GT300 News

[Gilly]

http://www.pcgameshardware.com/aid,696189/GT300-New-rumors-about-Nvidias-DirectX-11-chip-with-codename-Fermi/News/

maybe as soon as December?

[whismerhill]

hum this seems like it would be a repeat of the past

somehow this seems the most probable option but still since the HD4870 I would have thought NVidia would slow down on doing always bigger GPUs...

overall the rumours seems to be based on extrapolations from what happened in the past

[MobsterOO7]

It would be nice if they would start downsizing on the actual size of these cards.

[Gilly]

the die on the recent cards have been HUGE. it would be nice to see how this stacks up against the 5870, my money is on ATi this time around, and plus i have recently prefered ATi drivers (due to 0 problems and monthly updates), compared to multiple WHQL releases where things have been broken for nV. But lets not turn this into "flame gilly" or driver criticism. Point is... ATi is making money now on the new cards, and nV is still putting theirs togeather

[Paradigm Shifter]

Oddly, I don't really care about the die size, or the cost of GT300 (although if it's ridiculously expensive then there is no way I'd even consider it) but the bit that has me most horrified is that power consumption figure... 300w.

For those that missed it:

300w

Unless it's literally twice as fast as the 5870... it's not worth it.

...

Bloody hell...

After ATi did so well to get the power consumption of the 5000's down, it seems nVidia are going in the opposite direction.

[Frag Maniac]

I have just 3 words, BRING. IT. ON.!!! A nuclear powered Nvidia DX11 card that wipes the floor with any single GPU card and competes even with rival dual GPU cards.

This all comes down to what you believe in this article and/or what you want to put in your system. The power consumption for a single 5870 is 188 watts. According to this article the GT300 may give even the 5870 X2 a run for it's money, which some speculate will have 340 watt consumption.

If these numbers are true, personally I'd rather have a simpler solution in a single GPU card as powerful as the GT300, vs a dual GPU 5870 X2 that will no doubt run hotter and be more likely problematic in some games (As dual GPU cards often are), even if it means slightly less FPS in most games.

(EDITED)
Another article that sheds a bit more light. I wasn't happy hearing of Fudzilla being VR's main source. Even HardOCP is spreading false rumors about release dates.

http://www.tweaktown.com/news/13215/nvidia_says_40nm_yeilds_are_fine_gt300_on_schedule_for_q4/

Excerpts:

"They (Nvidia) are claiming that the reports of less than 2% yields were based on a mistranslated report from AMD’s competitive analysis team.

Also they say that the reports that the GT300 is a rebranded/respun GT200 are baseless. According to the report the GT300 is a ground up redesign and a new GPU."

I wouldn't expect anything better from ATI's reps, they often speak total garbage, like their CS reps do to their customers.

[Paradigm Shifter]

Indeed, I'd like to see something solid too. Because until we do, we have no idea. Except we know about the Radeon 5870, because it's out. So we have a rough idea, presuming clockspeeds aren't altered, how a 5870X2 will perform. But on GT300 it's all still FUD or fanboy supposition.

From the shots of the 5870X2, it's got 8-pin and 6-pin PCI-E power connectors... that would mean 75w from the slot, 75w from the 6-pin and 100w from the 8-pin.

How anyone managed to get 340w out of 75+75+100 is beyond me.

PCGH have something too. Although it seems to be mostly repeats of elsewhere (FUD via VRZone). If GT300 can match the 5870X2 (when each are released) as a single card, then it'll be exciting times. Let's hope both the X2 and GT300 hit before Christmas. :D

edit: I vaguely remembered seeing somewhere that video cards are allowed to break PCI-E specs and draw in excess of 75w from the slot, but the standards sheets are explicit: 75w max and only. Further, from other reports, most graphics cards will draw power preferentially from the 6-pin and 8-pin connectors, rather than the slot, as the power is cleaner coming straight from the PSU rather than through the mobo traces.

edit 2: There is a paper on PCI-SIG covering a proposed PCI-E 16x 150w specification. I don't belong so can't get to it, but it seems like PCI-E 3.0 might, potentially, allow 150w per slot...

