[Sweclockers] Titan X in SLI

[Mondozei]

Most of the reviews of Titan X in SLI, such as this one, had a weird and irrelevant selection of games.

Sweclockers have fixed that.

In addition, Sweclockers have also published their invaluble performance index. Here's their latest one:

One curiosity is that they maintain that AMD's crossfire is less buggy than SLI these days, has a better frame-pacing. This should be well-known by now, but it isn't really.

Also interesting to see how the 290 is continually outperforming the GTX Titan (1st version). Makes you wonder how the 390X will do against the Titan X down the road.

 

[Silverforce11]

This resolution is more relevant to compare such cards, as most would agree:

And 4K, which is also indicative of lower resolution with DSR/VSR:

 

[alcoholbob]

Titan X is just on the cusp. If it could hit 1600MHz with an AIO and better TDP it would be insane beast at 4K.

 

[Silverforce11]

Titan X is just on the cusp. If it could hit 1600MHz with an AIO and better TDP it would be insane beast at 4K.

If NV allow custom boards from AIBs, we might get that. NV's architecture is a beast when it comes to OC, it just needs the boards & cooling to support it.

It's very exciting to think what 14nm ff will bring if we're getting such good performance on 28nm.

 

[SolMiester]

This resolution is more relevant to compare such cards, as most would agree:

And 4K, which is also indicative of lower resolution with DSR/VSR:

Looks to me that Titan SLI has hit the CPU wall with 4k and 1440p the same!

 

[tential]

Looks to me that Titan SLI has hit the CPU wall with 4k and 1440p the same!

Huh? It's a 243% increase over the base card the R9 290x. It's not "FPS"...

 

[96Firebird]

If anything, it hit the CPU wall at 1080p...

Looks like a beast.

 

[iiiankiii]

The numbers don't add up. A single Titan X is 48% faster than a 290x but, in SLI, it's 143% faster than the 290x? WTH?

 

[Hitman928]

The numbers don't add up. A single Titan X is 48% faster than a 290x but, in SLI, it's 143% faster than the 290x? WTH?

It makes sense, it's just how things work when you have a baseline comparison rather than the raw data to look at. It may seem funny, but it's probably correct (I say probably simply because I haven't seen the raw data).

e.g.

290x = 60 fps
TitanX = 1.43*60 fps = 85.8
TitanX SLI (assuming 80% scaling) = 1.8*85.8 = 154.44

TitanX SLI improvement over single 290x = 154.44/60 = 2.574 or 157.4% faster.

 

[bystander36]

The numbers don't add up. A single Titan X is 48% faster than a 290x but, in SLI, its 143% faster than the 290x? WTH?

What is odd about that?

Perfect scaling would expect 48% faster from a single card to be 196% faster in SLI, but we all know that perfect scaling doesn't happen often.

Remember, that 48% faster means, it is at 148% of the 290x. Multiple by 2, if things were perfect, and you are at 296% of the 290x performance, or in other words, 196% faster.

148% to 243%, means it scaled 65%. While not ideal, isn't that bad.

 

[96Firebird]

For simplification, just think of the numbers as FPS.

 

[xorbe]

It's amazing that it takes 980 SLI or 290X CF to start really pulling away from a single Titan X. An overclocked Titan X is a beast.

 

[lehtv]

Huh? It's a 243% increase over the base card the R9 290x. It's not "FPS"...

It's a 143% increase, since the base card is 100%. 100%+143% = 243%

 

[gamervivek]

nvidia should be able to XDMA with the maxwell series since they have two DMA engines too.

 

[3DVagabond]

nvidia should be able to XDMA with the maxwell series since they have two DMA engines too.

What, and give up the market for these really nifty SLI bridges?

 

[flash-gordon]

Frametimes? Someone?

 

[Cloudfire777]

Ugh, Titan X SLI is actually required for 60FPS in 4K resolution.
Expensive stuff

 

[Leadbox]

Frametimes? Someone?

NOT a single review site has put Titan X Sli under the FCAT microscope
I wonder why?:sneaky:

 

[Headfoot]

NOT a single review site has put Titan X Sli under the FCAT microscope
I wonder why?:sneaky:

Seriously. Regardless of which company first proposed it, it's highly valuable data. I don't know why we're seeing a backslide into raw FPS...

