Poor performance with new 4890?

[myocardia]

crackshot, it's definitely your PSU. What you describe happening is what always happens with an overloaded, cheap PSU. They shut completely down, with no warning.

 

[GaiaHunter]

That's frilling wrong. The toxic version only has an 8-pin and a 6-pin which is limited by spec to 12.5a (150w) and 6.25a (75w) respectively. A pci-e can supply 6.25a (75w) as well. There is no way to get to 26.6a (320w). Max supply value is 25a 300w. A 4890 is never going to use that much, maybe near 200w OCCT SC3, 225w overclocked.

Sincerely I don't know if those numbers are video card only or system.

On the other hand, a 5970 is also a 8-pin+6-pin. You telling me an overclocked 5970 can't do over 300W?

http://www.hardocp.com/article/2009/11/18/amd_ati_radeon_hd_5970_video_card_review/7

 

[Schmide]

Sincerely I don't know if those numbers are video card only or system.

On the other hand, a 5970 is also a 8-pin+6-pin. You telling me an overclocked 5970 can't do over 300W?

http://www.hardocp.com/article/2009/11/18/amd_ati_radeon_hd_5970_video_card_review/7

A 5890 will do near 300w, they specifically limited the clocks to keep it under the ATX 300w limit. I believe gaming wattage was around 190w, though if the VRMs could handle it, it should OCCT at around 300w as a 5870 draws 160w there.

Edit: Do you realize their numbers (link) are total system wattage?

 

[GaiaHunter]

A 5890 will do near 300w, they specifically limited the clocks to keep it under the ATX 300w limit. I believe gaming wattage was around 190w, though if the VRMs could handle it, it should OCCT at around 300w as a 5870 draws 160w there.

But an overclocked 5970 sucks more on the same 8-pin+6-pin, so the system is capable of delivering over 300W by PCI-E x16 +8-pin+6-pin.

4890 Sapphire Toxic reviews:

http://www.guru3d.com/article/sapphire-toxic-hd-4890-vaporx-review/6

http://www.techpowerup.com/reviews/Sapphire/HD_4890_Toxic_Vapor-X/26.html

 

[Schmide]

Actual numbers that mean something

ht4u.net Power consumption of current graphics cards Specifically Attachment: Radeon HD 4870 1024 MB

 

[Schmide]

But an overclocked 5970 sucks more on the same 8-pin+6-pin, so the system is capable of delivering over 300W by PCI-E x16 +8-pin+6-pin.

4890 Sapphire Toxic reviews:

http://www.guru3d.com/article/sapphire-toxic-hd-4890-vaporx-review/6

http://www.techpowerup.com/reviews/Sapphire/HD_4890_Toxic_Vapor-X/26.html

Are you seriously arguing that a 4890 will reach 300w?

Typical

Idle 60w 5a
Load Gaming 130-140w 12a
OCCT Shader Complexity 3 190-220w 17a

Edit: added the amp figures.

 

[Painman]

Yup. Sounds like you tripped the OCP circuit in that PSU. A bit of a mea culpa here, because I recall thinking that the current PSU would be enough with its combined 36A. I guess XClio aren't the decent budget PSUs they used to be.

Well, don't mess around on the replacement... get something name-brand and solid. The Corsair650 is on sale again. That's a rebate price, but don't worry. Corsair does follow through on their rebates, eventually. That free shipping is of value too. That PSU is heavy (I have one).

After you've fixed that issue, I'd *still* consider a reformat... a new video card deserves a fresh start to be at its best. I don't do this if I'm staying with the same GPU maker but I do if I'm switching.

Edit: Side note, max TDP of a PCI-E vid card is 375 watts. Don't forget the 6.25 amps that can be drawn through the PCI-E slot. Not that any production GPU in the world is actually spec'd for this (yet).

 

[GaiaHunter]

.

Edit: Do you realize their numbers (link) are total system wattage?

I do.

I also read Total system Wattage without video card =165 Watts.

Overclocked total system wattage 540W. 540W-165W = 375W

375W is over 300W but you are telling me that a PCI-E 16x + 8-pin + 6-pin can't possible provide more than 300W?

Anyway - regular 4890 review http://www.guru3d.com/article/radeon-hd-4890-review-test/5 - Dude recommends 16A per 4890, 32A or so for system.