[Frag Maniac]

From what I've read even the 4870 X2 has been known to show as high as 370+ watt consumption in synthetic bench tests. I think you're oversimplifying your method of trying to determine max power consumption.

And as far as the talk of the GT300 being "all still FUD or fanboy supposition", not necessarily.

Check this out:

http://vr-zone.com/articles/-rumour-nvidia-gt300-architecture-details-revealed/7763.html?doc=7763

This would appear to verify 3 key things, 2 of which falsify earlier rumors. A 40nm process, contrary to ATI's so called report that was likely intentionally mistranslated ( I think Nvidia was being kind with those words), a non GT200 architecture, and 3 friggin billion transistors!

I say again, 3...BILLION... TRANSISTORS!

So I ask you at this point, does it really sound like the GT300 using 300 watts makes it inefficient, or that it makes it a stand alone, no SLI required freak of nature?

My only concern at this point is what speed a 920 will need to be clocked at to avoid a bottleneck. Preferably I'd like to avoid overvolting the CPU, but at this point I'm wondering if it will be necessary just to run this monstrous GPU without the system being unbalanced.

[Paradigm Shifter]

From what I've read even the 4870 X2 has been known to show as high as 370+ watt consumption in synthetic bench tests. I think you're oversimplifying your method of trying to determine max power consumption.

The only figures I can find that really hammer a 4870X2 (with Furmark) is on Overclock3D. 495w for a system about as close to my Core i7 rig as I've seen - 3.8GHz 920, 6GB Corsair, Gigabyte UD5, XFX 4870X2... I've got more HDDs though... that's from 336w idle in Windows, so that's a 160w idle/load jump. Now I know that the 4870X2 (and the 4000 series in general) isn't all that efficient at idle, but it's not horrible...

http://www.overclock3d.net/reviews.php?/gpu_displays/xfx_ati_4000_series_round_up/6

Just found some TechPowerUp data showing card power figures. 4870X2 idle is 73w, average 236w, peak is 273w and maximum is 381w, which yes, seems a little scary. I bet that was Furmark hitting all 1600 shaders for everything it could.

Compare that to the 5870 at 19w (idle), 122w (average) 144w peak, 212w (maximum)...

On paper, adding up the power provided from each possible source of power should be sufficient. However it's obvious that in specific scenarios designed to hit the cards as hard as possible, there is no limit on what they draw, short of blowing the PSU.

Does make me wonder what the maximum power draw of GT300 will be, though... :shock:

And as far as the talk of the GT300 being "all still FUD or fanboy supposition", not necessarily.

Actually, I meant about both the 5870X2 and GT300. Pics and figures are all very well, but I'll only be happy when we've got solid figures from multiple sources, rather than fifteen places all referencing one Fudzilla, Inq or BSN article... because all three of those places are good at 'imagining' stories for page hits. ;)

Check this out:

http://vr-zone.com/articles/-rumour-nvidia-gt300-architecture-details-revealed/7763.html?doc=7763

This would appear to verify 3 key things, 2 of which falsify earlier rumors. A 40nm process, contrary to ATI's so called report that was likely intentionally mistranslated ( I think Nvidia was being kind with those words), a non GT200 architecture, and 3 friggin billion transistors!

I say again, 3...BILLION... TRANSISTORS!

So I ask you at this point, does it really sound like the GT300 using 300 watts makes it inefficient, or that it makes it a stand alone, no SLI required freak of nature?

Yeah, looks sweet. You did notice the [Rumour] tag, though, yeah? ;)

Only solid info we've got to go on about performance of any of these cards is the fact that the 5870 is released. All the other cards are guesswork at best or outright lies at worst...

[Gilly]

the spec is the spec though Frag. The power consumption for the whole system might be that high, but depends on what you mean. Im sure if you work the wattage out of the card it would be no where near 370w.

