ATi Noob Crossfire Review - X38 vs P35 / Vista vs XP

Intro
Picked up my new HIS HD4870 cards today 8) I'm kind of unprepared as this is more a trial than anything, to see how much of a drama it is to run a multiple GPU setup. Testing if my PSU can handle it, checking how often BSOD's invade my personal space, seeing if it all fits in the case and so on. I'm unprepared as although my motherboard supports Crossfire, it supports it at 16x/4x bandwidth. So one card runs with full bandwidth available, the other card is limited to 4x bandwidth, which will no doubt hinder performance. If you're going to seriously Crossfire, you want to have dual 16x PCI-E lanes.

First impressions

Packaging and bundle was great. The boxes are very small in comparison to some other ridiculous video card boxes i've come across (cough, Asus 8800GT). Looks are decieving however, as although these boxes are small, you can tell they're packing some beefy hardware inside, just by picking them up.

Bundle

The HIS bundle doesn't come with any games or 3Dmark software you may find in other brands, but you do get everything you need including; DVI->HDMI adaptor, DVI->VGA adaptor, HDTV adaptor, Crossfire bridge, plus what appears to be an S-Video->TVout adaptor and your typical driver CD and manual.

Tool

You also get this handy little tool which has a screw driver with interchangeable heads and a bright torch.


The Cards

Well there's not a whole lot i can say about the cards. The first thing i noticed was the back of the card. It's crazy! I don't pretend to have a clue what it all means, but the back of these cards looks so busy it makes an 8800GT look like it was made in the 1950s. The Carbon fibre looking section on the coolers sticker i think adds some tasty character to the card. I thought it was interesting to note that nowhere on this card says "HD4870", which i thought was a little odd. The 4870 requires two 6-pin PCI-E power connectors, so if you're going to crossfire a pair, make sure your power supply has adequate cabling, either through 4x PCI-E power cables, or 2x PCI-E power cables and 4x Molex power cables with adaptors.

Installation

It's the first time i've felt as though the P182 is a little bit cramped! Unfortunately i had to remove my capture card (sorry GTAIV guys!) as i just don't have enough PCI slots available. Even the X-Fi got marched off down to the bottom of the board to make room for the overzealous installation that is "Crossfire". No biggie, let's slot them in, plug everything back in and fire it up already I did have to use 4 molex plugs as the Corsair HX-620 'only' comes with 2 PCI-E power cables, so please excuse the hasty installation and messy cabling. FYI, the board is a Gigabyte EP35-DS3P, but i'll go into that more in the performance section.


Test Setup
Q6600 @ 3.2GHz
4GB Crucial Ballistix @ 800 4-4-4-12
EP35-DS3P 16x/4x Crossfire & X38-DQ6 16/x16 Crossfire
(2) HIS HD4870 512MB GDDR5 @ 750/1800
Corsair HX-620 PSU
Vista x64 Home Premium & WinXP Pro 32-bit
ATi Catalyst 8.6 w/Hotfix

Neither of the cards were overclocked during any of the benchmarks.


Vista Performance

Well, it's quite clear cut that there's definitely some issues going on here. The dual 4870's barely scrape ahead of a single 4870 in the High test, due to what i'm tentatively assuming is a mixture of both average CF scaling and a severe bottleneck from one of the cards. When we hit Very High, there are major issues causing the dual cards to actually perform less than half of what the single card is capable of! I will say that there is a LOT of stuttering going on when viewing the CF benchmarks. I'm hoping that this stuttering is mainly being caused by the bottlenecked card, because it is quite honestly unbareable to watch. If i were to describe it, i would say it's like driving a Ferrari down a pot-hole filled dirt road at 300km/h, jamming your foot on and off the accelerator pedal...

I'll finish up the Crysis performance section by saying that the single 4870 high test was a pleasure to watch. 44.52fps puts the single 4870 30% quicker than my old 8800GT which scored 30.5fps on the same setup.

