nVidia GT 335M vs. ATi Radeon HD 5730 Mobility

Pretty simple question. Which card is better. I heard that the Radeon is better than the nVidia but the GT 335M has 72 Unified Cores while the 400 Unified Cores from the ATi is equivalent to 64 Cores right? Therefore, should the nVidia be a better graphics card due to the additional cores?

Therefore I was wondering which card is better for 16.1" 1366x768 gaming?

 

Depending on the RAM used in the GT335M, and it's core clock speed, it could very well be quite better, or worse than the HD5730.

The HD5730 is barely better then the HD5650, which is slightly better than the HD4670 GDDR3 right? So I don't think it will be hard to top.

IMO ATi dropped the ball with the HD5700 series. They should have been 600sp MINIMUM not 400sp.

The only worthwhile card from ati right now is the HD5870, or the HD5850 (with GDDR5, if it ever comes out in that form).

 

Since HD5650 is faster in most games than GT250M (96 CUDA cores and 360 gigaflops) I see no way how GT335M (72 CUDA cores and 233 gigaflops) could be any faster than 5730. Especially that even GTS 250M with GDDR5 wasn't (I mean that one in Toshiba)
That's based on test of 5650 by notebookcheck and notebookjournal.

 

You can't compare SP's across different architecture's like that - and that's not to count the differences in TMU, RBE, and other unit counts, much less clock differences.

As downloads wrote, the 5650 at 600/800 beat the GDDR5 GTS250M, which should be faster than the GT 335M (meant to replace the GT 240M). So the 5730 should easily beat the 5650.

And they didn't put 600 SP's in the 5650/57xx's cause it's based on the Redwood cores, and it makes no financial sense for ATI to cut down the 800SP Juniper cores when 40nm cores are at a premium right now (see: supply issues with 40nm)

 

Also I believe the OPs math is incorrect. The 5730 has 400 5 dimensional sp, so 400/5 = 80. That would mean it has more cores than the GT335m. I actually own an ASUS with the 5730 and it is very fast indeed!

 

Could a Vantage 3D score be used to determine "better"? I'm curious what the 5730 scores.

 

There's one, big problem with this comparison. The only 335M found in a notebook (M11x) is paired with a CULV processor. Also, it has purposely been clocked down to 450mhz.

A GTS 250M with GDDR5 would obviously defeat either, as would a GDDR5 5770M. If the 335M were clocked at 550mhz, and paired with the same CPU, it would be very equal to the 5730.

 

So your take on it is that if the CPU is equal and the cards are all at stock speeds designated by ATI/NVIDIA the 5730 and 335M should be about equal?

 

So the 335m, which is an overclocked 330m, which is an overclocked 240m, which is an overclocked 230m (goddamn, Nvidia, be a little more creative! Or hire a new development team!). It's underclocked in the M11x? What's the point of calling it a 335m, again?

I'd much rather go with a 5730, which should be faster than the Nvidia GT230m (or whatever they're calling that card in whatever laptop we're talking about).

 

^ histidine, I don't think that's the case though..

 

Alright, what's the difference between the abovementioned cards?

I'm no expert, so I could definitely be wrong.

 

Uh. The GT330M is a higher clocked version of GT325M.
The GT 335M has an entirely new combination of shaders and is not a overclocked part of anything. It's supposed to be about the same as the GTS 250M though.

(GT335M benchmark: http://it.sdinfo.net/article/1657/1657224.html)

AFAIK the 5730 in combination with a i7 quad core scores about 7700 - 7900, the GT 335M with the same CPU scored about 7500 at 1366 x 768. There's not a huge difference.

 

What about the ATI 5750m ? It has GDDR5 ram, how would this perform?

 

I haven't seen this card in any notebook so far.

 

You're right. It is only just appearing now.

A new Packard Bell has it:

http://geizhals.at/eu/a507174.html

I'm close to buying it, this is why I want to get an approximation of how fast it is, but I guess noone has a clue since it isn't out yet...