The card itself should only draw max 100 + 75 + 75 = 250w.

if the card draws too much im sure that would cause problems. I can understand however... you can get more than 75w from a PCI-e cable. Im sure if you stress it a bit, you can pull a lot more from that, you can draw more current than you are allowed I am sure.

As for how much I dont know. We could also only speculate on the connectors for the GT300.

300w - 75 leaves 225w for the PCI-e slots, that would be 2x 8 pins and a molex or something. Holy crap :P

[Frag Maniac]

Still though guys, if what's being said here is true of the GT300 competing with the 5870 X2 (and with 3 billion transistors why wouldn't it?), 300 watts is very reasonable power consumption given that it's much less than what two 5870s draw.

My point on the over simplifying was that though power/wattage calculations in essence are cut and dried, testing methods to determine them are not. That is why you have all these ways of measuring audio amps. Most reputable brands use A weighted, but some do not.

Point is we don't really know if that 300 watt figure was derived via an unrealistic max wattage scenario (PSU blowing as you put it) or via what the card typically draws under normal circumstances.

I also find it hard to believe that die shot even though blurred is photo shopped or something as far as the legible specs on it.

Could you guys please respond to my last paragraph which I edited in after you responded though?

"My only concern at this point is what speed a 920 will need to be clocked at to avoid a bottleneck. Preferably I'd like to avoid overvolting the CPU, but at this point I'm wondering if it will be necessary just to run this monstrous GPU without the system being unbalanced."

So what do you think, if what's been said about the GT300 is true, do you think I could do a moderate non overvolted OC on a 920 without having a bottleneck? I'm assuming I could hit at least 3.6GHz without overvolting.

Some say 3.4GHz is fine for any dual GPU setup prior to the debut of the 5870, but I'm not sure even 3.6GHz would suffice for a single GT300 with such specs.

[Gilly]

3.6 is well within reach i think yep :D

and other thing is that at very high res and/ or AA its more GPU limited anyway

[Frag Maniac]

3.6 is well within reach i think yep :D

and other thing is that at very high res and/ or AA its more GPU limited anyway

What I keep thinking too Gill is that with DX11 allowing way more processing on the GPU like multi threading, physics, etc, there will be less need to OC the CPU to balance the system load.

I seem to recall saying on one forum that the 5870 was well timed considering even more load will be handled by the GPU with DX11, but Nvidia appear to be addressing that fact even more so.

[Paradigm Shifter]

In multi-GPU scenarios or low detail settings, CPU frequency is important. Not so much in single-GPU or extreme detail settings. And sorry, I can't see many people buying GT300 then playing in 1024x768@low. :lol: ;)

You're right, Frag, in that if that 300w draw is 'max' then it's OK. If it's something like 'average' draw, then we could look forward to it trying to draw 400w+ as it's 'maximum'. Which, frankly, would be bloody dangerous if it's through a PCI-E slot, and a 6- and 8-pin power lead. Twin 8-pin might just cope. Might. Each of those 2-pins is supposed to carry 25w, but I seem to remember somewhere doing a power draw test on a 1000w PSU and finding that each of those 2-pin PCI-E bits could provide about 40w... that's 320w right there with twin 8-pins loaded above-spec. I'll try to find it. It was a 'take a PSU to a proper test facility' rather than a 'shove it in a PC and see if it runs'... might have been Johnny Guru, would make sense. Can't find it there, though. :( Might have been on a forum. Spent way too much time surfing hardware forums in my free time in recent weeks. :oops:

...

While this is a fairly educated guess on my part, in absolute terms I hit 3.8GHz without a voltage bump on my i7 CPU (I've since tweaked that 1.275v back down to 1.25v and it's still 100% stable) but as each chip is different I'd say 3.6 was the better target.