Well, to be honest, i didn't expect these results. A quick google of some HD48XX benchmarks will show you that these cards absolutely scream in COD4, especially when Crossfired. It's quite clear to me now that the issues is entirely with the 16x/4x bandwidth. The same jerky frame rate that was in the Crysis benchmarking could be seen in the COD4 benchmarking.

 

WinXP Performance

Well, this was quite unexpected. I re-ran the tests to make sure they were correct. Why are we getting an extra 10fps at full detail on the X38 board with a single card? One would assume it's because of the PCI-E 2.0 bandwidth, but would this extra bandwidth also show itself at 1024x768 with 0xAA? I don't think so... Who gives a shit about details when you're getting a 160fps average with the game maxed out though, amirite?

Needless to say i was a fair bit disappointed with the Crysis results. Seems as though Crytek don't give a flying firetruck about ATi with very poor Crossfire scaling. The biggest difference between the P35 and X38 can be seen in the 8xAA (may have been 4xAA, the tool sometimes lies, will confirm) tests. The "Ferrari on a dirt road" has been removed with the 16x/16x board and the playback was MUCH smoother. So, big thumbs up for that. I just really would have like to have seen more than 5fps gain going from 1 card to 2 cards at 1680x1050 on High...

Overclocking
Just going to pop this in here. I of course had to have a go at a small overclock while i was at it. So the cards were both 'auto-tuned' to 780/1090 and the Q6600 was ramped up to 3.6GHz. The results are as follows:

Of course 3Dmark see's significant gains, but it's probably moreso because of the CPU score than the actual FPS increase. With only an extra 30MHz on the Core clock i wasn't expecting any miracles and as you can see from the gaming FPS, there wasn't really any. Not to matter, i only really wanted to crack 20k


Temperature vs Fan Speed

.. 23% = 88/100c ..... 30% = 57/68c ..... 40% = 47/56c ..... 50% = 45/53c ..... 100% = 44/53c

Well, it looks to me as though i've actually got one card with a paste/cooler issue. The cards have even ventilation and if anything, the cooler card should actually be the hotter card, as it's got the X-Fi right near the fan intake. I've waved my hand infront of each exhaust and i'm 110% sure that those temp differences are correct as you can even feel the difference in air temp. A 15 degree variance is a little too much in my opinion and i'll be trying to get in contact with HIS regarding this.
EDIT: I swapped the cards around and now what was the cool card is actually the hot card. I think it's a mix of the 'hot' card getting it's air off the back of the other card and it also being the primary card, is doing more work. Nothing actually wrong with the card, it's just the primary card is always going to be hotter.


Final Thoughts


Unfortunately, with these results being the reverse of what they're supposed to be, i'm going to call it a day on benching these cards on this board. Let this be a forewarning to potential Crossfire'ers; make sure you've got a proper Crossfire board before you even consider Crossfiring. Don't do a half baked setup, as judging by the above results, you will be disappointed if you do.

I will say that i was very pleased with the compatibility and the setup's ability to play nice. I haven't recieved one BSOD or any lockups yet (touch wood). The ONLY issue i've had was during 3Dmark Vantage where the graphics were all skewed, but i think that was because i was changing from Single to Dual cards without a reboot (although that wasn't necessary when benching games). Being a complete ATi noob i did have a bit of Q&A with a nice chap from OCAU, just so i could be a bit more familiar with the Catalyst Control Centre (CCC for those in the know) and compare notes with. It's interesting to note that his CF4850 setup benches rougly the same as my CF4870 setup, however, his is on an X38-DQ6 board :wink:

Something else i meant to touch on was noise and heat. I'm sure there's many people out there who have video cards that spool up or "blow the cobwebs out" as i like to think of it, when the system first boots up. Whilst that's ok, you expect it, it is rather loud. So for those of you planning on jumping on the PC whilst the wife/girlfriend is in bed asleep, think again buddy! One thing that did get me, and gets me every time since then, was the cards seem to like blowing the cobwebs out twice, the second time being randomly. After the system has completely booted up and we're sitting at the desktop, the cards will both randomly shoot up to 4000+rpm and i tell you if it does leave stains in your pants every time i'd be amazed!

