Some see the light, and some feel the heat
|
|
|
ATI HD5870 in Crossfire on Asus P6T7 WS with Intel Core i7 975XE
THESE DAYS, many high-end mainboards with Intel X58 or AMD 790 chipsets offer three or even four PCIe x16 slots for multiple graphics cards. Here we test one of them.
Note that often the additional third and fourth slots may be only physically x16, but electrically only x8 using the on-board switches to share the signal lanes. Some boards like the Asus P6T7 WS Personal Supercomputer, use twin hot Nvidia Nforce 200 PCIe bridges to expand the two x16 signal links into a total of four x16 slots with each slot still having full x16 signaling.
That comes at the cost of some latency, as the Nforce will add about 100ns extra delay to the start of a data burst. Not a big deal for long data transfers common in graphics, like texture or display lists transfer, but potentially painful for small transfers like GPU computation operant fetch. So, if you have two cards, is it better then to put them onto the first and third true x16 slot, and let each of them have a dedicated x16 from the chipset - but three hops in between them over the PCIe bridges, or put them on the first two x16 slots to share data over the single Nforce 200, but have potential bandwidth contention towards the chipset?
As more boards, including the new entries from Gigabyte, start implementing the bridges, let's see what's the impact. Here we put together the Asus P6T7 WS mainboard mentioned above with the Intel Core i7 975XE processor and 6GB of the trusty Kingston HyperX DDR3-2000 memory, all running at the default clocks and settings. A CoolerMaster V8 heatsink fan handled the CPU heat, while a Corsair HX1000 PSU handled the power feed.
We ran two Asus HD5870 cards at the default speed - we will cover the voltage tweak overclocking options in a separate review. The point was to see what is the best choice for the high-end users when it comes to balancing the latency and accessible bandwidth per card in a bridged multi-PCIe slot system. Keep in mind that the new dual HD5970 GPU also uses a PCIe bridge on the card, this time from PLX. Therefore, its GPUs also have to go along with some extra latency to the rest of the system. That's of course in addition to the 15 per cent clock frequency drop in that new dual GPU card.
The test used was 3DMark Vantage under Windows Vista, with the most recent ATI Catalyst 9.11 driver suite. We ran both the performance and extreme benchmarks in both cases, the first one where both cards share a single Nforce 200 bridge by using the first two slots, and the second one where each one has its own bridge. All other configuration points were identical.
Case 1 - attachment on the first and the second full x16 PCIe slots, sharing a single Nforce 200 bridge
Performance
Extreme
Case 2 - attachment on the first and the third full x16 PCIe slots, one each per Nforce 200 bridge
Performance
Extreme
In the next round, we'll overclock these two babies using the Asus voltage tweak utility, and see how far - or how close to GHz GPU speed - they go together. µ
Share this:
Share
Ah, so you do come here just because you're desperate for attention then! Thanks for proving that for us.
As for buying Larrabee, if/when it comes out, if it's the best sure I may do so. Again, as I pointed out I'm not married to any brand like some slobbering fanboy. I guess this concept is a bit confusing for you to comprehend.
Well, you got some attention. Feel all better now?
That's a regular dayjob, now is it ?
Buy Larrabee, but not in 2010, maybe not even in 2011 or never ever. Its been postponed (again!)
Ps: my english isn't half as bad as drashtec's, and he's a real Brit as far as I could real through the techslang. The power that be @the Inq don't want me typing dutch/german/french/japanese here, now do they ?
Still an NV fanboi in sharkinfested waters ( do sharks eat trolls ? )
To Aryan: LOL. Well, I'll just say you used some interesting grammar in your post. And I also know that you could just be trolling for replies. If so, good job. I replied. I do hope your ego feels SO much better if this is the case.
Assuming you're not some troll desperate for attention, since you claim to be a Nvidia Fanboi, your loss. I am not a fanboi of any computer component brand. Makes it a lot more enjoyable to decide what to buy. Gives me a lot more options. To be blunt, why would ANYONE be a fanboi of a piece of electronics? It's a graphics card. It's not your wife or girlfriend. Or your sister. Or your mom. Or your favorite pet. The company that made it wants one thing from you. Money. But hey, if you want to attach your self worth to a hunk of electronics, enjoy. But if you ask me, that's kind of a pitiful thing to do.
Anyway, I do own two 5870's, and despite your claims the "software" side of things is fine. Only hiccup I've hit is less games support Crossfire, more games support SLI. But I also have one card in one PC, the other in my spare. When Fermi comes out, if I like it I may buy two GTX 380's (or whatever they will be called), put that in one PC, and Crossfire the two 5870's in the other. See, no fanboy philsophy here. I can enjoy BOTH major brands. If a third company started making graphic cards that were competitive with ATI & Nvidia for gaming..guess what, I'd take a look at them when I am ready to buy again and see if I feel they are a good investment.
Isn't it nice to have options, and not be married to a piece of electronics? :P
...respect.
i want a N64-2.
How scary is it when Drashek starts answering himself?....and the first comment made a little sense...but alas...it was all down hill from there :)
Hey Aryan, lean to spell or type and go buy yourself a new nvidiot card that is EXACTLY like your old one, just in a different box. Enjoy!
Why do people insist on using these 3DMark scores they are arbitrary numbers that do not translate into real world performance.
Lets see how some game performance on this beast. The least that could be done is these could be compared to two SLI cards.
Hi-Ultee' Here. what Point is: that ONE standard 5870 will scoresame as two Hd5870. If above machine & test is used on 2 crossfire hybrid, each on own cntroller with standard5870, get about 30,000 Vantage 3D.
drashek
ATi bores me. DX11, wow. DX10 didn't even arrive half baked in the DX9 dominated gamers scene, and now... liek gameing on DX11. Wow OMGWTFBBQSAUCE!!one1
The cold hard fact is that Ati didn't invest enough in the PC software side, like they have been doing for the last 2 decades or to; only to be surpassed in driver & additional software lameness by teh other one-trick-pony almighty chipzilla, ( eg: Intel believes Moore' law means: MOOOAAR CACHE ) therefore we are landed with ATi cards as one-trick-ponies strapped onto CPUs with gazillionbyte of cache, underclocked 80x86's; liek OMG teh progressss !1 . If you like the trick with the ponies as it has been the last 20 years; then gaem on. Put them ATis in Xfire all you want; but a set of one trick ponies, it still is.
An NV fanboi, has been for more then a decade. No, don't own stock.
12:59 How does one seat one minute, one single controller card that describes how to use one of the points is one of two ways. However, recent tests is difficult. One or two single standard, HD 5870 and 5870 show a 2 minute 1 fps. 5870,35 fps at HD - 70 + fps to get one one of the standard model has only been recognized as one. Permium & to reduce the number of big shots to the efforts of joy. How smart. One one one one one one the main one in front of the incredible graphics controller, one one one is used.
drashek2
Point well Explained about using one controller per card. However in recent Hard testing. HD 5870 scores 1/2 fps of standard 5870. HD getting only 35 fps while any make of standard model 5870 is 70+ fps.
Work so hard for great score, pay permium & find score cut in half. how smart is that.
NO WONDER USED MAIN WITH TWO GRAPHICS CONTROLLERS.
drashek
Well, assuming one run at for each case and setting, there appears to be a slight benefit to separating the GPUs across different bridges.
Multiple runs for each case and setting would be needed to identify if the perceived difference was statistically meaningful.