SLI tests at anand

Sell them ?

 

I agree and made that point earlier. It is a technology that needs to move down into at least the mainstream to be viable. The 6600 series has not validated that approach but perhaps the next generation midrange cards will. Until that time, SLI remains a trophy gimmick ala 3dfx in the PC market.

 

It is easier to sell one of something rather than two of something is it not? That was hardly my main issue with SLI. The 6600 series hype surronding SLI has proven to be misplaced. As a commercially viable product, SLI is relegated to less than a blip on the radar screen.

 

Perhaps Ati hasn't marketed that part of their cards that well.

I've been to their site just the other day and saw that they had a lot of cards for digital TV. I'm soon going to build a HTPC and i'm really interested in either DVB-T or DVB-C capable cards. I've only seen cards from Hauppage, Twinhan or other a bit more obscure manufacturers that supports these types and i'm a bit curious if Ati or Nvidia aren't planning to start make these types of cards. Cause they seem to be a bit overpriced at this moment.

Ati has the NXT2003-6 chips but i don't know what types of digital tv they support, and are there any cards available with these chips on them ?

As a sidenote, the Xilleon processor should sell like hotcakes now when WMCE 2005 are out . I'll buy one, as soon as i've bought a house and are in need of 1-2 extenders

 

I would disagree based on the comments that i've seen from most reviewing sites. Though i'm guessing that it will be better for marketing purposes then for actual sales.

 

Hmm to me 2 6600 GT's are pretty pointless.
If my actual 6800 GT would be PCI-E that would be a nice deal though. Add another GT in a year for pure ownage!
In my opinion SLI is a good option as soon the nforce 4 SLI boards drop below 150 bucks.
Great how well SLI works though.

 

In those tests posted at Anandtech, the SLI 6600GT's hang very close to the 6800GT with 4xAA and at all resolutions up to 1280x1024. The results at 1600x1200 are expected because the 6600GT tested has 128MB RAM.

It is silly to suggest that this renders SLI useless on the 6600GT. The primary benefit of SLI will be that consumers can add an additional card for a huge performance boost, typically at a marginal cost.

Keep in mind that any costs associated with getting a PC equipped for SLI are fixed costs, and will be necessary at one point or another if one is interested in trying SLI in the future.

The people who buy a 6600GT will not be seriously considering a 6800GT as it is way out of their original price range. However, these same people will in many cases be able to save up extra money to pay the marginal cost associated with adding a second 6600GT somewhere down the road.

It would also be pointless for NV to design two $200 cards in SLI that would significantly outperform their own single $400 solution. Assuming that most people don't want to spend more than $400 on video cards, and assuming that users who want to be SLI equipped will spend the fixed costs on such a PC setup, what would happen to the market of their $400 cards if two $200 cards in SLI would significantly outperform them?

Let's think of this another way: if one spends $200 on one 6600GT at the beginning, and then spends $100-150 on another 6600GT a bit later, they will have a solution that very closely approaches the performance of a 6800GT with 4xAA/16xAF up to 1280x1024, for a fraction of the original purchase price. They will also have a higher performing solution than those who purchased a $300 6800NU. They will probably also be in a better situation than someone who spends $200 on a 6600GT, sells it later for $100-150, and then buys a new card for $250-300. SLI may also mean that the cards in general will hold a higher resale value than non-SLI cards.

SLI is a means to help futureproof the cards to a certain degree, and a means to significantly increase the performance at any time for a marginal cost. This is always a good thing.

 

It's not cost effective to spend the money on a dual PCi nV4 MB and a 6600GT when you will save very little on that over a decient nF3/4 single slot(AGP or PCi-e) MB with a 6800GT right now - and for the forseeable future. Only an uniformed person or one of diminished capacity would even consider such!

On the other hand, the 6800GT /U SLi would be a pretty outragious system for those with the disposable income and /or no family to get...... It might take a bit of a family history of insanity to do.... Hmmm.....think they used to call my great great great great great uncle Vlad a bit crazy......

 

Well, it may be worthwhile for one to consider a 6600GT if they have an understanding of fixed vs variable and marginal costs.

Getting a PC equipped to handle SLI is a fixed cost. This is a cost that is necessary somewhere down the line if one expects to use SLI at any time in the future.

 

In case people haven't noticed, did anyone compare [H]OCP's data to Anandtech's data on the SLI setups?

Doom 3 at 1600x1200, 4xAA, was the only game that they both showed SLI results for:

[H]OCP:

6600GT SLI: 32fps
6800GT: 37.9fps

Anand:

6600GT SLI: 32fps
6800GT: 38fps

So clearly these numbers seem to be in line with each other.

Also notice that 6600GT in SLI may at times be slightly faster than a single 6800GT. In the [H]OCP piece, 6600GT SLI was faster than a single 6800GT in both Halo (1600x1200) and 3dmark05 (in fact, now that I take another look, the 6600GT SLI was faster than a single 6800U in both of these tests too!).

http://www.hardocp.com/image.html?image=MTA5NzYwMzIxNWNGVnNoMkFBMEZfMV8xX2wuZ2lm

 

Wooohoo!!! No wife and/or kids and word is that year end bonuses are sweet. Put me down for 2xGT Seriously though - they need to put some mainstream SLI capable boards on the market if they want to make any sort of impression.