Lastly, heat. I am a little disappointed with the furnace that's going on inside my computer at the moment. Whilst i haven't had the case closed up and seriously looked into the temperatures properly, little needs to be said other than if you place your finger on the exhaust, it burns! The cards IDLE at 80c according to both GPU-Z and CCC which is incredibly suck considering most of my previous cards load below that temperature. What's more disappointing is whilst you could bring that temperature down by upping the fan from it's current 25%, it's already loud enough as it is.

Now there's only one thing left to do. Spend another $250 on a GA-X48-DS4 16x/16x board, or sell the second card... decisions, decisions...

Cheers

 

Sell your old board, get the new!!!

 

I'm glad to see you didnt have any install issues. Thanks for taking the time to share your results!

 

Hi mate , where did you get those from ? (it would be only one of three stores today)

Could you run 3D06 too please ? Certainly weird scores there ey ? Im a bit less keen on these after seeing those results . Any chance of some tests with more cpu clock too ?

 

Ok, can somebody clarify this for me? You're getting worse results in CF than single card (excluding 3DWank which frankly have no interest in)? That's how I'm certainly reading it.

 

Using a board thats not 16x/16x might be doing strange things to the cards.

 

There are still some cf issues to iron out. Future driver revisions should fix it.

 

Nice - i think i'm going to swap my 2 HD3870's for one of the HIS HD4870's through work, The 16/4 crossfire is hurting your scores/fps's alot
Get a X38 board nad everything will be much better - drivers are a bit of a problem so early in the life of the cards also

My dual HD3870 setup the max i could get with my P35 board was 17878 in 3dmark06 - got my X38 board and with my cpu clocked 200mhz lower i got 20033 points

How much you get them for?

 

Pyscho, yep... sad eh! The only thing the Crossfire setup excelled in was Vantage... i don't know why Crysis or COD4 didn't like it as much.

I get the feeling that the issues isn't so much the 4x bandwidth on one of the cards, it's moreso the fact that there's different bandwidth. 8x/8x is much better than 16x/4x, and i can only assume it has something to do with 'stop and go' nature of feeding one fast card and then one slow card... as i said, it was jerky as fark!

Aussie, i got both from ITestate yesterday. When i read your post saying they won't have any till next week i cried, but i rang them up and they said they've got them I'll bung 3Dmark06 on now.

DM, cheers. I was quite pleased/surprised that it didn't give me any issues. It's certainly things like that which lean me more towards a proper crossfire setup.

 

Crossfire in a x16/x4 pci-e 1.0 board really really isnt worth it

 

yep thats what im thinking too, like synch issues or something, perhaps as the res increases the pci-e 4x cant handle it, the card on the 16x bus then has to wait for the 4x to catch up, causing the 16x in effect to act not that much better than a 4x but due to it having to wait for the 4x to complete it's work.

Shinanigans: Is there a way to set both slots to 4x on your board? If so it would be cool to see if you can test that and see how the performance compares

 

They better have mine to me soon

 

Reason I said that is I sent a mate there on friday with cash and they said sorry none until the 1st or 2nd July ... ffs damn them

 

Id put that figure near 50% quicker

X48 is the only way forward

 

I don't believe there is. I was able to run one card in 4x though, just by plugging the monitor into the 4x card and turning off Crossfire

This what i thought was interesting:

SINGLE 4870 @ 4x ... Average FPS: 34.71
SINGLE 4870 @ 16x .. Average FPS: 41.59
DUAL 4870 @ 16x/4x.. Average FPS: 44.52

If the 4x card is capable of 34fps on it's own, why is it only capable of 3fps when paired up?

 

Or X38 - Get the Maximus Formula or P5E of Asus and flash it to the X48 Rampage Formula bios
That's what i've done

 

I ordered mine on fr imonring (like 5am) and my order is in processing, i better call em on monday.

 

What chu mean? If i had a proper crossfire setup running, both cards would equal 50% quicker? I was just comparing single cards, with the HD4870 being 30% quicker than a single 8800GT, which is pretty damn awesome, i thought

 

Those aren't the results i was expecting.