Tech Report: Nvidia, Asus put the clamps on GTX 590 voltage

[AtenRa]

This is throttling due to GPU temps. Not at all the same thing nVidia is doing to the 590, which is what I said. I can't remember there ever being a card that has had the same issues as the 590. [sarc] Of course there is nothing wrong with the 590 it's all user error and a giant conspiracy [/sarc].

Unfortunately, if you want the type of performance that the 590 had the potential to offer, you are going to have to fork over a grand and use 4x slots on your mobo. You are going to have to buy 2x 580's. End of story, because this card is EPIC FAIL.

This is Throttling due to VRM temps

And what NV is doing with the GTX590 ?? plz enlighten us.
And what issues the GTX590 has ?? plz enlighten us again.

And dont say about an anemic VRM implementation (see my above post)

 

[3DVagabond]

This is Throttling due to VRM temps

And what NV is doing with the GTX590 ?? plz enlighten us.
And what issues the GTX590 has ?? plz enlighten us again.

And dont say about an anemic VRM implementation (see my above post)

Well, I guess it depends on the driver du jour on exactly what they are doing. Just minor overvolting though and the card throttles. They are using the drivers to OCP and save the card from blowing up.

There's nothing wrong with OCP. I've owned equipment that didn't need it (Krell reference series amplifiers driving Apogees.). At least not any sort of active circuits. It was terribly, terribly expensive. I don't want my video card, or anything in my PC to be built that way. (Well, I might except the PSU.). But face it, the 590 is fragile. It's weakly built. They cut it back to get it price competitive with what AMD could offer. They just couldn't build a substantially enough built component to compete though. This card is worse than the 480 and a major step back from the 580.

 

[Silverforce11]

AtenRa: Yeah there's nothing wrong with the 590 VRMs.. NV just decides to ban all vcore mods with recent drivers for no reason.

 

[notty22]

All of this fan analyzing of the gtx 590 is to distract from the fact, it competes for the fastest card in the world. Hocp, declared there is no chance of failure from normal use. I see 2 highly clocked Caymans and the supposed better AMD scaling barely keep up with the 600mhz gtx 590. Proving even more the gtx 580/Fermi superiority.

This overclock did improve performance for us, and is a safe bet for overclocking the GTX 590 without having to worry about lifespan and explosions. And the fabled "explosions" should not even be a remote concern if you are using the latest driver software.

 

[Keysplayr]

All of this fan analyzing of the gtx 590 is to distract from the fact, it competes for the fastest card in the world. Hocp, declared there is no chance of failure from normal use. I see 2 highly clocked Caymans and the supposed better AMD scaling barely keep up with the 600mhz gtx 590. Proving even more the gtx 580/Fermi superiority.

Nice. I like it. :thumbsup:

 

[Morg.]

All of this fan analyzing of the gtx 590 is to distract from the fact, it competes for the fastest card in the world. Hocp, declared there is no chance of failure from normal use. I see 2 highly clocked Caymans and the supposed better AMD scaling barely keep up with the 600mhz gtx 590. Proving even more the gtx 580/Fermi superiority.

-- errr I have trouble relating what you say to what you link.. your link says stock 6990 > stock 590 and yet you mention caymans in trouble /

The gtx580 fermi GPU IS better than a Cayman in terms of raw power per chip.
On the power side ... I can't get anything clear - officially they have the same TDP but most benchmarks show the 580 consuming much more -- meh.

As they're both in 40nm, comparison is quite simple :

GF110 (580)
520mm²
240 Watts (guru3d)

Cayman (6970)
389 mm²
207 watts (guru3d)

Ooh interesting .. and the Fermi does NOT have 34% advantage over the Cayman in terms of raw power while it does cost 34% more in raw materials.

On the other hand it consumes only 20% more power and it does not have a 20% advantage either.

So its relatively bigger, consumes a bit less in proportion and is less efficient overall.

Fermi rocks . but this ain't it yet, it's only rev 2.0 and it's already not bad at all - gotta remember it's far from being finished though.