 

[RussianSensation]

1. Wow, 960 falls apart at 1440P which means for future games this card is going to get creamed as it can hardly cope with increased graphical demands.

280X leads 960 by 30%, while a reference R9 290 is ahead by 61%. Considering 960 and R9 290 are just $50 apart and R9 280X hovers at the $200 range, it's shocking that some people are actually buying a GTX960 for gaming. That card is just as much of a turd as 8600GT/GTS were.

2. $2000 Titan SLI is just 14.4% faster than R9 280X CF. :sneaky: I guess one can say the Titan X is impressive in that it more or less matches Titan SLI but on the other hand Titan SLI performance is very underwhelming for their MSRP. More remarkable is that 7970Ghz CF were $0 due to mining. :biggrin: Good thing Titans overclock well, because at stock speeds the performance in modern games is a sad state of affairs considering their $1K MSRP.

3. When CF scales, R9 290X offers awesome bang for the buck. $600 setup today that's beating 780Ti SLI at 4K by 16% and isn't too far behind $1060-1100 GTX980 SLI! Looks like when gamers say that NV's cards start off a generation very strongly but towards the end of the generation AMD's cards last better/NV's advantage all but disappears is evident today with 7970Ghz > 680 and R9 290X > 780Ti and their SLI/CF combinations. That's something to think about when R9 390X launches.

Wish FCAT/frame times testing was done to evaluate the CF vs. SLI and while at it investigate 970 SLI VRAM stuttering (if it's there).

4. Upgrading to Titan Xs in SLI for 1080P seems almost pointless. We really need more GPU demanding next gen PC games. :thumbsup:

If NV allow custom boards from AIBs, we might get that. NV's architecture is a beast when it comes to OC, it just needs the boards & cooling to support it.

It's very exciting to think what 14nm ff will bring if we're getting such good performance on 28nm.

I am always amazed at how well NV's large chips overclock. I would imagine 6GB of VRAM and another revision would free up 30-35W TDP headroom which could result in a 1217mhz GM200 6GB chip. Take that and add MSI Lightning treatment and I can see 1.5-1.55Ghz OC on air! I think that's why NV is holding back GM200 6GB. Once they see where R9 390X lands, they can release a faster clocked better binned version that should be able to take the performance crown easily.

I am with you about 14nm/16nm. HBM 2 + lower node and a new architecture from NV -- it could easily end up > 50% faster than the Titan X! On one hand 28nm has held back Maxwell/R9 300 series, but on the other hand that means 14nm GPUs should be that much better in theory. ()

 

[omeds]

I don't think 960 was aimed at anything beyond 1080p tbh. It's a mainstream card for the most mainstream of resolutions. Kudos to AMD for rebranding Tahiti. It worked out well for them, albeit not ideal for an efficient next gen part.

And you're right, AMD are certainly the value option. Many people value more than simply price/perf though, like feature set, support, efficiency, cooling, noise.

Just out of interest, have you ever considered applying for a position at AMD?

Not helping.
-Elfear

 

[SimianR]

Impressive to see the R9 290 (non-x) crossfire pulling ahead of the 970 at 4K. It is crazy to see how far the 780 Ti's performance has fallen though.. had people known I wonder if some discussions would have gone a bit differently when the 290/290X were released

 

[omeds]

Large Kepler overclocks quite well in comparison though, but sure Hawaii has certainly improved more since launch, much like Tahiti did. Hawaii really shows its strength in higher resolutions too, largely thanks to ROP count and pixel fill rate. Keplers performance degrades much quicker due to a large deficit in this area, despite generally having faster processing speed (as is evident when fill rates are not impacting performance or creating a bottleneck). That is to say Kepler is better suited for lower resolutions, and Hawaii is better suited for higher resolutions.

Taking things like feature set and the displays I use into account, I'd still prefer Kepler atm. An upgrade of displays would change that though.

 

[Yuriman]

I don't think 960 was aimed at anything beyond 1080p tbh. It's a mainstream card for the most mainstream of resolutions. Kudos to AMD for rebranding Tahiti. It worked out well for them, albeit not ideal for an efficient next gen part.

And you're right, AMD are certainly the value option. Many people value more than simply price/perf though, like feature set, support, efficiency, cooling, noise.

