Samsung/Hynix Vs Alpida vRam

[Z15CAM]

nVidia advertizes Samsung and AMD advertizes Hynix. Both great products in my opinion but when you receive a card with Alpida vRam and you're an OC'r do you feel cheated - I do.

Is Alpida as good as Samsung or Hynix - What's your Experience?

 

[schmuckley]

I have Hynix/Elpida..
It tops out @ around 1800/1900
This is on AMD card..

 

[Z15CAM]

Perhaps you're missing my point.

When you order a nVidia card you expect Samsung and when you order an AMD card you expect Hynix but then you end up with Elpida vRam and everything changes in respect to the expected voltage/frequency scaling of the vRam and in most cases, from my experience, Elpida falls short.

Granted Card MFGr's guarantee a placid consistency; but, I didn't pay for Elpida, I paid for Samsung or Hynix scaling as outlined by the products reference design.

 

[Yuriman]

I think that "cheated" is a bit of a strong statement when the product is still capable of running at stock (advertised) clocks. I can concede your point on aftermarket boards advertised for overclocking - but that's not AMD/nV's fault.

 

[Z15CAM]

I think that "cheated" is a bit of a strong statement when the product is still capable of running at stock (advertised) clocks. I can concede your point on aftermarket boards advertised for overclocking - but that's not AMD/nV's fault.

Perfect-O answer: Then why do nVidia/AMD advertize their reference builds as using Samsung/Hynix vRam respectively then AIB's are free to use an alternative product compared to the reference specs (That's what we are buying) then we find it plainly does not scale the same nor does it have the over head of the components specified.

I have no argument when it comes to AIB alternate and modified advertized builds Vs reference board design and component builds where AIB board specifications DO NOT adhered to the reference spec's and it's a God Damn greedy shamble when consumers can't return a card because it does not consist of components advertized by the Principle MFGr.

Considering the Reference Spec's, I have no reason to believe Elpida is inferior but I know it's not what I paid for.

At least have AIB Partners inform the consumer as to the vRam Product they use when building a Reference Card Design whom advertizes a specific vRam MFGr.

Elpida vRam, from my experience, is 500 to 1000 Mhz short in frequency compared to either Samsung or Hynix irregardless of the voltage drop or current you push through it.

Both nVidia and AMD are guilty of this false advertizing in regards to their Reference Designs.

I've ranted enough and rest my case - What's your opinion?

 

[Headfoot]

Unless you are buying an OC-specified model you are only buying a product rated for the stock clocks and as long as the Elpida ram hits those stock clocks it is fulfilling the letter of the specification. It seems like the safe bet is to spend the $30-40 to get the OC-friendly version that will be guaranteed to have the higher end parts.

It's just like anything, you gotta pay a little more to get the higher quality. Nothing is free.

 

[ph2000]

i dont think AMD/Nvidia ever say they use a certain brand on their spec for memory chip

 

[PliotronX]

i dont think AMD/Nvidia ever say they use a certain brand on their spec for memory chip

Aye, just a lottery and I ain't ever gotten jack squattery from the lottery:

 

[daveybrat]

You should have posted this in the 'Video Cards and Graphics' section. This is for Cpu's and overclocking subjects.

 

[Leyawiin]

i dont think AMD/Nvidia ever say they use a certain brand on their spec for memory chip

Yeah, they don't "advertise" anything. The specs on their websites say GDDR5 and possibly a maximum amount and that's it.

 

[VirtualLarry]

DRAM / VRAM is a commodity, OP. Do you complain that your OJ from concentrate is partially from Bolivia instead of Florida?

 

[Homeles]

RAM is RAM. Really, memory overclocks tend to be much more dependent on the IMC, rather than the RAM. Sure, I'm sure manufacturer 1 may have better performance than manufacturer 2, but the difference is not really meaningful.

 

[PliotronX]

RAM is RAM. Really, memory overclocks tend to be much more dependent on the IMC, rather than the RAM. Sure, I'm sure manufacturer 1 may have better performance than manufacturer 2, but the difference is not really meaningful.

I have Winbond BH-6/5 and Micron D9 that sees different

 

[Z15CAM]

RAM is RAM. Really, memory overclocks tend to be much more dependent on the IMC, rather than the RAM. Sure, I'm sure manufacturer 1 may have better performance than manufacturer 2, but the difference is not really meaningful.

So in you opinion Samsung MV-3V4G3D-US_DDR3 is the same as all other Ram. Well it's not when it comes to over clocking and performance as it's far superior and that's why I specifically bought it.

 

[IEC]

I have some Elpida GDDR5 on a card that does 1700 MHz stable from a stock clock of 1250 MHz.

I also have some on a card that won't go above 1350MHz from a stock clock of 1250 MHz.

As with all things overclocking, YMMV.

 

[stahlhart]

If you are going to continue to pursue this subject here, please cite either corroborating third party sources or benchmark comparison scores of your own to support your position.
-- stahlhart

 

[Z15CAM]

Reference Gigabyte R9 290X at 1195/1500 with a +156mV off-set pegged with Alpida vRam on the PCB - Expected Hynix.

Needless to say the GPU Temp was 94C with a VRM Temp of 86C with the Ref-Cooler screaming at 95% but the card did not throttle.