 

[kevinsbane]

All of this fan analyzing of the gtx 590 is to distract from the fact, it competes for the fastest card in the world. Hocp, declared there is no chance of failure from normal use. I see 2 highly clocked Caymans and the supposed better AMD scaling barely keep up with the 600mhz gtx 590. Proving even more the gtx 580/Fermi superiority.

It can be put this way too:

All this fan analyzing of the GTX 590 is to distract from the fact that the 6990 is the fastest card in the world. Hocp declared that there should be no worries of burnout with the GTX 590 from normal use. I see 2 downclocked Caymans and the historically inferior AMD scaling keep up with and surpass an architecturally superior Fermi based card. Proving even more the shortcomings of the GTX590.

Not that I agree with how I said that, but it would be as valid of a presentation of facts as your statement, Notty.

 

[Lonbjerg]

-- errr I have trouble relating what you say to what you link.. your link says stock 6990 > stock 590 and yet you mention caymans in trouble /

The gtx580 fermi GPU IS better than a Cayman in terms of raw power per chip.
On the power side ... I can't get anything clear - officially they have the same TDP but most benchmarks show the 580 consuming much more -- meh.

As they're both in 40nm, comparison is quite simple :

GF110 (580)
520mm²
240 Watts (guru3d)

Cayman (6970)
389 mm²
207 watts (guru3d)

Ooh interesting .. and the Fermi does NOT have 34% advantage over the Cayman in terms of raw power while it does cost 34% more in raw materials.

On the other hand it consumes only 20% more power and it does not have a 20% advantage either.

So its relatively bigger, consumes a bit less in proportion and is less efficient overall.

Fermi rocks . but this ain't it yet, it's only rev 2.0 and it's already not bad at all - gotta remember it's far from being finished though.

You are talking only about "relative" performance in games, excluding tesselation, PhysX, CUDA ect...right?

 

[PingviN]

All of this fan analyzing of the gtx 590 is to distract from the fact, it competes for the fastest card in the world. Hocp, declared there is no chance of failure from normal use. I see 2 highly clocked Caymans and the supposed better AMD scaling barely keep up with the 600mhz gtx 590. Proving even more the gtx 580/Fermi superiority.

What superiority? The GF110 chip is bigger (way bigger) than the Cayman chip and the high power draw for such a huge chip (talking 'bout GF110) wont allow higher clocks. Antilles offers higher performance at a lower power consumption and with much smaller chip. This if anything proves the inferiority of the GTX590.

2 huge dies beaten by 2 smaller dies while at the same time using more power.

 

[Morg.]

You are talking only about "relative" performance in games, excluding tesselation, PhysX, CUDA ect...right?

Exactly. If I had more benchmarks .. who knows.
And I'm quite sure the Fermi is a zillion times better than the cayman in CUDA or other GPU-accelerated processing, also nobody cares about those benchmarks in a gaming card.

 

[SirPauly]

I dont even know why you guys are even debating the merits of the 590 reference PCB. It's woefully inadequate. It has 10x 35A VRMs at 80% efficiency for both GPUs. At its stock vcore, both GPUs have ~300W max b4 the VRMs get into stress zone. It's clear why NV has released drivers to prevent vcore manipulation as it will shorten the lifespan of these components to run well beyond specs (they are already running beyond specs at stock) when OC further.

The 6990 reference PCB is over-engineered. It's overpowered. For enthusiasts who decide to go water cooling, you can overvolt and get ridiculous OC to 1.1ghz. Even on air it can reach 1ghz (albeit noisy, but its stable and won't explode). Each GPU on the 6990 has ~ the same VRM capabilities as both GPUs on the 590. 4x 80A Volterra VRMs each GPU.

As for the 590 being similar to the 6990 in performance. No. Minus 1080p res, minus older games like COD, ME or Hawx where both cards get 200+ fps, the 6990 is >10% faster at stock. You can get reviews to skew the results how you like, but ultimately the performance in newer dx11 demanding games is all that matters when these cards are $700. Who really cares about a $700 GPU doing 250 fps vs 300 fps in old games?? Even in the event of SSAA in ME2, once its enabled, the huge fps lead the 590 has becomes insignificant.

