New anand article on SLI performance

[Rage187]

Rage187's claim that the V2 SLI "whooped any V3" is just flat-out wrong, given the benchmark results above. When not CPU-limited, the Voodoo3 and TNT2 were both faster than the Voodoo2 SLI.

Ummm I ran AMD and played CTF Q2 (I'm thinking it was faster in Q3 as well). Hell you even said it was faster in a previous post.

 

[Pocatello]

What kind of power supply do we need to run two GF6800 Ultra?

 

[Matthias99]

Originally posted by: Rage187

Rage187's claim that the V2 SLI "whooped any V3" is just flat-out wrong, given the benchmark results above. When not CPU-limited, the Voodoo3 and TNT2 were both faster than the Voodoo2 SLI.

Ummm I ran AMD and played CTF Q2 (I'm thinking it was faster in Q3 as well). Hell you even said it was faster in a previous post.

When did I say the V2 SLI was faster than the V3? 😕

And if I did, I was wrong. Look at the benchmarks above. In non-CPU-limited situations, the Voodoo3 benched equal to or faster than the V2 SLI.

 

[Rage187]

on the slower AMDs, the V2 SLI is better, though, at least in Quake2

and I'm pretty sure, not positivley, that it was faster in Q3.

 

[Genx87]

Here is my take on the upgrade argument.

Lets say next Fall I build a new computer with an A64, 2GB ram, blah blah.
I purchase a single NV50GT(7800GT??) for 400 bucks.

Total system cost is similar to this years game machine I bought. ~1300 bucks.

12 months rolls around and games come out that stress the system to the point where I cant play at my usual 1600X1200. Instead of purchasing a whole new machine for 1300 bucks with an NV60 variant. I go and purchase another 7800GT for 200 bucks. I get 40-100% increase in performance and prolonged the game machine for another 12 months.

Total system cost over 2 years is 1500 bucks or 750 a year. If I go my usual upgrade path and build a new machine for another 1300 bucks it will have cost me 2600 bucks or 1300 a year.

With the extra 550 bucks(1100 total) each year I can do other things like take a vegas vacation for 3 days 🙂

 

[ronnn]

I will wait until Nvidia produces their super drivers that solve all problems before investing any money based on todays results. Besides by the time the bugs are worked out, prices will be cheaper. If you want the best 3dmark05 results, sli is the way to go. If you want the best gaming platform today, sli is not a good choice. The argument of buying the same card in a year and upgrading than is risky. Very few cards seem to last a year without changes being made, though I guess you can always buy used.

 

[nitromullet]

Originally posted by: Genx87
Here is my take on the upgrade argument.

Lets say next Fall I build a new computer with an A64, 2GB ram, blah blah.
I purchase a single NV50GT(7800GT??) for 400 bucks.

Total system cost is similar to this years game machine I bought. ~1300 bucks.

12 months rolls around and games come out that stress the system to the point where I cant play at my usual 1600X1200. Instead of purchasing a whole new machine for 1300 bucks with an NV60 variant. I go and purchase another 7800GT for 200 bucks. I get 40-100% increase in performance and prolonged the game machine for another 12 months.

Total system cost over 2 years is 1500 bucks or 750 a year. If I go my usual upgrade path and build a new machine for another 1300 bucks it will have cost me 2600 bucks or 1300 a year.

With the extra 550 bucks(1100 total) each year I can do other things like take a vegas vacation for 3 days 🙂

Yep. this makes sense, since (theoretcially) there won't be a single card available when you purchase the "7800GT" that would yeild higher FPS than two "7800GT's". IMO, this is the key question of whether or not one should consider SLI as an upgrade path or not. If there is a card currently available that beats your proposed SLI setup, all you are doing is chasing the past, you will always have a slower video card than what is the current high end. You could do better by going with a less expesive cpu/mobo combo and a more expensive video card.

 

[kylebisme]

Originally posted by: Killrose
It will be interesting to see the final performance numbers on SLI once they get the grapic glitches cleared up. I guess it will either be way slower or way faster depending on what needs to be done to fix it.

