pack—includes three maps with amazing effects that fundamentally change the gameplay (requires full version of Unreal Tournament 3)
• A sneak peek at the upcoming Nurien social networking service, based on the Unreal Engine 3 (with built-in benchmark)
• A sneak peek at the upcoming game Metal Knight Zero (with built-in benchmark)
• All new NVIDIA “The Great Kulu” tech demo that showcases the use of PhysX soft bodies in a real game play environment
• All new NVIDIA “Fluid” tech demo—a simulation of realistic fluid effects with a variety of liquids

In conjunction with the Power Pack, new WHQL-certified drivers are available which enable PhysX acceleration for all GeForce 8, 9, and GTX 200 Series GPUs. This new driver also adds support for PhysX-accelerated features in Ghost Recon Advanced Warfighter 2.

The company is boasting with these drivers much higher levels of interactivity, effects, realm, and speed, stating "the PhysX-enabled levels of Unreal Tournament 3, the GeForce 9800 GTX+ runs 180% faster than on the AMD Radeon HD 4850."

“Seeing really is believing and is the reason why we compiled all of this great PhysX content into a free download for our end users,” said Ujesh Desai, general manager for GeForce GPUs at NVIDIA. “We want GeForce owners to experience for themselves these amazing effects to get an idea how PhysX will make games much more lifelike in the years to come. Physics-accelerated content is already here, and there are a ton more titles on the way. We can’t wait for our customers to jump in, get wet, and tell us what they think!”

If you're buyin' it, feel free to download via the location below.

Takes away some of the sting from watching GTX prices fall

The GTX 280 and GTX 260 came out about a month ago now, and they were both nice pieces of hardware. Fast pieces of hardware? Certainly. However one thing both of these cards were not, was inexpensive.

ATI came out of the left side of the left field with a zippy HD 4850 and a thunderous HD 4870, which stacked up well to the GTX 260 and GTX 280 for much less. Prices of these cards did not necessarily reflect relative levels of gaming performance. Originally, the GTX 280 was $650, the GTX 260 was $450, while the HD 4850 was $200, and the HD 4870 $299.

NVIDIA just could not compete effectively at their initial prices for their products. So, while prices often dropped quickly for video cards, the price of the GTX 280 has taken a precipitous plunge. Bi-weekly price cuts have led today for the GTX 280 going for $450 on online retailers, and even less in some places, with the offer of a mail in rebate. The GTX 260 as fallen to two thirds of its original price, now going for $300. 

Perhaps fearing a customer backlash, or perhaps out of the kindness of their hearts (or maybe a mix of both) board partners EVGA, XFX, and BFG have all offered rebates and $120 cash back if you bought one a GTX 280 or GTX 260 recently.

Here are the links to further information: EVGA fans go here, XFX followers can go here, and BFG'ers check out this page.

Competition leads to price drops

No doubt feeling the heat from ATI's fast recently released HD 4850 and HD 4870 video cards, NVIDIA has dropped the price on the GTX 280, and GTX 260. The GTX 280 is now selling for $500 at many retailers, which is $150 less than it was selling for when it was released only a few weeks ago. The GTX 260 is now going for $299, which is $100 less than it previously retailed for.

If you read our review of the HD 4870, you'll see that the HD 4870 definitely seems like a better deal at $300 than the GTX 280 was at $650 -- so this price cut was almost required. With these price cuts, and the 9800 GTX+ video cards, the NVIDIA / ATI battle is no longer as lopsided as it first appeared when the HD 4850 came on the scene.  

Unfortunately NVIDIA is in a bit of tight spot right now -- there GTX 280 and 260 cards are fairly expensive to be produced, yet are being beaten by the cheaper HD 4850 and HD 4870 cards in many benchmarks. The company's shares have taken a beating recently as well, perhaps as a direct result of this situation.

NVIDIA reportedly has plans to gain efficiency and lower costs by switching to a 55nm manufacturing process.

