HIS IceQ II 850XT PE - PCIe vs AGP...

In a field like 3D graphics cards, totally dominated by Taiwan players, it is a nice surprise to find a Hong Kong company doing well against the giants across the Taiwan Straits. Hightech Information System Ltd (HIS) focused quite a bit on the high-end "enthusiasts" segment for ATI cards, in the same way that Gainward would have on Nvidia cards. HIS uses unique cooling systems to boost the overclocking potential of their cards beyond the usual 540 MHz GPU/1180 MHz memory clocks. The flashy ice-blue package matching the ice-blue cooling assembly completes the picture, pun intended.

It was interesting to test the otherwise identical AGP and PCI-Express flavours of their flagship IceQ II 850XT PE card - not only see the differences om how it overclocks, but compare the AGP and PCI performance in two otherwise similar PC systems.

So, the PCI Express version ran on an Intel 925XE-based board with 3.46 GHz Pentium4 EE processor on 1066 MHz FSB and 1 GB of DDR2-533 CL3-3-3-8 RAM, while the AGP version ran on an overclocked 3.45 GHz Pentium4 EE on MSI 875P Neo board, with 1 GB of DDR2-433 CL2-2-2-5 memory. Both systems ran WinXP and the newest Catalyst official drivers. In both cases, the AGP and PCI-E ran at the standard frequencies without overclocking.

Both IceQ II cards are based on the 850XT PE platform, which, until the recent arrival of Geforce 7800 series, was the fastest card around. Their cooling system, designed by Arctic Cooling, requires a two-slot configuration but it supposed to let you run close or above 600/1200 MHz configurations in many cases. The cards also have the Rage Theatre option added as standard, providing HDTV output with provided component cables. I'd like them to have a HDMI cable version, too.

I tuned the two test systems to be as close as possible to each other in actual CPU and memory performance, so that the difference would mostly be in the graphics card department. We ran both cards at the default 540/1180 speed first, then overclocked them to see how far we go... in all cases, 3DMark05 was used.

The PCI Express card overclocked OK till about 593 MHz GPU, 1220 MHz memory, while the AGP card went to 586 MHz GPU, 1214 MHz memory. For simplicity, we just showed what should be a stable long-run configuration in the PCI-Express case, with in-sync 590 MHz GPU and 1180 MHz memory so that you can also see the effects of faster GPU without memory speed change - here are the results!




As you can see, the default config differences are minimal, but PCI-Express wins by a razor-thin margin in most cases - simply because the 850XT is a native PCI-Express GPU, maybe?

These are good cards quality wise, and they can overclock to near the limits of the 850XT GPU. The AGP version is still among the fastest stuff you can get for your AGP slot, probably. I'm keen to see how HIS R520 cards will perform, using similar cooling and in the Crossfire setup... another option would be to look at a dual GPU card with R520, later R580 chips? A good cooling setup could make that possible.... but for now.


Source . TheInquirer

There is absolutely no current need for the graphics card bus to be faster than AGP 8x, where PCI-E has made the most inroads.

There IS a current need for a faster Gigabit ethernet and RAID card bus, where PCI-E has failed to make any impact yet!

(in reply to SiliconFreak)

Totally agree with You, Matthew!

But the point is...that they "forced" us (or rather tricked us - well not me, but many other novice users)...that PCI-E is some State-of-the-art superstar from the future.......and that we cant live without it...and also that AGP 8x is some pre-historical dinosaur... But the reality is...that they needed something new to lift the prices up (cause they were falling already, because of high demand and rivalry in motherboard and graphic card markets)...and what did we get (well not me, cause i am sticking with AGP 8x until I see at least 100% higher performance at PCI-E than at AGP 8x....which isnt gonna happen in 2005 or 2006...)....we get exactly the same performance in new outfits...which would be ok...if the prices would stay the same....but since thats not the case....its a big shame to trick users-buyers like this....lying to them about how "great" this PCI-E is...

I am also aware that AGP 8x will soon be limited because of its max.bandwidth which is lower that PCI-E's...but...as things stand right now...PCI-E sucks...and even more...because in i.e. SLI mode....both cards work at only x8 instead of x16....bad bad bad...we need something new fast...for the right price offcourse...