Getting rid of the graphic glitchs won't nessacarly alter peformace much, but I don't see any reason to get excited about SLI until the glitchs are taken care of.

 

[apoppin]

Originally posted by: nitromullet
What Mathias99 is saying is true, it simply doesn't add up as an upgrade option.

For instance, say the 9800Pro had been SLI configurable... When the 9800Pro first came out is was around $400, and by the time NV40 and R420 were introduced they were around $200. So, if you had gotten a 9800Pro early on and then added another one when the next generation you would have spent a total of $600 for a setup that could be manhandled by 6800GT, which goes for $400.

Another case, say you get a single 6600GT now for $200 and then say the price drops on them to $100 when the next gen cards come out. That would have you spending $300 for a setup that performs about as well as a then last gen 6800GT, which probably will run for $300 or less by the time the next gen comes out. Not to mention that NV50 and R500 cards will be available at that time, and there will most likely be a $300 option that performs better than either dual 6600GT's or a single 6800GT.

IMO, SLI is the way to go if you get dual 6800 series cards, which would yield very high performance. SLI (or SLI plans) only becomes interesting when a dual solution offers more performance that any single card on the market at the time of purchase of the first card, otherwise I would always go with the faster single card.

look at it another way . . . . i did buy a 9800xt before Summer for $223 (the 6800GT was UNavailable) . . . . let's say i have a SLI (type) ATI MB available . . . .

i could sell my 9800XT for ~$180 and buy a x800xt for ~$350 MORE

. . . . OR . . . .

i can get another 9800XT for ~$200 to get very similar performance to one x800xt
(and i saved a few bucks). 😛
:roll:

edit: there are MANY ways to look at an upgrade path (now) there weren't before SLI . . . . MORE options are ALWAYS better. 😉

 

[EightySix Four]

Plus their is always the mislead average consumer who thinks 2 of them equals 2x the performance in everything...

 

[apoppin]

Originally posted by: crazySOB297
Plus their is always the mislead average consumer who thinks 2 of them equals 2x the performance in everything...

that's marketing's job 😛
:roll:

. . . and the perf increase IS pretty impressive to those of us above-average.
😀

 

[nitromullet]

look at it another way . . . . i did buy a 9800xt before Summer for $223 (the 6800GT was UNavailable) . . . . let's say i have a SLI (type) ATI MB available . . . .

i could sell my 9800XT for ~$180 and buy a x800xt for ~$350 MORE

. . . . OR . . . .

i can get another 9800XT for ~$200 to get very similar performance to one x800xt
(and i saved a few bucks).


edit: there are MANY ways to look at an upgrade path (now) there weren't before SLI . . . . MORE options are ALWAYS better.

This makes sense also, at the time you bought the 9800XT there was no single card that could out perform two of them in a theoretical ATi SLI setup. What doesn't make sense is buying 1/2 a 6800GT now that requires a new motherboard and then buying the other 1/2 later. You could just buy the whole 6800GT now with the money you save on either keeping your current mobo or opting for a less expensive option (like maybe a socket754/A64 combo).

 

[Pete]

Originally posted by: Genx87
Here is my take on the upgrade argument.

Lets say next Fall I build a new computer with an A64, 2GB ram, blah blah.
I purchase a single NV50GT(7800GT??) for 400 bucks.

Total system cost is similar to this years game machine I bought. ~1300 bucks.

12 months rolls around and games come out that stress the system to the point where I cant play at my usual 1600X1200. Instead of purchasing a whole new machine for 1300 bucks with an NV60 variant. I go and purchase another 7800GT for 200 bucks. I get 40-100% increase in performance and prolonged the game machine for another 12 months.

Um, or you just sell your NV50 to finance the NV60. I didn't realize you had to purchase a new system to buy a new video card (tho, ironically, you do in the case of SLI--sorry, couldn't resist 😉).

SLI is nice as an option for new systems or total upgrades, but I still think it's most attractive at the high end as an immediate way to offer performance beyond volume manufacturing and profitabilty contraints. IoW, NV60 performance at NV50's release is pretty cool. The promise of NV50 performance in the future at NV50's release? Not so exciting.