If the OP 4890 sucks 16A on its own and the OP PSU is around 24A, that means the rest of the system can only suck 8A (96W) from the 12V rail. Add to that age of PSU and it is completely feasible to be a faulty PSU.

 

[Schmide]

I do.

I also read Total system Wattage without video card =165 Watts.

Overclocked total system wattage 540W. 540W-165W = 375W

At the wall! ~80% efficiency

432w (540W)- 132w (165W) = 300w (375W) Ironically exactly.

375W is over 300W but you are telling me that a PCI-E 16x + 8-pin + 6-pin can't possible provide more than 300W?

Anyway - regular 4890 review http://www.guru3d.com/article/radeon-hd-4890-review-test/5 - Dude recommends 16A per 4890, 32A or so for system.

If the OP 4890 sucks 16A on its own and the OP PSU is around 24A, that means the rest of the system can only suck 8A (96W) from the 12V rail. Add to that age of PSU and it is completely feasible to be a faulty PSU.

The original poster is overclocking on I believe a 6-phase board. I would look for some discoloration around the CPU VRMs. I doubt a new PSU will change whatever is going on with his system. I've been wrong before, but his power footprint (gaming) just isn't high enough to not look elsewhere first!

Edit: You can't subtract the limitation on the 3v and 5v to determine the 12v. Each rail is designed separate and regulated so. Although on some PSUs they have one shared 12v rail with two regulators.

 

[GaiaHunter]

Are you seriously arguing that a 4890 will reach 300w?

No.

I'm saying that a PCI-E 16x lane, a 8-pin and 6-pin can provide more than 300W.

If you have a problem with the 300W figure for the 4890 toxic, talk with the poster of that thread.


Clear?

Now - I'm just trying to help the OP as I did in this thread http://forums.anandtech.com/showthread.php?t=2027324&highlight= in this post http://forums.anandtech.com/showpost.php?p=28970518&postcount=14.

Edit: You can't subtract the limitation on the 3v and 5v to determine the 12v. Each rail is designed separate and regulated so. Although on some PSUs they have one shared 12v rail with two regulators.

I can do that to do rough estimate. How many Ampere do you think that PSU will provide considering the max wattage for the 3V and the 5V rail is 210W?

This will give me the minimum amperage.

Additionally you can't discard age of the PSU.

This guy had his system with a 8800GT. No problems.

He replace his video card, system freezes and crash. Performance is inconsistent.

He wipes the nvidia and ati drivers - still problems.

Yeah, he can format and he can try that card on another PC - all stuff that I told him to do.

And your big interest on this thread is because I linked something, that doesn't state if it is video card draw or total system draw, that says a 4890 toxic uses 26A?

 

[vj8usa]

Furthermore, I have never heard of a case where a stronger power supply would increase framerate. If power supply was the issue, then shouldn't we expect more crashes under normal stock conditions?

I actually encountered this with my old 450W Smartpower2 PSU when I tried to run my 4850 on it - my framerates sucked in everything, and were worse than the X1900 I was replacing. Nothing ever crashed though, and every game ran properly albeit slowly. As soon as I upgraded my PSU, the 4850 began performing as expected.

 

[Schmide]

Now, saying it can't consume more than 300W because it is impossible for a PCI-E 16x + 8-pin+ 6-pin, to provide over 300W, then you aren't proving anything because it can provide over 300W.

Not by ATX standards. If a company produced a card that at stock values draws more, I think they would be in trouble with some regulation.

Now - I'm just trying to help the OP as I did in this thread http://forums.anandtech.com/showthread.php?t=2027324&highlight= in this post http://forums.anandtech.com/showpost.php?p=28970518&postcount=14.

You're making extreme assumptions and doing some calculations that in my opinion are a bit off.

The Edit:

If you have a problem with the 300W figure for the 4890 toxic, talk with the poster of that thread.

You injected those numbers into this thread to back up your assumptions. I took exception to that and showed better tested values that show otherwise.

 

[GaiaHunter]

You injected those numbers into this thread to back up your assumptions. I took exception to that and showed better tested values that show otherwise.

