GeForce GTX 650 Ti Boost Review: A Budget-Oriented GK106-Based Boss

[blazorthon]

the 7770 was not direct competition for the 650ti... the 650ti was placed and priced right in the middle between the 7770 and 7850, in both price and performance. if the 650ti comes down to 7770 prices, then the 7770 becomes overpriced. The direct competitor for the 650ti from AMD was meant to be the 7790.

At a little lower price it offered better performance, and was meant to be the 650ti killer.

The 650ti boost is clearly aimed straight at the 7850, and by extensionthe 7790. priced about the same as the 7790, and performing about as good as a 7850, it's a formidable challenge to both cards. Which will require AMD to cut a number of their card prices to remain competitive.

Radeon 7770 and GTX 650 Ti were competitors until the Radeon 7790 showed up. The GTC 650 Ti Boost is a competitor for the Radeon 7850 and merely has a somewhat lower reported price. The Radeon 7790 is still not its competition because it is specifically competition for the GTX 650 Ti now.

That the 7790 was meant to be competition for the 650 Ti is totally irrelevant in the context that we were in because the 7790 is still not here and we didn't even know about it until recently. Before it, the only competition for the 650 Ti was the Radeon 7770.

The 7790's MSRP is not nor was below that of the 650 Ti.

Besides, about all of this middle of the road crap with the 650 Ti between the 7770 and the 7850/650 Ti Boost... Do keep in mind that in decent quality settings and such, the 7770 is literally right behind the 650 Ti. The 7790 makes a better case for being in the middle because it's actually in the middle instead of only a little bit above the 7770 and far below the 7850.

 

[Evolution2001]

I don't know that I'm completely onboard with these types of news articles. They are good and fun to read, but until the cards are actually in retail channels, I think it's unfair to AMD to repeat whatever nVidia has stated on a phone call.
If memory serves me correctly, nVidia has been known to spout off about new hardware and expected pricing, only to see the product launch date arrive and..."but it's only available in (insert country name) and there are only 3 units available. And the price has been marked up due to limited availability."

So when we get info 'leaked' like this, many of us on the verge of purchasing an AMD card suddenly decide to wait. Granted, waiting always guarantees a lower overall price regardless of which brand you choose, but it seems unfair to AMD.
This scenario can apply to either mfg, and I have no allegiance to either of them, but it seems like this scenario happens more with nVidia than AMD.

On the other hand...hey, it's guerrilla warfare marketing. And if nVidia is better at it, then so be it! If it means better prices for the consumer, that's good. But not-so-good in the long run (for consumers) if nVidia squeezes everyone out because of their aggressive ("questionable"?) marketing tactics. Then as the only major player, their prices are not likely to benefit the consumer.

Meh...whatever. As you were...

 

[ericjohn004]

The 7790 needs to drop to 130$. The 7850 1GB needs to drop to 145$ to compete. And I'm not talking about after rebates and games. This needs to be their starting price. AMD's starting price needs to be below what Nvidia is asking because perception is everything.

People perceive Nvidia to be quality and they perceive AMD to be second best. Therefore AMD needs to be cheaper and better performing to match the "quality" Nvidia is putting out.

And maybe Nvidia is higher quality, if you read the review on SLI and Crossfire 660Ti's vs 7870's the 660Ti has practically no tardy/dropped frames while the 7870 has quite a few. The 660Ti's graph is perfect while the 7870's graph is jumping up and down.

So maybe there is something to this perception that people have. Not all people have this perception. But clearly, from the sales numbers highly favoring Nvidia while the price/performance seems to favor AMD. So it doesn't make sense. And now that price/performance is favoring Nvidia, AMD better watch out.

I can't see anyone buying the 7790 over a 650Ti w/Boost 1GB.

 

[ericjohn004]

BTW the 650Ti with boost is on Newegg right now for 170$. So Nvidia was right, they are available now and at the price they said it would be.

 

[de5_Roy]

No, it doesn't make things worse for AMD. I'm not going to just repeat myself in explaining that because that was included in the earlier post.
Enduro and such didn't take any longer to get working properly than it took AMD to get a proper driver for the desktop cards (Catalyst 12.6 and 12.7). Furthermore, the start for Nvidia's competing technologies such as Optimus was much worse.

amd damaging the brand new lineup's user perception by neglecting to support cost amd and nvidia took advantage of it (and gained). nvidia's timing and execution had also been better than amd's.
nvidia's switchable gfx tech matured better than amd's, which led to oems demand more of nvidia's gpus while amd... well, they did whatever they did. i don't remember what amd even had at that time. but i do remember optimus' launch and p.r.
in gcn launch's case, the cards had gone without amd's attention long enough for people to withold purchase or fall back on older and/or less performing nvidia cards that provided better overall experience.. it was not for long, because kepler launched, but it certainly happened. a lot of those people later went straight for gtx 680, 670 when those launched. i think, for example, this may be one of the reasons why gtx670 and 680 rank to high on steam's video card rankings in hwsurvey. in amd's defense, may be more nvidia users use steam than amd users.

