[OC3D.NET + Guru3D] MSI R9 290X Lightning Review

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

24601

Golden Member
Jun 10, 2007
1,683
40
86
I dont know about AMD side but I can prove that atleast on the 780 there is no difference between Hynix & Samsung.

I guess we have to get TheStilt's input on 290x lightning.

Although Gibbo @ Overclockers UK had Samsung memory on his Lightning.

Maybe someone who buys the lightning here & also owns a Reference 290x can do the same comparisions as I did so we can have some Idea.

The differences would only show if they actually set Samsung's specced timings on the Samsung equipped 780s or just got lazy and just set Hynix's or Elpida's specced timings even when they are using Samsung.

Samsung is specced for lower timings period end of discussion.

What timings Nvidia/AMD(ATi) and/or AIB set is their own doing. It's not what the memory is specced for automatically.

If someone changed the timings for the 780s equipped with Samsung they could likely lower the timings and get higher performance at equal clock speeds compared to looser timings.

TheStilt's bios gave large boosts in 3dmark even though he halved the ROPs on the card specifically because of the timing tightening.
 

rtsurfer

Senior member
Oct 14, 2013
733
15
76
The differences would only show if they actually set Samsung's specced timings on the Samsung equipped 780s or just got lazy and just set Hynix's or Elpida's specced timings even when they are using Samsung.

Samsung is specced for lower timings period end of discussion.

What timings Nvidia/AMD(ATi) and/or AIB set is their own doing. It's not what the memory is specced for automatically.

If someone changed the timings for the 780s equipped with Samsung they could likely lower the timings and get higher performance at equal clock speeds compared to looser timings.

TheStilt's bios gave large boosts in 3dmark even though he halved the ROPs on the card specifically because of the timing tightening.

I mostly agree with what you have said.

But, if Samsung has had looser timings than it needed to have than shouldn't it clock higher.? Because Hynix clocks Higher due to looser timings as you have said. We have usually seen Hynix out clock Samsung.

And if Samsung can't clock higher due to some Silicon limitation then consider this. Suppose if Samsung can manage 1800 Mhz at tighter timings, if the Hynix can mange 1900 Mhz +, shouldn't the extra 100 Mhz be able to offset the looser timings disadvantage & manage to achieve the same performance as Samsung.
 

24601

Golden Member
Jun 10, 2007
1,683
40
86
I mostly agree with what you have said.

But, if Samsung has had looser timings than it needed to have than shouldn't it clock higher.? Because Hynix clocks Higher due to looser timings as you have said. We have usually seen Hynix out clock Samsung.

And if Samsung can't clock higher due to some Silicon limitation then consider this. Suppose if Samsung can manage 1800 Mhz at tighter timings, if the Hynix can mange 1900 Mhz +, shouldn't the extra 100 Mhz be able to offset the looser timings disadvantage & manage to achieve the same performance as Samsung.

That would be true if there weren't shader limited workloads.

GPGPU and current AAA games are extremely heavily shader limited.

Unlike the fixed function side of things, shaders workloads are extremely sensitive to memory timings.

Thats the reason GPUs even HAVE L1 and L2 cache in the first place.

If they just needed bandwidth then they would just save money by just having no L1 and L2 cache, since the GDDR5 is capable of providing plenty of bandwidth.
 

rtsurfer

Senior member
Oct 14, 2013
733
15
76
That would be true if there weren't shader limited workloads.

GPGPU and current AAA games are extremely heavily shader limited.

Unlike the fixed function side of things, shaders workloads are extremely sensitive to memory timings.

Thats the reason GPUs even HAVE L1 and L2 cache in the first place.

If they just needed bandwidth then they would just save money by just having no L1 and L2 cache, since the GDDR5 is capable of providing plenty of bandwidth.

Somewhat makes sense.:thumbup:
 

norseamd

Lifer
Dec 13, 2013
13,990
180
106
And the marketing to go along with it - DRMOS 4 DrMOS 4 is the next-gen Mosfet which is used to deliver the power for the GPU. MSI DrMOS 4 does not waste much power, stays cool and is more power efficient than other powersaving solutions. Because DrMOS 4 is a 3-in-1 chip it delivers much more power with less components needed. This solution gives you the power to break world records.

so either drmos have problems or nvidia is paying them to keep them off amd cards. actually nvidia does not just do that. they often yank the right to even make nvidia cards away from suppliers. look what happened to xfx. at least they gave the 290x good memory. so how would a 290x with a custom pcb and hynix memory with a water cooler overclock?
 

rtsurfer

Senior member
Oct 14, 2013
733
15
76
so how would a 290x with a custom pcb and hynix memory with a water cooler overclock?

Expect numbers from users on air on Wednesday or Thursday if any delays.

As for Water, wait a few months as a waterblock for these hasn't even been announced yet, let alone in the hands of a user.
Should be here in a couple of months eventually.
 

Face2Face

Diamond Member
Jun 6, 2001
4,100
215
106
so either drmos have problems or nvidia is paying them to keep them off amd cards. actually nvidia does not just do that. they often yank the right to even make nvidia cards away from suppliers. look what happened to xfx. at least they gave the 290x good memory. so how would a 290x with a custom pcb and hynix memory with a water cooler overclock?

The response I received from an MSI rep was this -

First, it's a cost issue, second, 290X is a completely different beast power architecture wise than the 780 so the engineers adjust for that as both Hawaii and GK110 have different challenges.
 

Face2Face

Diamond Member
Jun 6, 2001
4,100
215
106
Very poor factory OCs on these things. The power draw is wild...

With only +100MV they are not putting down impressive core clock numbers, but with up to +200MV most are hitting 1200MHz+. And the Rbby tool works with these cards, so the sky is limit underwater. Even with the red mod they can hit nice clocks - https://www.youtube.com/watch?v=hArgTaPQpRc The power design and PCB on these cards are unrivaled by any other Hawaii card.