 

Huh, you're right. That's abandoning Nvidia's previous naming conventions, with the 5 in the 3rd digit usually meaning an overclocked version (ie 9650, GTX285).

 

Packard Bell exists???

 

lol, did you follow the link?

 

Just read up on PB, apparantly it was bought by Acer in 2008..

 

I see, interesting. yeah I followed the link and saw the name =) hehe that and some foreign characters. Nice to know.

 

It's in German

As far as I've been able to read, the 5750m is basically a 5650m with GDDR5 ram. The PB I linked to only has 512 MB, but should still be ages better? It has 51,2GB/sec memory bandwidth, double of the 5650m.

And the 5650m already gets 7300 3dmark06.

The PB costs around 700 € (euros), which would be around 955$. Seems like a sweet deal to me, knowing that some more expensive notebooks only have a 5730m..?

http://www.amd.com/us/products/notebook/graphics/ati-mobility-hd-5700/Pages/hd-5750-specs.aspx

 

Except that in the notebookchecks review, the GTS250M w/ GDDR5 actually loses to the 5650...

3dMark06 scores are an awful way to compare actual in game performance, unfortunately. First of all, the bench is incredibly CPU dependent. As an example, my 4870X2 on my desktop drops by 4k points by downclocking my CPU by 25%. Downclocking my GPU by 25% only drops my performance by 1k or so.

Just looking at the 5650 benchmarks from the notebookcheck review, the 5650 matches or beats the GTS 250M and the Mobility 4670 pretty often, and even gets close to the GTX 260M in some scenarios. The GT 335M is meant to replace the GT 240M, but is under the GTS 350M in clocks/shaders/TMU's/ROPs etc. The GTS350M being the GTS 250M should mean that the 5730 will beat the GT 335M pretty often.

Of course, the GT 335M is in a different form factor altogether than typical notebooks with the 5730

 

Well... thing is, though, the benchmark scores I mentioned are of the two cards with the SAME CPU. Under that circumstance I think I can argue that the video card would have made the most significant difference.
In any case, right now I seem to find the notebookcheck's reviews a little odd, especially their collection of user submitted game statistics. Nobody knows what the resolution used is, and nobody really knows what "low", "medium", and "high" are.

Although, your last comment was interesting about the form factor. The HD5650/5730 both have very low TDP, with the 5650 being under 20W and the 5730 not that much over 20W. Should stand to reason that someone COULD put a 5730 in a 14"... Hmm.

 

Same CPU or not, 3dMark06 will limit a card's full performance in it's benchmark because of the CPU limitation compared to real world performance.

An example again is when I used my QX9650. At 3.6GHz with a default 4870X2 (at 800Mhz core), I scored ~19000 or so. I downclocked my 4870X2 by 25% to 600MHz core while keeping CPU standard. My scores dropped to ~18000. Then I kept my 4870X2 at 800MHz core, but downclocked my CPU to 2.7GHz. Lo and behold... 3dMark06 went down to ~15,000.

Yet in real world games, a 2.7GHz vs. 3.6GHz would've seen far less of a difference than 600Mhz vs. 800Mhz.

In other words: it's nice to compare 3dMark06 for laptops with the same CPU, but don't take it as any indication of how real world performance relates.

You can look up 5730 benchmarks in the ASUS N61 thread / other laptops with 5730's out now. A lot of benches put it somewhere above the GTS 250M in real world performance figures.

And yeah, I bet a 5730 (and certainly a 5650) can fit into 14" notebooks. I bet the current issue is cost

 

Hmm. I think you have a fair point. I'll read up on these benchmarks a bit more.

 

This may be news to some, but a benchmark isn't always required to know which GPU is faster. In fact, benchmarks from sites like Notebookcheck often just lead to confusion. Notebookcheck is nothing like Anandtech or Tom's Hardware, they don't test GPU's on a base system. So pretty much anything can be skewed, doesn't mean everything is, but it certainly means to check everything from that site twice, and then thrice. Really.

 

That is what I think too. It feels like it's just like the Futuremark's score database on their site. All user submitted with no standardization.