Pitcairn was and still is priced better than Nvidia's Kepler competition up until this GTX 650 Ti Boost (even that is still debatable), so any argument about Fermi over Pitcairn works worse for Fermi. Besides, THat sort of thing is perfectly normal. Many of the older cards are usually sold at a discount around the time of the next generation and their sales often improve in the short term as a result. The same happened to Nvidia too.

i agree about pitcairn's pricing. but nvidia's timing with deals was better.

My point about VCE not being a big deal is that both it and Nvidia's competitor, like Quick-Sync, were junk for quality from what I'd been reading. Some people have told me that this has been solved with Ivy Bridge and current drivers, but that's still debated online and I can't guarantee the truth of it.

quality has almost nothing to do with sales. if it had, no one woulda bought gtx550ti. my point was that amd put in the hardware, hyped it, then left it dark for months. even reviewers were vocal. that hurts image.

Your argument about Kepler with gamign versus compute is totally irrelevant. What I said is true and making excuses for it solves nothing.

Lawl, it's quite relevant to my point - how nvidia sells more cards (i.e. why people want to buy nvidia cards more than amd/how nvidia makes more money) than amd. you picked up only on "drivers". it's okay if you didn't understand it. darn semantics, eh?
nvidia making money by getting customers to buy one pro card with limited compute perf. and then get them to buy another with high compute perf. is just good business. all these despite cuda being locked and proprietary. otherwise amd and nvidia's position would be much different in pro/hpc market and nvidia wouldn't sell as many cards - gaming or pro. in mainstream situation it'd be like using an nvidia card a for physx, even when one uses an amd card as main gfx card.

This isn't even an argument against Nvidia.

you're right, finally.

 

[xomm]

So basically, the true successor to the GTX 550 Ti?

 

[flyflinger]

I look forward to next month's Buyer's Guide ranking... should be very interesting.

 

[slomo4sho]

You clearly bought cheapshit then. XFX let me guess???

XFX actually has some pretty good coolers. It's probably Powercolor's Myst Edition. I don't think there even is an XFX 7870 LE/XT card.

Yeah, think you are right. Powercolor is not great either.

How am I quoted for something I never said

 

[Justarius]

I don't understand why this review is comparing the 2GB 650 Boost to the 1GB 7850. Both cards come in 1GB and 2GB variants, why aren't we comparing apples to apples? It's not like the 2GB 7850 wasn't readily available, if anything it's the 1GB that is hard to find... It first I assumed this was why Anandtech's review scored higher for the 7850 but I see some of the other sites did a proper comparison of the two and still score the 650 Boost higher. Still, seems sloppy Tom's would use unbalanced memory and then make such a big deal over the comparison. None of the other reviews did this.

 

[Duke Nucome]

You clearly bought cheapshit then. XFX let me guess???

XFX actually has some pretty good coolers. It's probably Powercolor's Myst Edition. I don't think there even is an XFX 7870 LE/XT card.

Yeah, think you are right. Powercolor is not great either.

How am I quoted for something I never said

I don't understand why this review is comparing the 2GB 650 Boost to the 1GB 7850. Both cards come in 1GB and 2GB variants, why aren't we comparing apples to apples? It's not like the 2GB 7850 wasn't readily available, if anything it's the 1GB that is hard to find... It first I assumed this was why Anandtech's review scored higher for the 7850 but I see some of the other sites did a proper comparison of the two and still score the 650 Boost higher. Still, seems sloppy Tom's would use unbalanced memory and then make such a big deal over the comparison. None of the other reviews did this.

It's because the 1GB 7850 is the same price as the 2GB 650ti boost. That said these card don't even have the core power to warrant more than 1GB Vram.

 

[ingtar33]

It's because the 1GB 7850 is the same price as the 2GB 650ti boost. That said these card don't even have the core power to warrant more than 1GB Vram.

bingo. the 7870 is the lowest end gpu to warrent 2gbs of vram. neither the 7850 nor the 650ti boost will saturate 1gb... unless they're in xfire/SLi, but that's about the only time they might saturate 2gb. Get a 2gb version if you plan to xfire/sli low end gpus in the future. otherwise skip them.