Nvidia caves in, will allow SLI on upcoming x58 motherboards for Intel

Nvidia blinked first.

Intel did not want to give Nvidia a CSI license because Nvidia would not allow SLI on intel motherboards with multiple GPU slots.

Nvidia did not want to give Intel an SLI license because it wants to sell motherboards... it would REALLY prefer if people could only get SLI if they bought Nvidia motherboards.

Unless there is an unannounced backroom deal giving Nvidia CSI, Nvidia caved, and there will be SLI Nehalem motherboards - with an Intel x58 chipset and an Nvidia nForce 200 chip that has "patented SLI technology for graphics bandwidth management and multi-GPU peer-to-peer comunications both required to optimize graphics performance" - which is marketspeak for we would only let Intel do SLI if we got to add a small chip for which we will charge an arm, leg and reproductive organs for.

Technically, without driver games/restrictions (read DRM) there is NO reason why SLI should not work on ANY multi-slot PCIe motherboard that supports at least 8x on the slots. None. Zip. Zilch.

Remember the SLI patches that were floating around on the net for non-Nvidia chipsets a while ago? Legal threats and driver changes killed those.

Actually, the nForce 200 chip WILL have one technical result: it will slow down the GPU's - instead of talking to 16x PCIe 2.0 from the x58 directly, a single 16x PCIe 2.0 slot will be "splt up" by the nForce 200 to two PCIe "16x" slots, just like on the 780 motherboards.

Continuing problems plague the company even after after the merger

While it's no secret that many felt that the AMD / ATI merger wasn't going to be good, I don't think many saw it spiraling out like this, today reports come in from the AMD camp about Q2 bringing down a $950 million hit. The company plans to take an $880 million impairment charge on parts of it's ATI acquisition relating to handheld and DTV units of their Consumer Electronics Group, this added with other charges bring the total filing to just short of the $1 billion mark with the SEC.

The apparent upside is that while this is only a preview of things to come the 17th the company is planning to make an impact on that number when it closes sales of some of their 200mm wafer fabrication tools for a $190 million jump allowing the company to hopefully spin it into something productive as they are in the midst of restructuring the company to get it back in the game. Shareholders seem to have felt some reassurance with the company stock rising from $4.80 to $5 although this falls short of even their double digit pricing last year.

The news would probably bring some bit of joy to NVIDIA but they currently face their own issues with a price war taking tolls, entire batches of GPU's being labeled as defective and a new lawsuit on the floor from Rambus relating to patent violations in their memory controllers. It seems that all camps are taking hits right now in the marketplace in the midst of an already troubled economy. One would hope that these companies find resolve to keep technology going and rise out of the trouble they're currently dangling in right now.

We'll follow up with more as AMD makes their disclosure on July 17th. Hopefully there's more details to what entirely brought about the $880 million charge.

Troll Alert

Don't you just love it when standards bodies are snookered into adopting standards where there are submarine patents awaiting?

Rambus claims that Nvidia (and everyone under the sun probably) is violating 17 Rambus patents by interfacing to SDR, DDR, DDR2, DDR3, GDDR and GDDR3 memory before paying it to license "the technology".

Hopefully the claims get tossed out of court. Rambus in the past has run into trouble for trying to collect on patents it owns but did not disclose ownership of when sitting on memory standards committee's.

Cooler and faster GPU's ahead?

Digitimes reports that in order to get more chips out of a wafer, increase yields, and hopefully reduce the cost of production, Nvidia is expected to go to a 55nm process in the second half of this year, and is actually expected to have 4-5 new GPU's being produced on that 55nm process by the end of the third quarter of this current year.

Nvidia and AMD are also both expected to complete the tapeout of some 40nm parts by the end of the year - it will be interesting to see who gets the first working 40nm parts to market.

Apparently Nvidia's GT200 and G92b GPU's have already been "ported" to the 55nm process and the mid-range G96, expected to launch by mid July, will be made in both 65nm and 55nm varieties. Upcoming G94b, G96b, and G98b GPU's scheduled for post-August launch will be 55nm parts.

