Reviews - ZOTAC GeForce GTX 470 - tested in SLI and compared to Radeon HD 5870 XF

Quote:

---

So how do two perform? Read on to find out.

---

Read more.

Still happy with my GTX260 SLI & Q6600, but would make a good new build when the new Phenoms x4 come out using asus unlocker to make x6 cores and 5850 XF.

Interesting artical, it's interesting to see that both ati and nvidia high end cards are getting decent performance gains from xf/sli.
However one thing this artical shows is why would you sli the 470? if you can afford that sort of money on graphic then go the full way and get sli 480's.
The 2x 5850 in xf vs single 480, works because the price is around the same and while the 470 and 5870 cost the sameish, the 5870 xf is a more possible simply because it's been around longer so people with a single 5870 may want an upgrade.

Also odd to see the dual gpu cards performing below that of xf, the card? or is it the case that a single dual gpu card is actually starting to saturate the pci-e 2.0 x16 bandwidth?

Still annoying to see Hexus use nvidia's own list of benchmarks almost exclusively without considering many other titles. Despite this fact though, the GTX470 still comes off pretty badly. It can barely match the HD5870CF setup in nvidia's own test suite, and in Bad Company 2 actually loses to HD5850CF, let alone HD5870CF. SLI scaling may be good, but CF scaling is far from bad, and with such weak performance to start with, SLI does not save the GTX400 architecture from its badness.

Quote:

---

Originally Posted by sammorris

Still annoying to see Hexus use nvidia's own list of benchmarks almost exclusively without considering many other titles.

---

I've almost accused them of doing similar in the past, but the current suite seems slightly better balanced than some in the past, so I wouldn't be too quick to say the same this time. I'm sure Hexus would happily be able to confirm there are no conditions/recommendations on games they test nVidia cards on.

Check the GTX400 series benches. HAWX, Far Cry 2, Bad Company 2, DiRT 2 and a Crysis, be it the original or Warhead. Even in sites that never normally use any of those games, that's always the list.
HAWX because ATI messed up performance in it with the HD5 series, DiRT 2 because there was a driver problem that caused ATI cards to be slow (meant to be fixed pretty soon), Far Cry 2 because it shows off the GTX4 series' extra memory, and Crysis because it's innately heavily biased in nvidia's favour. Bad Company 2 is also biased in nvidia's favour, but less so, and it would appear drivers have since removed most of that bias.

ever thought that game work better on some nvidia cards becuase they are better than their ATI counterparts? :p

im impressed by the GTX470 SLI results, bar bc2 they keep up with the 5870, and even the single GPU results. i think now the GTX470 atually justifies its price. (slightly)

Considered it, but the number of games it happens in is not high enough to suggest that being the case. Personally, I would want to read more benchmarks that contain results from games other than those suggested by one of the brands being tested before I'd call that decision. Ultimately the HD5870CF is still marginally superior even given this. Add that to the fact that they're pretty much the same price, the GTX470 is still some way from being justifiable.

Hello, all.

Battlefield: Bad Company 2 - a game that runs much faster on ATI hardware and is heavily promoted by the red team.

DiRT 2 - provided with most high-end Radeon 5-series cards, and here's a quote from AMD:

"AMD has worked tirelessly with Codemasters to collectively transform realism in PC gaming through new DirectX 11 technologies only available today on the latest ATI Radeon graphics cards," said Matt Skynner, vice president & general manager, GPU Division at AMD.

Crysis - because everyone complained when we took it out. It's semi-useful to see how GPUs perform with a hideous amount of load imposed via the enthusiast setting.

Far Cry 2 - Provides decent scaling in multi-GPU setups.

H.A.W.X - initially added to demonstrate benefits of DX10.1. Produces consistent results each time and is also useful for gauging multi-monitor 3D performance.

You will see these games benchmarked in ATI's briefing documents, as well as NVIDIA's, so we'd say the line-up is pretty fair. In fact, if anyone has a case of complaining, it's NVIDIA - we don't go out of way to add a PhysX-enabled title (much to NV's chagrin) and have rebuffed attempts to add specific NVIDIA-optimised games to the suite.

As always, feel free to let us know what you would like to see added, and why :)

As Tarinder says, make some suggestions for games you would like to see in there!

thanks for that reply, clears that up :)

so ultimately tarindar, CF5870s or SLI470s?