 

[Isaiah4110]

I'm pretty sure common knowledge states that the 2GB version of the 7850 gives much better performance at 1920x1080 than the 1GB version, especially if you want to use any sort of AA...

 

[Duke Nucome]

I'm pretty sure common knowledge states that the 2GB version of the 7850 gives much better performance at 1920x1080 than the 1GB version, especially if you want to use any sort of AA...

http://www.bit-tech.net/hardware/graphics/2012/11/05/ms...
In Battlefield 3 at 1,920 x 1,080, the MSI HD 7850 1GB was actually able to outperform the 2GB version thanks to its factory overclock, albeit by a single frame per second on the minimum frame rate. This suggests that the 2GB's extra 1GB of GDDR5 is all but useless at this resolution.

http://www.legitreviews.com/article/2048/13/
When it comes to gaming performance the XFX Radeon HD 7850 1GB Core Edition did better than expected with a screen resolution of 1920x1080! We were able to easily play all the current game titles at 1080p with the image quality settings cranked up! There were a few game titles like Skyrim where we felt a minor hiccup here or there when the frame buffer would fill up, but they were few and far between.


http://www.techradar.com/reviews/pc-mac/pc-components/graphics-cards/msi-hd-7850-1gb-1129390/review
"After all, the 2GB frame buffer of the stock HD 7850 must be partly responsible for its impressive showing at hi-res, right? And that's putting post-processing such as anti-aliasing in the mix.
How wrong we were. It seems that the bottleneck to the Pitcairn Pro's performance is not the memory at all - the GPU itself bottlenecks performance well before the memory gets anywhere near making things crawl.
What does that mean though? Well, it means the 1GB version of the HD 7850, especially the slightly overclocked version supplied by MSI, is more than able to keep pace with the stock 2GB version of the Pitcairn Pro card."

 

[Isaiah4110]

Heh, interesting. I guess "common knowledge" can be wrong sometimes. Good to know.

 

[Duke Nucome]

Heh, interesting. I guess "common knowledge" can be wrong sometimes. Good to know.

It often is when the hype machine gets all revved up and people start to spread somewhat inaccurate information. Truth is Vram is more of a marketing tool used to sell more product than anything. The only place that I have noticed where more Vram makes a difference is triple monitors.

 

[ingtar33]

The only place that I have noticed where more Vram makes a difference is triple monitors.

bingo

multiple monitors and huge resolutions will saturate 2gbs... on a single monitor at 1080p you only need 1gb.

I think gaming across 3 monitors at super high res might require 3gbs... nothing will really need more then that.

 

[Slavens]

I think I missed something here. This isn't an attempt to troll, I'm trying to figure out how the benchmarks lead to the 650 ti Boost besting the 7850.

I'm not an 'ATI or NOTHING!' freak. I buy what makes sense at the time.

First, the 7850 is missing from Conclusion, Performance per dollar, so you can't just make a quick direct comparison:
http://www.tomshardware.com/reviews/gaming-graphics-card-review,3107-performance.html

Second, Tom's compares the 2gb 650 ti Boost against the 1GB 7850 and the 7850 bests the ti Boost in three of the five titles

As for the OpenGL, I'd like to see actual openGL games benchmarked. There's more to an openGL title than pure rendering. There's far more to a game than rendering. We've seen the impact of ambient occlusion, 'god-rays', and extreme anti-aliasing setting on the raw framerate of video cards.

I also have to question the use of Autocad as a benchmark, mainly because I feel that it places too heavy of an emphasis on pure rendering. While synthetic benchmarks are always part of of a good review, we benchmark performance in specific games using specific video configurations because we know that these provide two important bits of information:
- a 'real-world' expectation of performance of a given product
- a point of comparison between competing products in a meaningful manner, across price points and time (as new cards are released)
If we're going to do OpenGL/CL benchmarks in Bang-for-the-buck gaming cards, I'd like to duplicate the D3D efforts. If we benchmark the D3D systems synthetically, do the same for OpenGL. If we benchmark the D3D with games, please use games for OpenGL.

I also have to express disappointment in the lack of discussion over the memory bandwidth. If you look through reviews of video cards that have a similar GPU but varying memory buses, you'll see that cards with higher memory bus widths see smaller framerate drops as video settings are increased. Similarly, cards with high memory bus widths typically see longer service lifes - take the Radeon x850 series for example. I'd give an nVidia example, but I'm not as familiar with their cards.


My problem is that the 1GB Radeon 7850 bested the 2GB 650 ti Boost it in 3 out of 5 games with half the RAM and has a 75% wider memory bus. I don't see how the Boost 'wins'.