I am looking forward to the next steam update and see how well either cards have penetrated the market.

When does the protection mechanisms come into play with such impressive clockings?

 

[Lonbjerg]

Exactly. If I had more benchmarks .. who knows.
And I'm quite sure the Fermi is a zillion times better than the cayman in CUDA or other GPU-accelerated processing, also nobody cares about those benchmarks in a gaming card.

DirectCompute is used in Civ5.
CUDA is used in Just Cause 2.
PhysX/APEX in more games.

Your claim is false.

 

[AtenRa]

Well, I guess it depends on the driver du jour on exactly what they are doing. Just minor overvolting though and the card throttles. They are using the drivers to OCP and save the card from blowing up.

Again your luck of information and knowledge continues with misinformation,

Both Hard OCP [H] and LAB501 have Overclocked the cards at high frequencies without throttling, i suggest you first have a look at data before you talk fud.

HardOCP 732MHz Reference cooler 0.963V no throttling
http://www.hardocp.com/article/2011.../asus_geforce_gtx_590_overclocking_followup/2

LAB501 upto 900MHz (non reference cooling)
http://lab501.ro/placi-video/nvidia-geforce-gtx-590-studiu-de-overclocking

AtenRa: Yeah there's nothing wrong with the 590 VRMs.. NV just decides to ban all vcore mods with recent drivers for no reason.

The driver you are talking about (270,51) is a BETA, the official NVIDIA driver for the GTX590 is at this moment the 267,91 which allows OV.

 

[Keysplayr]

Again your luck of information and knowledge continues with misinformation,

Both Hard OCP [H] and LAB501 have Overclocked the cards at high frequencies without throttling, i suggest you first have a look at data before you talk fud.

HardOCP 732MHz Reference cooler 0.963V no throttling
http://www.hardocp.com/article/2011.../asus_geforce_gtx_590_overclocking_followup/2

LAB501 upto 900MHz (non reference cooling)
http://lab501.ro/placi-video/nvidia-geforce-gtx-590-studiu-de-overclocking




The driver you are talking about (270,51) is a BETA, the official NVIDIA driver for the GTX590 is at this moment the 267,91 which allows OV.

Begging your pardon here, while it is the latest certified driver for GTX590, I do believe the 267.91's are where the over voltage was disabled.

 

[AtenRa]

Begging your pardon here, while it is the latest certified driver for GTX590, I do believe the 267.91's are where the over voltage was disabled.

hmm are you sure ?? i believe only the 270,51 does that. plz advice if you can thx

 

[Keysplayr]

hmm are you sure ?? i believe only the 270,51 does that. plz advice if you can thx

Sure np.

Goes like this:

267.71 (Vista and Win7 Support GTX590)
267.85 (Added XP Support GTX590)
267.91 (Disables overvoltage)

I'm pretty certain that is the case as I remember a discussion I had with NV engineers. 270.51 continues what the 267.91's started, but also offered game bug fixes such as Dragon Age II and various improvements in SLI setups in many games.

 

[Keysplayr]

I agree with you, I don't see Fermi's superiority either here as it looses against the HD 6990 at stock settings most of the time according to HardOCP link that notty22 posted. He said a draw when the HD 6990 beat the GTX 590 in 5 of 8 games.

Also PhysX and CUDA is a moot point as they had lost so much relevance lately due to the OpenCL hype and the fact that in the last 6 months, only a couple of games had been released using CPU accelerated PhysX and besides of a Badaboom update, nothing new in the GPGPU front had been released, a pity, as GPGPU has promie.

While the HD 6970 is still behind in the GPGPU front in average, some apps like MilkyWay@Home and SmallLux shows that the gap between the Radeon and geForce got much smaller and is a step into the right direction, it even support Command and Queue Out of Order Execution that allows parallel kernel executions from different threads. AMD showed being a bit more serious in the GPGPU arena since the HD 5x00 series launch and above.