DiRT 2 is indeed provided with most Radeon cards, which is why it's embrassing for ATI that they got it so wrong. Recent drivers have boosted the performance of Radeons hugely, almost up to the standards set by the opposition, not quite. They'll get there eventually I suppose.
Bad Company 2 running faster on ATI hardware? Must have missed something here, not seen much evidence of this elsewhere.
Crysis should always be included in a benchmark, simply because of its cult status and legendary system requirements. No complaints from me it being included.
FarCry 2 is rather biased for multi-GPU testing as it cannot support any more than 2 ATI GPUs. I don't think this affects nvidia cards to my knowledge.
HAWX is something that gave the HD4 series a huge advantage due to DX10.1. Sadly, DX10.1 was completely broken on the HD5 series when it was first released. I'm not entirely convinced they've fixed it yet.

PhysX titles should be avoided at all costs, and they are at least excluded from Hexus tests. More neutral tests like Assassins Creed 2, and ATI-biased titles such as The Settlers 7 and Napoleon Total War should ideally be included to show comparisons from both sides.

Hexus reports are still missing minimum fps figures as well, which is a fundamental part of benchmarking. 59fps average is no good if the minimum is only 21 (DoW2).

yeah minimum FPS figures are the only thing missing IMO.

maybe MW2 becuase thats played quite a bit. and once in a while a massive round up including some 8800GTX to the newest cards. maybe a £1,000,000 give out for me aswell.... its not much to ask :P

Yup, noted. The graphing system is changing to accommodate the minimum fps - we have all the numbers saved - and you will see that very shortly. :)

Somewhat agreed, though games like MW2 are so undemanding that really a single 5870 or 470 should be up to the job even at 2560x1600. It is appreciably difficult to decide what titles should be included for reasons like that. However, a fully comprehensive test once in a while would be very much appreciated, for example:
http://gamegpu.ru/images/stories/Abz...2/1920_9_2.jpg
These are incredibly helpful. They do, however, lack SLI/CF results as the site has to buy the cards itself. Now granted, not all of these cards will be easy to find without substantial expense, but such a review, preferably including a wealth of titles, not just the usual half a dozen or so, would be amazing. There are lots of very low-key titles that a large number of people play that are surprisingly demanding. CPU performance is also something to consider. It need not be that comprehensive, but for example strong dual core, weak dual core, strong quad core, weak quad core, might be good (for example, i5 650 vs E5200, i7 930 vs Athlon II X4 620 or Q6600). Peak memory usage would be handy too.

There are loads of things that could be done to make an amazing benchmark suite that seem to be picked up on in bits and pieces by different sites but never quite strung together into one big benchmark. The resources for this sort of undertaking may be quite large, but hell, I'd donate a small sum of money to see something on that sort of scale!

oo yes. like use the post powerful single and GPU and the most powerfull multi GPU setup and bang it on a high res and see the effects of CPU scaling, from a lowely E5X00 to a 980X and then select a few older ones and recent ones and see if overclockign the CPU will help, ie take an old c2d and see if OCing will help it to the same FPS as a newer dualcore, same with a Q6600/Q9450. legion hardware did this once.


ofc i can understand the workd being these would be immense and itd take long time, but once a year this would be superb!

otherwise the game selection IMO is very good currently.


love that you have the minimums ready, is there a way for us lowly people to measure them in game ourselves?

Lol, doubt they kept a 980X hanging around for benchmarking purposes :P
Wish I had the money to get a load of stuff like that, I waste enough free time to warrant doing all this myself :P

Fraps will benchmark games fully, considering the fps every refresh (graphable), and the amount of time it took to render each frame, so you can check for spikes (also graphable). This allows you to track average, minimum and maximum fps in each game.

i see it now, fraps does minmum aswell. nice.

hmm...

I'd like to see minimum fps charts (ideally replacing av. fps charts as it will give a better perspective on what these cards are actually capable of doing), Metro 2033 benches would be nice (might involve getting ATi to actually work on the drivers for it though), maybe 3D results when the DDD kit is released, or using the IZ3D monitor and also try and get a stable OC at the beginning of the review and do tests with OC and at stock or OC at the end of your review process and go over the tests again.

I know a lot about benchmarks and the current suite is slightly tailored towards favouring nVidia.

BC2 favours ATI because nVidia's driver team hasn't had a chance to work on it yet.