 

I think people get the idea of fixed cost but in the long term it's irrelevant if you have the money now. Check these scenarios over a 12 month period.

$100 (board) + $400 (6800GT) == $500 (no upgrade)
$150 (SLI board) + $200 (6600GT) + $150 (6600GT) == $500 (SLI upgrade)
$100 (board) + $200 (6600GT) - $100 (6600GT) + $200 = $400 (sell 6600 / buy next gen)

Although it depends somewhat on the quality of next gen's $200 parts which scenario do you think provides the best bang for the buck over that one year period? The 6600 upgrade sacrifices performance for the initial $150 savings and many would say that it's best to put out all $500 now. Of course for those who really don't have that extra $150 some smart planning could work out well for them.

 

The 6600GT will naturally cater to those people who do not wish to spend the extra money up front for 6800GT-type performance, yet still want to have the option to obtain 6800GT-type performance in the future for a marginal cost.

As for an analysis of your scenarios, typically we do not see a 2x increase in performance in 1 year for a $200 part. Remember that 6600GT SLI is giving 6800GT-type performance, which would be difficult for a $200 card to beat in one year's time. Essentially, SLI capability is in a sense buffering the user from uncertainty in the future. It is impossible to know exactly how much increase in performance and features will be evident in 1-2 years time, yet it is much more certain how much increase a user can expect from SLI. And don't forget that SLI products should hold their value fairly well, and eventually the enthusiast users will sell the SLI ready card for an even more capable SLI ready card in the future.

A user who purchases an SLI ready motherboard and card will have plenty of good options to choose from in 1-2 years time. If performance and features of the refreshed cards is not high enough to eclipse an SLI setup, than they simply need to purchase a second card (at a fraction of the original purchase price) and run it in SLI config and then upgrade to a new SLI ready card sometime even later in the future. If performance and features of the refreshed cards is high enough to eclipse an SLI setup, then they simply need to sell their SLI ready card for a more capable SLI ready card. So they really are, in a sense, buffered from uncertainty in the future regarding graphics performance and feature content.

The more one thinks about it, the more one realizes why ATI is working on a multi-gpu solution. On top of the fact that it will benefit users, they really don't have a choice if they want to compete on a level playing field against NVIDIA

On a side note, there should be a 256MB variant of the 6600GT, so I imagine that performance in this SLI config would be even closer to the 6800GT at 1600x1200 vs the variant we saw in Anand's article.

 

I agree on the 6600GT not being worth it, how hard is it to save up another 200 bucks? That's less than a weeks pay, and since the 6600GT has only 4 ROPS, it will never be as fast as a single 6800GT at high res and fsaa, not to meantion 128MB standard.
SLI is for those who like to spend 1000$(200 for SLI mobo, and 400 a piece or a 6800 GT) just to have the best, and maybe play the few games at the highest setting that can't be played , like doom3 with 4AA at 1600x1200, or 2048x1536 etc.
SLI is not economical, only for the rich, or people who like to spend alot to have the best GPU(s) for 1 product cycle, then throw them away for the next product cycle.

 

Link

 

Interesting comment made by [H]OCP:

If this is true, than SLI becomes even more attractive than it already is...

 

Personally I don't see how it could happen? Unless the connection between the two boards is fast enough to push texture data back and forth at extreme speeds etc.
Or am I missing something here?

 

Considering that the driver dynamically changes the load of each card, even assuming the (default) split screen mode there can be opportunities for the load to be 70/30 and 5 frames later 30/70. Both cards must have loaded the textures they're using AND the textures they might have to use in case the load changes. I can only see some savings on framebuffer mem (even more with Triple buffering enabled) but even then the savings will be nothing to write home about and that's assuming nVidia's SLI uses dynamic sizes for framebuffers.

 

I dont either, It's possible a bit of memory frees up (As illustrated earlier) Due to the way each cards handles rendering. But it seems unlikely that Nvidias method can double memory..

 

One of the false assumptions is that an SLI MB is wasted. An SLI MB might last me for 2-3 years, during which, 2-3 generations of vid cards will come and go, and all during that time, I have additional flexibility in my upgrade path. We're talking $2-$4 amortized over 24-36 months in extra cost for an SLI MB, for the additional flexibility to handle SLI if you want it. Or, to have multiple vid cards for 3+ monitors. Or, to have a separate vid card to deal with dedicated desktop. Think of it as a KVM switch on your own computer.

On the video processor, Nvidia's claim was that compression/encoding didn't map to pixel shaders, and I agree. Motion estimation can't be done by pixel shaders, and can't be done by a stream/block type processor like the iDCT components on current GPUs. It requires somewhat random access and search techniques.

I personally think decoding/encoding are commodities and should be included in every GPU. Decoding, because compressed video is everywhere. And encoding, because so many people have digital camcorders now that it provides a *significant* speedup to working with video. OEMed beginning systems for home use should frankly have it integrated and come by default rather than forcing people to buy a separate encoder card, or deluxe tuner card.

Part of the purpose of the VP is to commodify the video features so that every card has them, and take them out of the realm of separate "Theater" or "Compression" cards.