Also so called "graphic card war" gives me a headache...because they do that intentionally...for users to believe they have big conflicts between them (i.e. Nvidia and ATI)....but reality here is....that they do that intentionally...just to get our focuses out of main things....which are...performance performance performance...and all that at lowest  prices possible, offcourse.... And each time they say how much better and faster their new products will be...and what we get each time? maybe 5-10% in speed gains...nothing more....and on the other side...at what price? MUCH higher than those 5-10%....badddddddddddd....we dont need that....if they would only want...they could give us 100% or better speed(performance) increases each time they release new product....but they dont...cause they wouldnt make such big profits that way....so they rather feed us with those breadcrumbs....

Anyway....enough....i am tired already from all this...lets just hope for the best...cause sooner or later they would need to change their policies....cause what goes around...comes around....and thats 100% true!


Greetings,

SiliconFreak.

(in reply to Matthew)

A "good" accelerated graphics card does more work on the card, and less across the bus.

About the only really significant gain from PCI-E, is that it makes the "hypermemory" approach usable, where the card only carries enough memory for the frame buffer, and all the textures are across the bus - AGP could do this as well (graphics aperture / DME) but with greater penalty, so it was normally only an overflow measure.

There has been some development of a motherboard or motherboard chipset (SiS) that supports both AGP and PCI-E, without using a bridge - maybe one way of bridging the generation gap, where you have a more than adequate AGP card, but want a motherboard and CPU upgrade that doesn't lock in to the old standard.

(in reply to SiliconFreak)

Yes I know that....but....with or without Hypermemory...all that I (and many others) want at the end of the day...is to have better performance...and dont really care if the technology to deliver it is called Hypermemory,Crossfire,SLI or Donald Duck...

But like I said...so far...I havent seen any major improvements in that area (cause those 5-10% increases dont count here)....so...still waiting for "big players" to deliver that to all of us...but I guess I will be very old by then...

(in reply to Matthew)

My mistake, it was ULI not SIS for the AGP / PCI-E combo....

Hypermemory is NOT a performance solution, hypermemory is a costcutting, cheap-ass solution that places excessively high demands on bus and system memory bandwidth - a bit like having a low-memory AGP card and hoping that the "graphics aperture" of system memory will make up for it.

(in reply to SiliconFreak)

Yes I know Matthew what Hypermemory is...and thanks but no thanks...not for me...

Because look...they could offer us 1Gb of Ram, at least 100% faster GPU's and many other new technologies inside Graphic card EVEN TODAY! and all that for maybe 20-30% increase in price (and still make good profits!!!)...BUT...they rather constantly telling us how hard is it....then here are those ATI-NVIDIA so-called "wars"...etc..etc...so the average user focuses more on all that....than on what is most important....PERFORMANCE / PRICE ratio.... Well I also know that we are somehow already used to all that (to get small gains instead of 100%+ ones - because they forced us into that in all those years of playing their dirty games) and that we dont have many choices here...its either you take some of their cards or not....but since we ALL need them (they know that VERY WELL!)....we need to buy them.... And what do they do....they just quietly count $$$...make 5-10% increase with each new card they release (instead of at least 50-100% increase with each one)...and thats bad (exept for their shareholders )...because....i.e....you may buy the latest GPU on the market today....then add some water or sub-zero cooling....and easily skip next 2 generations...and still be very competitive (if not faster) with the fastest GPU at that time...orrrrrrrr....change card with each new that is released....spend at least $1500 in that time(3x$500)....and...what would you get at the end?....MAYBE 30-40% faster graphics than related to first card you bought....and for what price? At least $1500.... Thats what makes me mad!

So...at the end of the day...who needs hypermemory...when they could easily add at least 128 if not 256mb on each new card (for few extra $$$ only!)...and instead....to lower their prices...they "invented" this Hypermemory...which forces us to buy more system ram....andddd....whats most important....since we all know its much slower than the one on Graphic cards....also cant handle all data and transfers so fast (even fastest bus doesnt help here)....so its all just some marketing trick to be able to sell cards with 32 or 64Mb of ram and trying to convince us that they handle everything as fast as ordinary cards with 128,256 or 512Mb can....

Nevermind....I know I triple-check everything before buying new cards so even I am also "tricked"...at least I am not that much as many average or novice users....so good luck to all with new products (because its more or less the same with all other manufacturers and their products)....

(in reply to Matthew)