Again, I don't see what's so impressive about two 6600GTs that perform about the same as a 6800GT--but with half the RAM. Something's also slightly amiss if two 500MHz 8 pipe cards can't beat a single 350MHz 16 pipe one--they have a 40% theoretical advantage in pixel shader power (assuming 100% SLI efficiency, which I grant seems unrealistic). It may be the missing ROPs to blame, as the SLIed 6600s will have 8 to the 6800GT's 16. I'd love to see more shader-heavy tests, like 3DM03 and 05 at ultra high res, to stress SLI's shader power.

 

[apoppin]

Originally posted by: nitromullet

look at it another way . . . . i did buy a 9800xt before Summer for $223 (the 6800GT was UNavailable) . . . . let's say i have a SLI (type) ATI MB available . . . .

i could sell my 9800XT for ~$180 and buy a x800xt for ~$350 MORE

. . . . OR . . . .

i can get another 9800XT for ~$200 to get very similar performance to one x800xt
(and i saved a few bucks).


edit: there are MANY ways to look at an upgrade path (now) there weren't before SLI . . . . MORE options are ALWAYS better.

This makes sense also, at the time you bought the 9800XT there was no single card that could out perform two of them in a theoretical ATi SLI setup. What doesn't make sense is buying 1/2 a 6800GT now that requires a new motherboard and then buying the other 1/2 later. You could just buy the whole 6800GT now with the money you save on either keeping your current mobo or opting for a less expensive option (like maybe a socket754/A64 combo).

the only reason i can imagine for buying half of a 6600GT setup . . . . is simply not having the extra $200.

a single 6600GT will satisfy most (non lcd) gamers for most current (and probably next year's) games. When the 6600GT fails to satisfy, an additional $80 NEXT year will put off a MAJOR upgrade for still ANOTHER year. 😉 That's a TOTAL of $280 - spread out in two purchases - compared to a single 6800GT for ~$380.

Kind of an extreme(ly poor) budget gamer's upgrade path.

SLI is good for the two extremes - the HW fiend who must have the latest and greatest - two 6800Ultra Extreme OC's in SLI . . . . and the budget gamer.

i think nVidia has a winner!
:thumbsup:

 

[Drayvn]

Why was it that ppl who bought the 6800 series and for that matter nVidia was blasting ATi for not being future proof as they didnt have the SM3 feature.

In a way SLi is future proofing, but in no way can it compete with next gen cards with their new features.

As the guys saying theoretically. Lets say ATi had an SLI out now and u bought an X800XT now, then next year bought another X800XT to increase performance.

But next year games are coming out with SM3 and the next gen fully support SM3. What would u do, in terms of FPS ur up there with the next gen cards. But as the next gen cards mature, the further ur current SLI setup will slip further behind.

And this goes for all SLI setups which will go up against the next gen cards, which are typically every year. So neither is it an upgrade path nor a one time buy.

Also, do u think nVidia will be evil enuf to make underperforming single cards but when put in SLI work just as good as the competitions single card... OOOO that would be evil!

Or in the future until ATi being out their own SLI type solution. Do u tink benchmarks will always have SLI instead of the single card, or will it be fair game and have the SLI as an extra part of the benchmark review?

 

[Genx87]

Um, or you just sell your NV50 to finance the NV60. I didn't realize you had to purchase a new system to buy a new video card (tho, ironically, you do in the case of SLI--sorry, couldn't resist ).

I build entire systems. I dont upgrade piecemeal. If I went out and got an NV60 then chances are Ill build an entire new machine. It wouldnt make sense to me this round to have sold my 5900 for 200 bucks and got a 6800GT for 400 bucks but keep the 15 month old computer. It was an XP@2200Mhz with a GB of PC2100 memory and a seagate 7200 RPM drive. I would have a major bottleneck in the CPU,RAM, and disc. Basically the computer itself was a generation behind. And the benchmarks between an XP and A64 is like night and day. Roughly 30-50% faster on the 6800s according to Toms.