Further, at the moment, the 1gb versions of these two cards are both $149 on NE and the
2GB 7850 is $10 cheaper than the 2GB Boost at $159 and $169 respectively. Given the $10-$20 difference between the 1GB and 2GB versions of the cards, I'd skip the 1GB versions entirely.

 

[Slavens]

I think I missed something here. This isn't an attempt to troll, I'm trying to figure out how the benchmarks lead to the 650 ti Boost besting the 7850.

I'm not an 'ATI or NOTHING!' freak. I buy what makes sense at the time.

First, the 7850 is missing from Conclusion, Performance per dollar, so you can't just make a quick direct comparison:
http://www.tomshardware.com/reviews/gaming-graphics-card-review,3107-performance.html

Second, Tom's compares the 2gb 650 ti Boost against the 1GB 7850 and the 7850 bests the ti Boost in three of the five titles

As for the OpenGL, I'd like to see actual openGL games benchmarked. There's more to an openGL title than pure rendering. There's far more to a game than rendering. We've seen the impact of ambient occlusion, 'god-rays', and extreme anti-aliasing setting on the raw framerate of video cards.

I also have to question the use of Autocad as a benchmark, mainly because I feel that it places too heavy of an emphasis on pure rendering. While synthetic benchmarks are always part of of a good review, we benchmark performance in specific games using specific video configurations because we know that these provide two important bits of information:
- a 'real-world' expectation of performance of a given product
- a point of comparison between competing products in a meaningful manner, across price points and time (as new cards are released)
If we're going to do OpenGL/CL benchmarks in Bang-for-the-buck gaming cards, I'd like to duplicate the D3D efforts. If we benchmark the D3D systems synthetically, do the same for OpenGL. If we benchmark the D3D with games, please use games for OpenGL.

I also have to express disappointment in the lack of discussion over the memory bandwidth. If you look through reviews of video cards that have a similar GPU but varying memory buses, you'll see that cards with higher memory bus widths see smaller framerate drops as video settings are increased. Similarly, cards with high memory bus widths typically see longer service lifes - take the Radeon x850 series for example. I'd give an nVidia example, but I'm not as familiar with their cards.


My problem is that the 1GB Radeon 7850 bested the 2GB 650 ti Boost it in 3 out of 5 games with half the RAM and has a 75% wider memory bus. I don't see how the Boost 'wins'.

Further, at the moment, the 1gb versions of these two cards are both $149 on NE and the
2GB 7850 is $10 cheaper than the 2GB Boost at $159 and $169 respectively. Given the $10-$20 difference between the 1GB and 2GB versions of the cards, I'd skip the 1GB versions entirely.

 

[somebodyspecial]

One day, maybe they'll remember OpenCL isn't cared about by NV at all because nobody tries to make money with this crap. You run CUDA in AdobeCS6 (video or photo etc), Blender, 3Dsmax, Vray, etc etc. Pro apps use CUDA (with perhaps an option for opengl or opencl).

Start running CUDA vs. whatever AMD runs fastest with (opengl or opencl) in apps testing gpgpu. Nobody buys a NV card to run opencl stuff. NV doesn't care about an opensource driver when they already can do the same things faster in Cuda or OpenGL in EVERY pro app that makes you money. Testing gpgpu without showing cuda is pointless. A real pro app user would not act like 7yrs of cuda groundwork doesn't exist. IT DOES exist, and it's a HUGE difference in some things. IE your own ps cs5 tests shown here:
http://www.tomshardware.com/reviews/adobe-cs5-cuda-64-bit,2770-8.html
3mins vs. 12mins (6core cpu)...WOW.
http://www.blenderguru.com/4-easy-ways-to-speed-up-cycles/
Only 12x faster for ticking a box for cuda...12x faster sucks right?

I would only turn on OpenCL if there was NO other app that did the same job with Cuda. But then, that situation doesn't exist so... Which explains why NV pushes cuda and openGL and basically ignores OpenCL. I mean, why would they want to foster AMD optimizations in apps? Helping OpenCL means AMD gets into the game as apps add support. They don't desire this. Not to mention cuda in the same app would (should) never be beaten by opencl, since it's kind of an abstraction layer above directly talking to the hardware with cuda so to speak.
http://www.heatonresearch.com/node/2487
"If you want to use nVidia, ATI and Intel, then you will use OpenCL. Sounds like a slam-dunk to use OpenCL? Right? CUDA is much more advanced that OpenCL. In OpenCL you are programming your graphics kernels in C (actually C99, but still C). In CUDA, you are using C/C++. In OpenCL, you are dealing with a higher-level abstraction. If you are on an nVidia card, OpenCL is essentially compiling to CUDA. Because of these reasons, I decided to go directly with CUDA for Encog's GPU implementation. This limits me to nVidia cards, but that is something I am willing to accept. At least for now. I may add an OpenCL version at some point in the future, but that is not planned at this point."