Lets see:

Things for you to do:

- Run another game and see if you get better performance;
- Reinstall Windows;
- Try to to see if you can either borrow a better PSU or use the 4890 on another PC with a beefier PSU.

To OP: While I think your problem probably is the PSU, since you are in a tight budget, I would try to at least to format drive and re-install windows as this poster is saying.

That OCZ is ~42A and should be more than enough.

Any other option you want to add?

- Run everything at stock;
- Disconnect drives;
- Take ram modules off.

I just gave him all the steps and provided additional information not only for him but for other posters concerning PSU data and GPU amperage data, both not that easy to come by.

You gave me data of a 4870 for example, guess finding amperage data isn't easy.

EDIT: Actually more consumption data for those interested http://benchmarkreviews.com/index.p...k=view&id=403&Itemid=47&limit=1&limitstart=14 .

 

[Schmide]

Lets see:
Any other option you want to add?

- Run everything at stock;
- Disconnect drives;
- Take ram modules off.

Switch the leads to the video card. Although his CPU isn't that hungry, 50w (4a) sharing that with half the video card may cause some drop on his CPU overclock.

Lets see:
You gave me data of a 4870 for example, guess finding amperage data isn't easy.

Do you really think the values of a 4870 are that far from a 4890? There are other sources that back up the other numbers, finding an upper bound using synthetic benchmarks that can cause the VRMs to regulate isn't exactly commonplace. Most numbers you find will equate to the 130-140w number.

The OP is really only jumping 50w (4A) from his 9800gt.

 

[Schmide]

EDIT: Actually more consumption data for those interested http://benchmarkreviews.com/index.p...k=view&id=403&Itemid=47&limit=1&limitstart=14 .

All those numbers are at the wall which is 25% more than the actual numbers!

 

[GaiaHunter]

Do you really think the values of a 4870 are that far from a 4890?

A 4890 consumes more.

But, again, I'm not interested in knowing if the Angels have sex or not.

I'm interested in helping the OP.

Are you completely certain his PSU isn't faulty?

I'm not.

Am I completely sure that the problem isn't with is drivers?

When he formats his PC and reinstall Windows I'll be sure it isn't due to drivers.

If he takes his 4890 to a different PC running another 4890, I'll be sure it isn't the video card that is faulty.

When he attaches a different PSU that is working fine and has enough power to his system and if it stuff keeps shutting, then I'll know it isn't the PSU.

Now, I wont discard that the problem is the PSU because ATX standards don't let graphic cards suck 300W.

Again and for the last time - if you think that those 300W+ for the 4890 toxic are bogus you go and say so to the author of the thread.

Even if it is 18-20A instead of 26A, I really don't care and won't change the fact that his PSU can be faulty or not providing enough juicy. A rough estimate says that PSU is around 24A at max peak. Even if it is 28A-30A it can still be faulty or not be delivering enough amperage.

Even a excellent brand PSU with 70A can be faulty and not provide enough juice to the 4890 - hardware breaks.

I know if it was me I wouldn't have risked plug a 4890 into my system with that PSU, that probably is a couple of years old, in the first place.

 

[GaiaHunter]

All those numbers are at the wall which is 25% more than the actual numbers!

So, the 4870 number you provided, http://ht4u.net/reviews/2009/power_consumption_graphics/index29.php, is meaningful - 187W.

But the number I provide, http://benchmarkreviews.com/index.p...k=view&id=403&Itemid=47&limit=1&limitstart=14, 166W - sucks, because it is at the wall?

Actual numbers that mean something

ht4u.net Power consumption of current graphics cards Specifically Attachment: Radeon HD 4870 1024 MB

So, do your numbers mean something or are your numbers at the wall? If they are at the wall, why did you post them, after accusing the numbers I linked to of being not meaningful? If both aren't at the wall, can you please stop screaming At the wall?

 

[Schmide]

A 4890 consumes more.

O really? If you survey 4870s vs 4890s they trade places by around 10w and mostly sit around the 130w figure. The 4890 has a much better idle 40w vs 65w.

But, again, I'm not interested in knowing if the Angels have sex or not.

I'm interested in helping the OP.

Then get your numbers and figures right. Exaggeration helps no one.

Are you completely certain his PSU isn't faulty?

Nope. It could very well be a power issue.

Am I completely sure that the problem isn't with is drivers?