It seems to me that GTX 590 in overall lost this round by a nose unless if someone releases a Non-reference GTX 590 which it might be feasible ala ROG, with powerful VRM, water cooling solution and probably an independent power brick. (Mini-nuclear reactor, just kidding, :biggrin: )

Problem is, HOCP isn't the only review site out there. Take them all in and see what the average brings. I bet you'll come up with trading blows more often than any of this superiority BS talk.

 

[AtenRa]

Sure np.

Goes like this:

267.71 (Vista and Win7 Support GTX590)
267.85 (Added XP Support GTX590)
267.91 (Disables overvoltage)

I'm pretty certain that is the case as I remember a discussion I had with NV engineers. 270.51 continues what the 267.91's started, but also offered game bug fixes such as Dragon Age II and various improvements in SLI setups in many games.

OK thx, so 267,71 is the best driver untill now with OCP and OV suport but not found in NV site

 

[SirPauly]

The HD 6990 is a compelling choice, with some strengths and advantages -- great to see. And the GTX 590 competes with it very well, with some strengths and advantages as well.

They both offer much more performance than a GTX 580.

I just don't understand how anyone can form a view of dud, epic fail and turd if they read all the reviews. I certainly can understand someone forming this view if they cherry pick.

What I think is unfair in a way, is maybe the expectation was to have GTX 580 Sli performance at this price-point, based on its a full core. A victim of the success of what the GTX 580 could do I guess.

 

[SlowSpyder]

notty22, no one is taking anything away from the performance of the GTX590... it's a very fast card. I'd say it's very near a 6990, depending on the game and resolution it can be faster or slower.

This thread is about Nvidia locking down the voltage on the GTX590 because they kept failing when users overvolted/overclocked them. It appears they ran the components on the PCB to the ragged edge in an attempt to keep up with the 6990. Which, on the other hand, the 6990 is the same price, offers very similar performance, and appears to be a more stoutly built piece of hardware.

I just can't see getting a GTX590 over a 6990, at least not when talking about reference cards.

 

[Skurge]

It is faster there, but I DOUBT the 6990 even at 972 core would be pulling as much power as 580SLI. It would be literally impossible to cool it on the stock cooler.

They don't show temps though.

 

[thilanliyan]

They don't show temps though.

Wow...I stand corrected.

 

[notty22]

-- errr I have trouble relating what you say to what you link.. your link says stock 6990 > stock 590 and yet you mention caymans in trouble /

The gtx580 fermi GPU IS better than a Cayman in terms of raw power per chip.
On the power side ... I can't get anything clear - officially they have the same TDP but most benchmarks show the 580 consuming much more -- meh.

As they're both in 40nm, comparison is quite simple :

GF110 (580)
520mm²
240 Watts (guru3d)

Cayman (6970)
389 mm²
207 watts (guru3d)

Ooh interesting .. and the Fermi does NOT have 34% advantage over the Cayman in terms of raw power while it does cost 34% more in raw materials.

On the other hand it consumes only 20% more power and it does not have a 20% advantage either.

So its relatively bigger, consumes a bit less in proportion and is less efficient overall.

Fermi rocks . but this ain't it yet, it's only rev 2.0 and it's already not bad at all - gotta remember it's far from being finished though.

The old mm comparison again. Its useless. Your proposing that AMD could have made a all more powerful gpu if they had made a big a die as GF110. The problem is they couldn't. It underperformed for its die size if you compare it within AMD's own lineup, to the redesigned Barts core.

That got smaller than Cypress

6870 255mm²
6970 389 mm²

The performance of Barts had some assuming Cayman would have the same performance to mm efficiency. Its not that easy it seems.

 

[wahdangun]

Problem is, HOCP isn't the only review site out there. Take them all in and see what the average brings. I bet you'll come up with trading blows more often than any of this superiority BS talk.

but the fact is notty use HOCP benchmark as a tool to show fermi superiority, maybe he mean a superiority in drawing more power? Lol

 

[badb0y]

Cayman was designed for 32 nm so that's probably why it wasn't as fast as expected. It's like the 2900 XT, it sucked when released but it's evolutionary successors were pretty good.