There's an example of what many devs think, especially when NV funds some help (like all big apps) to code in Cuda. Besides this why do I care how fast I can help someone solve cancer etc? Bitcoining is over for all but large botnets as easy blocks are gone.
https://community.rapid7.com/community/infosec/blog/2012/12/06/skynet-a-tor-powered-botnet-straight-from-reddit
Just an example...Oh look it uses opencl...LOL. Well a botnet would have to be cross-hardware compatible right?

So what is it with testing open source JUNK? I get that a lot of it's free and works on a lot of stuff, but does anyone use this stuff to make money? Based on registrations 163,000 folding@home site, folding@home is pointless to most and so is bitmining. There are 352MILLION computers sold each year and you're benchmarking for 163000 people (and why the heck do I want to run up my electric bill for that? Will they pay me when solved?)? How about some meaningful REAL MONEY MAKING apps start get tested for showing gpgpu stuff? You can use luxrender plugins for stuff like 3dsmax, blender 2.6, cinema4d etc, but again as soon as you do that I turn on Cuda. OpenCL should always be a last option if forced when there is a cuda way to do the same thing on NV. You guys are always forcing NV cards into the worst gpgpu situation they could be in, which nobody would actually do, then acting like it's normal instead of retarded This amounts to this lame excuse:
"Previously, when the GPGPU universe was divided into CUDA (Nvidia) and Stream (AMD), we faced the problem that most applications supported only one of the two environments, and could thus not be directly compared to each other."
http://www.tomshardware.com/reviews/graphics-card-benchmarks-charts-review,3154-8.html
It's AMD's tough luck most apps skip them because they haven't invested for 7yrs like NV has right? Take the best app from both and compare those running the same scene render etc if this is the excuse you'll give for an answer. Then again everyone uses adobeCS, so it's hard to imagine not using it as a benchmark while showing a bunch of stuff nobody uses for real life instead. Odd...If stream has fallen so far behind cuda that it's useless that isn't NV's problem, nor should you act as though the situation doesn't exist. 500 universities in 26 countries teach cuda. How many teach stream or opencl use? You can use OpenCL in Adobe and anyone with NV hardware will immediately turn on cuda. The # of people using AdobeCS vs all of what you've shown combined is ridiculous. They sell 4Bil worth of the stuff yearly. Surely you can do better than this platform agnostic crap.

OpenCL is rapidly replacing CUDA as the go-to language for GPU compute and Direct3D replaced OpenGL almost completely several years ago. You're being so far out of the loop that it's not even funny. Also worth mentioning is how Direct Compute tends to beat both OpenCL and compute while, like OpenCL, not being limited to Nvidia cards.

It's also worth mentioning that Adobe is one of the biggest companies ditching sole CUDA support and switching to OpenCL/Direct Compute.

Tom's tests a lot of compute workloads that many people uses. It's generally Anand whom tests a lot of non-important things such as years outdated application suits.

Any proof? IN what? Cuda is still in adobe so basically AMD is handing over some cash to get it in there. No surprise, it's a good move as lots of content comes from that suite.

Proof of rapidly replacing cuda please. I disagree with Toms workloads on opencl also. When there is a way to test cuda (via some other plugin etc) you should be comparing them both to each other, not showing how bad opencl is on NV (well duh, because they already can do X in Cuda so why waste on opencl?). F@H/Bitcoin etc is uncommon. Opencl for crap like blur, photo stuff, vid editing etc is uncommon, people use adobe and cuda. You will finally be able to compare these soon, but they should compare NV running cuda on adobe, vs the same photo or vid job on opencl for AMD. Pretending cuda isn't used like this (adobe) is misleading. It amounts to FINDING a way to do these operations in opencl so AMD can actually attend the party, instead of just using Cuda in adobe like the majority of content people do currently.

 

[renz496]

why everyone assuming Adobe already drop CUDA? for all i know the latest Adobe still have CUDA in them.

http://blogs.nvidia.com/blog/2013/06/19/5-reasons-to-choose-nvidia-gpus-for-stunning-performance-in-adobe-creative-cloud/

OpenCL might be much more known than before but it doens't mean it will kill CUDA just like that. both CUDA and OpenCL have their own advantage over another.