FC2 is so far in favour of nVidia it has to be considered proprietary. It's old and nobody really plays it, so why keep benching it?

Dirt 2 is fair because even though it's an ATI game, nVidia wins at it. Expect nVidia to get even better as time goes by. As a dx11 game, this is a good win for nVidia and a problem for ATI - it should be included because of that.

Crysis - fair, it's a killer of all gpu's.

HAWX is just old and bad and doesn't deserve to be used as a benchmark for dx11 cards.

I would wager that most of those games made it into nVidia's "reviewer's guide". All except BF2 and maybe Crysis, those were probably replaced with Metro and Metro with PhysX. :D

Quote:

---

BC2 favours ATI because nVidia's driver team hasn't had a chance to work on it yet.

---

I doubt that very much, the game's been out for quite a while now.

I agree with FC2, though it's more biased pro-SLI than pro-nvidia per se.

Yeah Metro 2033 is the obvious one, but when I found out nvidia were only faster because they hacked 'very high' details back down to standard 'high', I lol'ed - that probably isn't nvidia-biased either, truth be told.

The only time minimum fps is valid in terms of gaming performance is when it's shown as a timegraph like Hardocp does it. As a single number on a chart, it can be abused.

Example -

A 5870 runs BC2 at 50fps average, 80fps high, but one time in the benchmark it falters and scores 10fps. The true 'low' is closer to 30fps over most of the benchmark.

A 480 runs BC2 at 40fps average, 70fps high, but one time in the benchmark it falters and scores 20fps. The true 'low' is closer to 30fps over most of the benchmark.

Does the 480 really deserve to be twice as good on minimum framerate? Of course it doesnt, and that is why only a timegraph is useful when minimum framerates are included.

why not avg the minimums?

Well the overall average is just the average minimums subtracted from the average highs. :p

Time graphs aren't difficult to do, all it needs is fraps and a spreadsheet.

I dont think there is anything wrong with just averaging the whole range of numbers, you guys need t stop moaning and let them get on with it.

Some of the games are a bit odd in the review ine up i do agree with that. All the games that are int he review line up should always be the top played games of the time. BFBC2 is a definite win, where is MW2? I dont liek it but a lot of people do.

Company of heroes should never have been taken out, its old but its still popular and with full settings is pretty demanding.

Jimbo: It's not difficult, remove anomalies like that, and state the minimum figure that is often observed. Full frame rate graphs are nice of course. Out of curiosity, who reported the figures you state?
Biscuit: Using minimums raw might add some strange readings (which would no doubt be explained) but just using average alone misses out on a lot of the detail. It ignores a large number of issues.
Call of Duty, I'll be honest, doesn't belong in a benchmark of high-end dual GPU setups. The frame rate limiter is set pretty high in it, and everything here would exceed said limit, the game just isn't that demanding.
Company of Heroes on the other hand, whilst far older, is still worthy of consideration, as RTS games often suffer lower frame rates.

Hardocp does time graphed benchmarks. They are pretty cool so long as there aren't too many cards on the graph, at which point things can start to get messy. Nobody reported the figures, I just used them as an example of how a single minimum drop can give a false impression.

Best way to do it - raw data in a script - choose which cards to display and not display on the graph, render it on demand. A good anti-hotlinking tool too :P

Concerning the Price of the GTX 470 you can purchase one from ARIA.co.uk, arias on version for £278.84 including delivery which is 30 to 40 GBP cheaper than the HD 5870 - and more importantly is in stock :)

Quote:

---

Originally Posted by Smacks

Concerning the Price of the GTX 470 you can purchase one from ARIA.co.uk, arias on version for £278.84 including delivery which is 30 to 40 GBP cheaper than the HD 5870 - and more importantly is in stock :)

---

5870 is still in stock, but you're right, they're about £15 more expensive than 470s excluding outliers. But that seems to be a worthwhile price to pay for the advantages.

Yeah, Aria have some decent prices on the 400 cards, and better stock levels than most, though the only cards below the standard pricing points (i.e. £300 for the 470 and £450 for the 480) are OEM cards, so it's not an amazing deal. That, plus the prices still haven't droped to appropriate levels. The GTX470 is an HD5850 rival, not an HD5870 rival, and thus it should really be at the £250 mark, for good brands, not just OEMs.