Now, I wont discard that the problem is the PSU because ATX standards don't let graphic cards suck 300W.

I won't either. I'm giving realistic figures backed up by good sources.

Edit snip: A bunch of figure on a post won't supply my figures either. <- I'm getting tired. I have no idea what I was trying to say. Ha.

Again and for the last time - if you think that those 300W+ for the 4890 toxic are bogus you go and say so to the author of the thread.

I'm not on a never ending quest to fix all power values on the internet. Just give the correct ones when they show up on this forum.

Even if it is 18-20A instead of 26A, I really don't care and won't change the fact that his PSU can be faulty or not providing enough juicy. A rough estimate says that PSU is around 24A at max peak. Even if it is 28A-30A it can still be faulty or not be delivering enough amperage.

Even a excellent brand PSU with 70A can be faulty and not provide enough juice to the 4890 - hardware breaks.

I know if it was me I wouldn't have risked plug a 4890 into my system with that PSU, that probably is a couple of years old, in the first place.

That's fine. My 12v upper bound power opinion on his system, game loaded.

50w-60w C2D e6750 @ 3.2
40w p35
140w 4890
5w ram
25w hd dvd fans

270w 22.5a

Enough to investigate elsewhere.

But yes it could be his PSU.

On a side note: I would like to see him run OCCT for like 20min on his CPU just to eliminate it.

 

[Schmide]

So, the 4870 number you provided, http://ht4u.net/reviews/2009/power_consumption_graphics/index29.php, is meaningful - 187W.

But the number I provide, http://benchmarkreviews.com/index.p...k=view&id=403&Itemid=47&limit=1&limitstart=14, 166W - sucks, because it is at the wall?

I don't exactly remember that number, but actually it's a bit low. Furmark on a 4890 should be a bit more than 133w (166w wall) The VRMs there may be throttling.

So, do your numbers mean something or are your numbers at the wall? If they are at the wall, why did you post them, after accusing the numbers I linked to of being not meaningful? If both aren't at the wall, can you please stop screaming At the wall?

FFS those are direct measurement under a synthetic benchmark which is known to give cards an artificial high number. 187w on board would be at the wall that 233w. I'm not screaming, but you can't equate at the wall with actual on board power.

Maybe this will help

xbitlabs: PC Power Consumption: How Many Watts Do We Need?

 

[cusideabelincoln]

As others have stated, your power supply is most likely the problem as to why your computer shut down.

However let's examine the performance problem: You said you weren't getting any better performance over your 8800GT. My initial thought would have been to check V-sync. But then you said you overclocked the card and it finally performed. What this tells me is that the card was not increasing its clockspeeds beforehand, and thus why it was performing a bit worse than your 8800GT.

Now why wouldn't it raise it's clockspeed? Well, the answer is probably the power supply.

I hope the ModXStream Pro solves your problem. Be sure to tell us if it does or doesn't.

 

[shangshang]

hey OP, be sure to report back here what you found after putting in the new power supply (and don't format your drive yet). I'm curious to know myself if this was indeed your power supply. Unlike many in here, I dont think it is. I have not read any specifications on any 4890 that would "throttle back" the framerate because somehow your power supply cannnot supply enough juice. There is no such throttling. As far as I know, the 4890 clock speeds are invoked in discreet units, eg., 240 Mhz at 2D, 700 Mhz at 3D. It's either 240 Mhz or 700 Mhz. It does not work at 500 Mhz because it somehow senses that your power supply is not able to give it enough juice.

And those who say that if the power supply is weak, then the 4890 will "slow down" are essentially arguing that the 4890 has the ability to throttle back to the point that it is not capalbe of doing. Does my reasoning sound right here?

If the power supply is the issue, then the 4890 should still attempt to work at the programmed 700 MHz in 3D, only that it will crash (or we should expect it to) much more often. But the OP does not experience any crashing under stock Mhz conditions. He only experiences crashes under overclocked conditions (which may not necessary be a power supply issue alone as the video card is not operating out of specs.).

So I still say it's probably Windows that's causing the slowness. And it's not beyond reach to blame Windows for many freezing and slowness issues, right?

So OP, please report back so we all can know.

 

[shangshang]

As others have stated, your power supply is most likely the problem as to why your computer shut down.