The die shrink from 65nm to 55nm is expected to lower costs by approximately 20%, and will also lower power consumption and will also likely allow for an increase of clock speeds.

New and different RV770s and RV700s coming in August

According to the website Fudzilla, ATI has been holding back the floodgates of non-standard HD 4850 and HD 4870 cards in order to make sure all the reference boards clear out the warehouse first.

So far, not many non-stock and standard HD 4850s and HD 4870s have come out, but there have been a few exceptions. MSI released a HD 4870 with a mild, 30 MHz overclock, and as he talked about yesterday, Diamond is releasing a Black Edition O/C card. It also seems that Sapphire will be coming out with a HD 4850 with 1GB of GDDR3 -- this is probably going to be a popular configuration, and will probably be copied by a few other partners.

As you may have heard, the HD 4870 X2 (RV700) should be coming out within a few weeks. Supposedly the dual-GPU on one PCB HD4870X2 is about 10% faster than two separate HD4870's working in a CrossFire setup, thanks mostly in part to a whopping 2 GB's of GDDR5 memory. Of course, if it comes out, we'll review.

A little bit further into August, we can expect more unique offerings to be brought to market by ATI's board partners. Similar to the HD 3000 series, ATI will be allowing a lot of leeway with designs. One possible product might be a HD 4850 X2 card -- last generation, Asus was the only company to try its hand at the the HD3850X2, and this card was surprisingly nice. A HD4850X2 would fit well in the current line-up between a HD4870 and a HD4870X2, so with this generation doing so well for ATI, it seems likely that it'll be made, to give consumers an oppourtinity to buy a HD48xx card at whatever price-point they're interested in, be it $200, $300, $400, or anywhere in between.

Last time around Unika (a far Eastern card maker) put out a HD 3850 with GDDR4,so it seems likely that another possible upcoming design could be a HD 4850 with 512MB or 1GB of GDDR5.

On the NVIDIA side of things, you can certainly expect partners such as XFX, BFG and EVGA to offer extremely overclocked products, in order to compete with ATI's momentary upper-hand. EVGA has already unleashed a flurry of SSC and FTW branded cards, and water-cooled GTX 260 and 280 cards willl likely crop up from a few board partners as well (BFG springs to mine). Much further down the road (say a few months) a super-charged dual-GPU GTX 280 could become a possibility, but this is probably not likely due to the prohibitive cost of such a venture. However a sort of GTX 280 Ultra product (a GTX 280 [290?] with significantly quickened shader cores) seems more likely -- but this is just conjecture at this point.

Hot off the Internets

About a week ago, Diamond Multimedia announced that they would be offering an overclocked, "XOC Black Edition" of the HD4870. Initially, this card was only going to be offered through systems builder SmoothCreations, but the response to the super-charged card got many people excited, so, Diamond decided to release it to the masses.

There has hardly been any overclocked editions of the just-released and wonderfully fast HD 4870, so it'll be interesting to see what this baby can do. The XOC Black Edition features a core clock that will be running at 800 MHz (for a 50MHz O/C), while the quick GDDR5 will be flying at 1100MHz, which is 200MHz above stock (and can be considered 4400MHz effectively, when taking the HD4870 architecture into account.) Until the release of the HD4870X2 within the next month, this might be the fastest single card you can pick up.

Diamond sent us the first sneak-peek image of this new HD 4870 a few moments ago.

Sometimes skulls and flames are tough to pull off -- but with this card, and the look of the less expensive HD 4850, personally I think they pulled it off well enough.

No word yet on the price for XOC Black Edition.

Nvidia

Digitimes reports that Nvidia's partners in Taiwan are pretty low key about the defects Nvidia reported when taking a $150-$200 million charge this quarter for expenses related to unspecified mobile GPU problems.