Just out of interest, have you ever considered applying for a position at AMD?

Not helping.
-Elfear

I can think of plenty of times in the past where AMD had significant features NV didn't, for a while. Even now NV doesn't have everything AMD does, so that's not a one-sided argument.

And, as for efficiency, GTX 480?

That said, I just put a 960 in my sister's computer because of its performance per watt.

 

[RussianSensation]

I don't think 960 was aimed at anything beyond 1080p tbh. It's a mainstream card for the most mainstream of resolutions. Kudos to AMD for rebranding Tahiti. It worked out well for them, albeit not ideal for an efficient next gen part.

Right but once future games come out, there will be increased GPU load. That's the reason a lot of us look at higher resolutions and how cards fare vs. one another even if we ourselves use a lower resolution. Since know that future games will increase GPU load 50-100%, as an example, seeing how a card fares at 1440P and 4K allows us to see what happens in a GPU-limited scenario and how the card can handle the extra load. In this case, the 960 falls apart compared to the 280X and even more so compared to the R9 290.

And you're right, AMD are certainly the value option. Many people value more than simply price/perf though, like feature set, support, efficiency, cooling, noise.

It's very difficult to defend the $200 GTX 960 2GB because the gamer who buys it might upgrade to a 60-70% faster Pascal x60 series card in 1.5-2 years for another $200. Why do that when you can have 60%+ more performance for the next 1.5-2 years with a $50 more expensive R9 290? The 960 is the worst buy in NV's line-up today. Casual gamers would be OK with a 750/750Ti. $200 960 2GB and $240 960 4GB sit in no man's land. Buying a 960 means going out and wasting yet another $200 in 2 years as I said, which actually ends up costing $150 more and for the next 1.5-2 years you end up with a card 50-60% slower than an after-market R9 290.

Your insinuations about noise and temperatures do not fly because after-market R9 290s run in the 70-80s range and are quiet. The PSU argument doesn't work either because one can easily find solid PSUs such as the Antec 750W Gold rated with 5 year warranty for only $50. This PSU will easily last 7-10 years, if not more.

Efficiency and perf/watt on the desktop on their own are just marketing words without context. With a GTX960 in demanding situations you literally get half or even less than half the performance of an R9 290. What difference does it make if your card is more efficient but it's unplayable in modern titles? Might as well use Wii U or Intel IGP to argue perf/watt.

Also, care to tell me what these amazing killer features that GTX960 has that warrants a 61% loss in gaming performance and half the VRAM against a $50 more expensive after-market R9 290 that is also cool and quiet? I am all ears. No need to discuss my career choices btw. If you have a rebuttal to the points I made earlier and in this post, please stick to GPUs please from a technical point of view. Even SLI can't save the 960 because an after-market R9 290 = 290X reference and that's roughly as fast as 960 SLI. It's interesting how you are quick to dismiss this 50-61% performance advantage but then why get a 960 over a 750Ti?

Also, many on this forum seem to be confused about preferences for value and gaming performance and price/performance that AMD brings with the gamer preferring AMD. If NV brought all of that to the table, I would be recommending their product(s), like I recommended 6600GT/6800GT/GTX460/GTX470/8800GT, get it? Is it my fault now that $2K Titan SLI is only 14% faster than R9 280X/7970Ghz CF or that a $400 R9 290 > Titan or that a $550 R9 290X > $699 780Ti?

If I bought dual 780s for $1300 or dual 780Tis for $1400 or $2K Titans and I saw today $800-1100 R9 290s/290Xs beating those cards, I would feel ripped off/upset because initially those cards had an advantage and hence their pricing premium was somewhat justified. Whether or not NV intentionally stopped optimizing drivers for Kepler and diverted more attention to Maxwell or if Maxwell is simply a superior architecture long-term for compute isn't even the point anymore. The reality is 1-1.5 years from 780/780Ti/Titan launch, the far cheaper AMD CF setups are outperforming Kepler, including the uber expensive $2k Titans; and this was already evident 6 months ago. This review today just supports what we saw months ago. Imo, those PC gamers who were able to score R9 290/Xs for free due to bitcoin mining are the real winners in the Kepler vs. R9 200 series generation. Too bad where I worked there were no R9 290 cards or I would have had 2-3 free R9 290Xs for 15-18 months already.