However let's examine the performance problem: You said you weren't getting any better performance over your 8800GT. My initial thought would have been to check V-sync. But then you said you overclocked the card and it finally performed. What this tells me is that the card was not increasing its clockspeeds beforehand, and thus why it was performing a bit worse than your 8800GT.

Now why wouldn't it raise it's clockspeed? Well, the answer is probably the power supply.

I hope the ModXStream Pro solves your problem. Be sure to tell us if it does or doesn't.

1) his computer shuts down because he overclocked his video card. When his vid card is running over specs, then shutting down is not out of norm.

2) you are essentially saying that his 4890 has the ability to throttle its Mhz down when it thinks the power supply can't give it enough juice. I don't believe this is how the 4890 works. Furthermore, assumes that the 4890 can throttle down the Mhz due to a weak power supply, then why would the 4890 attemp to overclock the Mhz even if the user overclock it? Shouldn't the 4890 hava prevented all overclocking too because of inadequate power juice?

3) you think that the 4890 is slow because it is not increasing its speed due to a weak power source, but how do you know? it's just a guess on your part, a guess that I have explained would make no sense considering how the 4890 operate at discreet speeds. It's an all or nothing event. There is no "in between" speeds.

 

[WildW]

I had a similar issue when I upgraded GTS250 to HD5850 - Windows Experience Index dropped to 6.0 with the 5850. Retested several times, reseated card a couple of times, eventually the problem disappeared and numbers shot up to 7.something. It may be a PSU issue or a Windows driver issue.

Honestly was never sure if my weak power supply was the issue, but I replaced it anyway, it was too borderline.

 

[GaiaHunter]

1) his computer shuts down because he overclocked his video card. When his vid card is running over specs, then shutting down is not out of norm.

But the OP also said

Actually, it completely shutdown at stock clocks as well as decreased performance versus my 8800gt.

I was able to play COD4 a little longer than crysis (15min) befor it shutdown as well. Like I unplugged it from the wall.

And those who say that if the power supply is weak, then the 4890 will "slow down" are essentially arguing that the 4890 has the ability to throttle back to the point that it is not capalbe of doing. Does my reasoning sound right here?

And this guy said.

I actually encountered this with my old 450W Smartpower2 PSU when I tried to run my 4850 on it - my framerates sucked in everything, and were worse than the X1900 I was replacing. Nothing ever crashed though, and every game ran properly albeit slowly. As soon as I upgraded my PSU, the 4850 began performing as expected.

Also, from http://www.guru3d.com/article/radeon-hd-4890-review-test/5

There are many good PSUs available, over the years we reviewed a lot of them and have loads of recommended PSUs for you to check out in there, have a look. Things that can happen if your PSU can't cope with the load?:

* bad 3D performance
* crashing games
* spontaneous reset or imminent shutdown of the PC
* freezes during gameplay
* PSU overload can cause it to break down

Can the 4890 just get stuck at 2D clocks? And then when he OC the speed actually changed?

I would still format Windows before changing PSU.

 

[Painman]

Well, I'm curious as well. It's always been my understanding that a power-starved PC will simply error out, or shut down due to OCP kicking in. But, lately we have this idea floating around that power-starved systems will throttle themselves somehow.

If video card makers are so keen on pointing out that they have OVP/OCP and thermal throttling functions, why don't they vaunt these supposed Undervolt and Undercurrent throttling capabilities? Has anyone hooked a 5870 up to a known shitty power supply, played a game while running MSI Afterburner's OSD, and actually observed the card's clocks/voltage throttling down in response to insufficient power?

Unless/until a scientific investigation takes place, then this idea of video cards throttling themselves in response to insufficient power needs goes into my myth list.

And it's not just video cards. If a system experiences a voltage dip, particularly on a single rail PSU, it's the entire system experiencing the voltage dip. Yet somehow, the components across the entire system have some magical ability to sense this dip within nanoseconds and respond by clocking down rather than simply erroring out.

That means the RAM, the CPU, the northbridge, the southbridge, everything, having some herertofore unknown ability to sense a voltage dip and immediately respond with nanosecond synchronicity and kep the system humming along.

Forgive me, but you'll have to color me in a vivid shade of skeptical.