Apparently the problem is in one of three places:

• GPU packaging (chips)
• solder bumping process (dip)
• PCB substrate problem (plate)

Digitimes sources, some investors, suggest that the problematic GPU's were the GeForce 8500M's, and that the actual problem was bump processing.

55nm 8800 GT

Really, we are in a sort of video card nirvanna right now. With cards as powerful as the HD 4850 for about $200, and the 9800GTX+ for not much more, this second half of 2008 is a good time to get back into PC gaming if you've been away for a while. 

One card that should be even less expensive then the aforementioned products is NVIDIA's upcoming 9800 GT. The 9800 GT is basically a die-shrunk 8800 GT. The 8800 GT has a 65nm GPU where the 9800 GT has a 55nm part. Apparently this video card has been peaking through the hedges for a while, turning up here and there, and should be coming not too long from now, although no official announcement has been made.

Chinese website Zol is responsible for the pic of the card above.

The 9800 GT apparently is reportedly set to the same clocks as the 8800 GT ( 600/1500/1800 MHz, [core/shader/memory]) but has Hybrid Power as well. Having a 55nm GPU, it would seem likely that this 9800 GT has the potential to be a good over-clocker. Looking at the prices for cards now a days, a price around $149 mark seems likely -- perhaps falling to $139 shortly thereafter.

The 9800 GT will probably slowly replace the 8800 GT -- although there is still a great deal of 8800 GT cards around, so it could take a while.   

Gamers rejoice!

Digitimes reports that AMD is planning to introduce the Radeon HD 4870 X2 by the end of July.

Graphics card vendors apparently expect evaluation samples in mid-July, and reference boards with a whopping 2GB of GDDR5 by the end of the month, at which point AMD will also start supplying the GPU's to video card vendors who want to make their own boards instead of rebadging the reference board.

The 4870 X2 is expected to be $499

Company uncovers problem to potentially cost them $150-200 million.

NVIDIA seems to be taking a bit of flack today, with stock prices dipping ever lower as the company announces a lower Q2 expectation due to pricing pressure and delayed product ramps. And they may also be taking another hit when announcing that they'll be earmarking $150 - 200 million for repair and replacement costs on their laptop line which is taking an unexpected toll from high yields of  GPU failure in consumer products.

The immediate question for many laptop owners is "which chips are they?" which would be a great question, unfortunately we're left only with the official word being "older graphics chips" -- which could range from a 9500GS and 9650 since those are technically old in the pipeline, or it could be a flurry of 8400 chips and 8600 chips as well as a huge slew of FX5200's from way back in time. I would say that getting their latest driver patch might be a good idea though given that it's set to kick in the GPU fan earlier with more sensitive thermal scaling.

For those already experiencing these failures, the company is pulling out $150-200 million to help repair and replace systems affected, this should hopefully translate into fast repairs and hopefully more focused cooling solutions to prevent it from occuring in the future. For me I've actually been checking out NVIDIA based laptops but I guess I'll be more focused on each cooling approach over how sleek or impressive it comes off on the outside.

Hopefully NVIDIA picks themselves up, they just started on their latest generation the GTX series and it looks to be doing them well, one would hope that we see some of that performance translate into the mobile market as the 9 series in the mobile line never really saw a dominant chip like the 8800.

Promised to be pretty slick and real fast

Diamond Multimedia (video card mavens), SmoothCreations (a high-end, "boutique" system builder), and a company called DangerDen (who are masters of component cooling) are teaming up for a triple tag-team on the recently announced HD 4870 XOC Black Edition.

Available only through SmoothCreations starting tomorrow, this card will have the potential to be the fastest HD 4870 yet released, with the core and shader clock set to a maximum of 950 MHz (which is 200 MHz above stock HD 4870 speeds.) Speeds this high were reached by using an new custom firmware for the GPU, and a hardcore cooler crafted by DangerDen (water cooled versions will be available.) You'll have to the extreme overclocking yourself though -- the default overclock of the XOC Black Edition will only be 50MHz above stock (800 MHz.)