 

[BenSkywalker]

I agree 100%. I'm just saying that claims that the V2 "dominated the market for a very long time" (your own words) are a bit misleading -- there were single cards running about 75% as fast as it within 6 months

Not quite. A TNT1 pushing 40FPS 1024x768 while a Voodoo2 SLI setup was pushing 75(both running Q2 demo1- 87.5% faster for the V2) with Unreal numbers showing the TNT pushing 21FPS running 10x7 with the V2 SLI pushing 36FPS(71% faster) and Incoming showing the TNT pumping 48.6FPS with the V2 SLI pushing 72.1FPS(48% faster). Expendable running 1024x768 the TNT does manage to get quite close(it actually edges out the V2 SLI running 640x480) managing 40.8FPS vs the V2 SLI's 42.8FPS(a mere 5% edge for the V2 SLI there).

If you bought a single V2 at release, it would not have made much sense to upgrade with a second one 12 months later (by which time the Voodoo3 and TNT2 were available).

$100 for another V2 vs $250 for a TNT2Ultra or V3 3500 wasn't always clear cut and many of the members here back in those days debated which way to go. Actually, back then the big plus going with the newer cards was improved 2D and 3D IQ(getting rid of the pass through cables was a major issue for a lot of people).

Rage187's claim that the V2 SLI "whooped any V3" is just flat-out wrong, given the benchmark results above. When not CPU-limited, the Voodoo3 and TNT2 were both faster than the Voodoo2 SLI.

Absolutely, although when looking at retail availability it was closing in on fifteen months between them.

Now -- will this be what happens this time around? That's still up in the air. If the next-gen parts (NV50/R500) don't show big performance increases over NV40/R400, SLI might actually make sense as an upgrade path.

It made sense before for a lot of people too.

I'm just saying that, in the past, that was not the case -- and with the current supply issues with the NV40 and R400 cards, prices seem unlikely to drop dramatically over the next 12 months (so the new cards wouldn't have to show all that much performance gain to win out in terms of price/performance).

ATi is going to be trying to launch their R500 core shortly(which is actually their old R400 core) which could have nV lowering their prices a decent amount on their higher end parts if the R500 comes through.

 

[Acanthus]

As it looks now, PCI-E 6800s are going to be a problem to find as well.

I am planning on moving to AXP3000, 6800GT, Asus SLI and upgrade to a 2nd GT in 6 months or so.

 

[drpootums]

ahh...this is good news for me and anyone else putting a single card solution on a socket 754 with AGP around christmas. I'm going with a Asus or MSI socket 754 board, a AMD Athlon 3400+ 64, and a PNY or eVGA 6800 GT AGP, and being this is out for socket 939 and people are gunna be making the bump to PCI-e, everything ill be getting will just come down in price. I know this performance is good, but i think single card AGP will still be good for quite a while yet.

SLI does look cool though, to bad i dont have extra money...

 

[touchmyichi]

Originally posted by: apoppin

Originally posted by: nitromullet
What Mathias99 is saying is true, it simply doesn't add up as an upgrade option.

For instance, say the 9800Pro had been SLI configurable... When the 9800Pro first came out is was around $400, and by the time NV40 and R420 were introduced they were around $200. So, if you had gotten a 9800Pro early on and then added another one when the next generation you would have spent a total of $600 for a setup that could be manhandled by 6800GT, which goes for $400.

Another case, say you get a single 6600GT now for $200 and then say the price drops on them to $100 when the next gen cards come out. That would have you spending $300 for a setup that performs about as well as a then last gen 6800GT, which probably will run for $300 or less by the time the next gen comes out. Not to mention that NV50 and R500 cards will be available at that time, and there will most likely be a $300 option that performs better than either dual 6600GT's or a single 6800GT.