As for what would eliminate a CPU bottleneck? Hm. I'm going to presume you'll be running 1920x1200@either 4 or 8xAA/16xAF, as there really seems little other point in getting GT300/5870 otherwise. ;) From that, I'd say that you're unlikely to encounter a CPU bottleneck with a single card. If you went dual card (for whatever reason) you might begin to encounter bottlenecking there.

However, it's also totally dependent on the games you play - as many people have pointed out, Half-Life 2 continues to show framerate improvements even as you scale a Core i7 into the 4.6GHz region. But the Source engine is probably the most CPU sensitive engine in common use now.

I dunno how DirectX 11 will take GPU works. If it puts each task onto the processor that can deliver with best efficiency, then GPU physics would probably be a good idea. As for overclocking CPU to balance system load... who knows. The CPU will still need to assemble it all, particularly if everything is threaded off neatly. DX11 might finally deliver games that actually use these multicore CPUs properly! :)

[skeeder]

my take on this is...well, we've seen this before. ATI has just launched to 9800pro. The 4870 was the 9600pro.

Now NVIDIA is claiming that the 5950U will kill it.

Any one of us knows that when NVIDIA is behind, they are not in their A game. The 8000 series was fantastic...why? the 7000 series was incredible. Just something I've noticed...they don't like losing, and they tend to cut corners to win.

[Paradigm Shifter]

Hm. Interestingly, I've not seen a great deal of bragging from nVidia about this launch.

What I have seen is them rubbishing the competition (standard practice now, unfortunately) and downplaying how much use DirectX 11 will be... (I suspect once they have DX11 hardware out their tune will suddenly change...) I've not seen leaked benchmarks or NDAs being 'broken'.

Which says one of two things to me; either GT300 needs more time to mature on the process they're using, and they're scrabbling to get clockspeeds up high enough to compete (like ATi did with the X1800 cards and the soft grounding issue they had which delayed it by a good eight months...) or it is very good, but they learned that mouthing off about "opening cans of whoop-ass" doesn't work when the competitors products are significantly better than anticipated. Because you're not telling me that nVidia thought the 4000 series was going to be as good as it was, or they wouldn't have done that quick about-face on pricing of the GTX260/280. A $200 price drop is a serious market miscalculation.

[Gilly]

http://www.tomshardware.co.uk/radeon-hd-5850,review-31692-14.html

interesting benches, the 2nd one in particular.

shows that even @ 4x AA and 8x AA @ 1920x1200 the CPU does play a role if its got a hefty OC.

but only on some games. And this may change in the near future with the advent of the DX11 game

[Paradigm Shifter]

Three issues I have with that data:

1) It's Tom's Hardware.
2) Left4Dead is Source engine based and therefore known to favour increasing CPU speeds.
3) Other differences are within statistical margins of error.

I'd be saying the same no matter whose CPU/GPU was on the bench - show me differences in the 5%+ region and I'll say they're significant. But show me 1fps difference and I'm just not gonna call it significant when FPS are >30, sorry. That said, there is a 6% difference with RE5, too, which is just entering into significant territory. But Capcom's last game (DMC4) gobbled CPU cycles for increased framerate as well, so again it's not totally unexpected.

But most of those games show <3% difference, which for a 36.5% clock boost is... a poor indication that they're CPU bound, sorry.

[Frag Maniac]

Not to mention that where differences are shown the frame rates are way higher than you'd notice in game anyway. Not really worth it IMO to ramp up the CPU higher than you need to just for bragging rights or some misconception over bottleneck worries. It's gonna run smoother, cooler and longer at a comfy 3.6GHz.

[whismerhill]

shows that even @ 4x AA and 8x AA @ 1920x1200 the CPU does play a role if its got a hefty OC.

Agreed with Paradigm Shifter & Frag Maniac
TomsHardware isn't what it used to be...

The CPU can play a role in some conditions while gaming but this doesn't show the right conditions
(games that require a lot of CPU power: X3 in big fights and/or with AI mods, GTA4 :roll: , Supreme Commander, a good % of strategy titles like Civ4 when pushed to their limits and so on ...)