While not stated explicitly in the press release we received, it is highly likely that the HD 4870 XOC Black Edition will only be sold as a part of complete computer from SmoothCreations.

The HD 4870 XOC Black Edition should become available sometime soon only from SmoothCreations.

Update: A few websites are reporting that due to the positive response to this product, Diamond will be offering the HD 4870 XOC Black Edition as a retail card, in less than two weeks.

NGOHQ reveals current progress on hack allowing 38xx cards to run AGEIA accelerated code.

It seems like once everyone really dedicated themselves to the programmable GPU idea the floor got opened to many possibilities, from new ways to handle graphics progressing to things like folding and data processing acceleration.

NVIDIA had taken their own route with how to expand, using CUDA they started on new ventures one of which included the purchase and integration of AGEIA into their company, later taking the PPU out of the focus and instead accelerating physics through CUDA on their hardware instead. The GTX was originally the first series to see this with some possible backwards compatibility to the G80 / G92 but they later changed their mind and fully supported the previous generation chips, apparently there in that opening NGOHQ got their shot at playing with the code needed to tweak the AGEIA code to support ATI 38xx series cards. Unfortunately because they have no working samples of the 48xx there will be no inclusion of it and one can hope that they do a release down the road for it when they get a chip.

I guess the big question would be of course, does it work? From the present screen released, it seems to work quite well given the huge CPU score on 3DMark Vantage.

While it's a bit unfortunate that PhysX doesn't run on a whole lot of games in the mainstream, it hopefully will pan out to be a good addition later. Either way I'd say it's safe to jump on the code as fast as possible seeing as NVIDIA struck fast at the website the last time it introduced the SLI patch for SLI on non-NVIDIA motherboards.

For those pondering the fate of CUDA on ATI/AMD, I wouldn't say it's safe to bet on it being 100% compatible anytime soon, but I would say that people doing apps on CUDA apparently will not be too taxed producing software that can scale to both chips as ATI/AMD has adopted OpenCL officially for their software support.

NGOHQ has no finalized date for the PhysX hack but I'd say it'd be safe to keep checking in on them through the coming weeks.

AMD's 4800 series to propel AMD forward?

Digitimes reports that AMD may be increasing its market share in the GPU marketplace from 30% to 40% based on the success of its 4800 series of GPU's.

With the ATI Radeon 4850 priced at $200 - and in some cases less - often outperforming the recently introduced Nvidia 9800GTX sales of ATI cards are expected to increase; even with Nvidia cutting the price of the 9800GTX to $199 the 4850 is likely to sell in very high numbers, and let's not forget that the 4850 has noticably lower power consumption than the 9800GTX.

The upcoming Radeon 4870 with its higher clock speeds and faster memory may very well outperform the 9800GTX+ that Nvidia introduced as a response to the 4850.

One thing is for sure - the introduction of the 4850 at its $200 price point is definitely good for us consumers, as it puts unprecedented GPU horsepower in our hands for two bills or less - from both AMD and Nvidia.

The TV Wonder brand is resurrected!

The new ATI TV Wonder HD combines a very capable TV tuner with a lower end Radeon GPU to bring a one slot solution to home theatre PC's.

The TV tuner portion is quite capable:

• Analog TV (off the air or cable)
• ATSC off the air digital SD/HD input
• ClearQAM unencrypted cable QAM SD/HD input

The TV output portion is quite nice too:

• enhanced DVD upscaling
• HD decoding
• HDMI with Dolby 5.1
• accessory kit with additional I/O jacks available

The GPU portion is not meant for heavy gamers, but is sufficient for casual lower resolution gaming

• 120 stream processors
• DX 10.1 support
• 128 bit memory interface

The board has a PCIe 2.0 x16 interface, and comes with more software than you can shake a stick at.

The only thing that it really lacks, which would have made it a killer card, is the ability to record HD content from a component video input - which is conspicously absent from the device.