IMO, SLI is the way to go if you get dual 6800 series cards, which would yield very high performance. SLI (or SLI plans) only becomes interesting when a dual solution offers more performance that any single card on the market at the time of purchase of the first card, otherwise I would always go with the faster single card.

look at it another way . . . . i did buy a 9800xt before Summer for $223 (the 6800GT was UNavailable) . . . . let's say i have a SLI (type) ATI MB available . . . .

i could sell my 9800XT for ~$180 and buy a x800xt for ~$350 MORE

. . . . OR . . . .

i can get another 9800XT for ~$200 to get very similar performance to one x800xt
(and i saved a few bucks). 😛
:roll:

edit: there are MANY ways to look at an upgrade path (now) there weren't before SLI . . . . MORE options are ALWAYS better. 😉

I completely agree. Sure there will be times when the single card option will prevail, but then again you never really know. There might be a large mark off on a 6600 GT to complement the one you already have. Maybe 2 100 dollar 6600 GT's will beat any 200 dollar mainstream solution in the future (imagine if you could chain up dual 9500 pros right now...think of how that is a 2 year old card also). In the end, SLI adds another upgrade path even though it is one that does not necessarily have to be taken.

 

[Xernex]

SLI is looking promising, but I think we need to wait and see more reviews on this until we can make final conclusions, there are still unanswered questions like PSU requirements, how well overclocking works and whether or not nvidia can tune its drivers to fix some of these graphical glitches and maybe even squeeze a tad more performance out.

 

[BFG10K]

Interesting but I still easly prefer upgrading single cards more often.

 

[nRollo]

Originally posted by: BFG10K
Interesting but I still easly prefer upgrading single cards more often.

No way! For a guy like me who keeps his cards a long time- this is the way!
😉

Anyway, I'd love to bridge the cards again just like the good ol' days, we'll see if I can afford it.

 

[SonicIce]

::hugs his NV50 that he brought back from the future:: SLI, what a waste of money

 

[RussianSensation]

Originally posted by: gururu

the 6600GT, even in SLI mode, is still generally outperformed by a single 6800GT. At lower resolutions or with AA disabled, the performance of two 6600GTs would definitely be more similar to that of a single 6800GT

i wouldn't want to wait two years to get 6800GT performance. I'd rather just pay the price and get a GT today. two GTs is really the pudding. I'm a little diappointed because the 6600GT option was attractive.

Good point!! especially since 6800Gt is already at 350-360 right now and 6600GT is slightly above $200. Add to that the cost of SLI motherboard, and you can see how buying 6600GT is pointless for SLI setup.

Also Anand mentions that the upgrade path is bright since you could pick up a 2nd SLi card cheaper. I disagree with the fact that there is a benefit to that for 6600GT at least. Cards that retail at $200 take 6 months just to drop to $150 and then another 6 months to drop to $110 => just look at 9600Pro or 9600XT which barely dropped from their mainstream introduction levels over the last 1.5 years. Who is going to pay $200 for 6600GT now and then say $100 in 1 year, when you can get FASTER 6800GT for $350 NOW....and have the chance of overclocking it to 425/1100......? Did anyone think about overclocking in SLI <<< will it be stable? Are you going to need 600Watt power supply>>>>?

SLI upgrading can make sense in 6800GT case because the speed almost doubles in intensive benches. Still those people that consider 6800Gts now probably want to have close to the cutting-edge hardware, latest features, etc. These new features, which enable better graphics (ie. DX8 vs. DX9 => DX10, PS4.0 etc.) are important to them (No?) So I still question the usefulness of this solution since hardware enthusiasts could grab NV50 instead of a 2nd 6800GT based on feature set even if the 2 6800Gts performed slightly faster. But, 2 6800GTS improve 50-80%, and from last introduction 6800GT doubled 5900/9800P cards. So NV50/R500 are most likely to be faster than 2 SLI setups.

In my eyes, the only people this will appeal to is people with higher incomes who could pick up 2 6800GTs or 2 6800Ultra now, and then 2 NV50s/2 R500s later. Isn't SLI motherboard rumoured to retail at $199? I can only imagine the price of 6800s in PCIe considering there is a huge shortage of high-end PCIe cards, making SLi setup even more expensive.