My i7 2700k was idling at 1600mhz on an Asus P8Z68V-Pro Gen3 MB with 16GB's of Samsung MV-3V4G3D-US_DDR3 @ 1866Mhz/9-9-9-24-1T-1.3v, 4GB's of RAMDisk used for Win7 [64] on a Samsung 830 Pro 256MB SDD for secession sluff.

Increasing Frequency over 1500Mhz with this Alpida vRam hindered the score.

These runs are artifact free and can't push the Gigabyte Reference 290X any harder before artifacts show.

Waiting delivery mail of a XSPC Razor WB with Back-Plate.

Do you want my GPU-Z log of these runs?

 

[eric_98]

If you are going to continue to pursue this subject here, please cite either corroborating third party sources or benchmark comparison scores of your own to support your position.
-- stahlhart

Stahlhart, you've done a good job reducing the bum fight nature of these forums. But if a thread is pointless let it die on it's own. No benchmarks are needed here to make the OP's point that a bait and switch may be going on. I personally will look for it in any future cards I buy.

Warning issued for moderator callout.
-- stahlhart

 

[Z15CAM]

you've done a good job reducing the bum

You have completely lost me - What's your point unless Sponsoring has something to do with it?

If that's the case, I will concede to Anandtech.

 

[3DVagabond]

Reference Gigabyte R9 290X at 1195/1500 with a +156mV off-set pegged with Alpida vRam on the PCB - Expected Hynix.

Needless to say the GPU Temp was 94C with a VRM Temp of 86C with the Ref-Cooler screaming at 95% but the card did not throttle.

My i7 2700k was idling at 1600mhz on an Asus P8Z68V-Pro Gen3 MB with 16GB's of Samsung MV-3V4G3D-US_DDR3 @ 1866Mhz/9-9-9-24-1T-1.3v, 4GB's of RAMDisk used for Win7 [64] on a Samsung 830 Pro 256MB SDD for secession sluff.

Increasing Frequency over 1500Mhz with this Alpida vRam hindered the score.

These runs are artifact free and can't push the Gigabyte Reference 290X any harder before artifacts show.

Waiting delivery mail of a XSPC Razor WB with Back-Plate.

Do you want my GPU-Z log of these runs?

It's possible your preferred Hynix/Samsung only clocks higher because it runs looser timings.

 

[Z15CAM]

You have a point but from my experience, Samsung and Hynix vRam will do 1600+Mhz where as Elpida STOPS at 1500Mhz, irregardless of the voltage you pump through it.

You're saying Timing (Something we have no control of, one time we did but not now for vRam). So you say Elpida at 1500Mhz is equal to Samsung or Hynix at 1600mhz's - If so why doesn't it show up in Benches?

 

[3DVagabond]

You have a point but from my experience, Samsung and Hynix vRam will do 1600+Mhz where as Elpida STOPS at 1500Mhz, irregardless of the voltage you pump through it.

You're saying Timing (Something we have no control of, one time we did but not now for vRam). So you say Elpida at 1500Mhz is equal to Samsung or Hynix at 1600mhz's - If so why doesn't it show up in Benches?

No, I'm not saying that it's the same. All I'm saying is that we only see one spec. We don't see or know the timing spec. Elpida just might aim for a different balance of metrics between clocks/timings/power, or one brand might be better than another. We don't have enough evidence to decide one way or another.

Also, how much of a real world difference do these extra clocks on RAM give us? 1fps? 2fps? Personally I'm more concerned with the quality and design of the power circuitry and the PCB. In the long run crap capacitors will matter more than an extra hundy or two on the memory clocks. This is only my own personal biases, YMMV of course.

 

[Z15CAM]

Great answer and perhaps my Old School experiences are now extinct.

But I can't pass on what I see when one pays for and expects what's advertized from these VGA Card MFGr's.

And don't give me an excuse that AMD doesn't profess they don't flag Hynix vRam for thier reference boards or nVidia don't insinuate they use Samsung because that's BS.

 

[blackened23]

Everyone uses Elpida GDDR5 because of availability. Not because they want to. Not because they want to screw you over. They use it because they are required to unless they want to stop production of their GPUs altogether.. I mention this because JacobF was asked this over at the EVGA boards (don't have a link, was a long time ago) and he stated that Samsung GDDR5 has been very limited in volume in the prior 6 months and presumably still is limited. A lot of classified 780 cards at the time were popping up with Elpida memory. And naturally, some people who heard bad things about Elpida through forums and word of mouth freaked out at the EVGA forums. And that's when JacobF mentioned the scarce availability. It (Samsung GDDR5) is hard to obtain, therefore as a backup option SK-Hynix and Elpida are also used, as those are more readily obtainable.

I don't know what kind of GDDR5 the PS4 is using. If it's samsung, that could be the reason. But it's pretty clear that Samsung availability is far worse than that of the alternatives.

That said, i've had GPUs that overclocked just fine in terms of VRAM with Elpida. I think people are blowing the entire thing out of proportion. Elpida is fine. IMO. And if you do get Samsung, maybe it's a bonus - i'm not convinced that overclocks all that better, but if you get it consider yourself lucky. Because it is not easy to obtain and has had scarce availability for some time now.

Also, no GPU manufacturer advertises a specific brand of GDDR5. You get what you get basically. It could be SK-Hynix, could be Elpida, could be Samsung. This is done by necessity. No one is trying to screw you over, although i'm not convinced that any brand is better than another. I could be wrong on that, but i've had elpida GDDR5 cards overclock just fine on the memory. So...not convinced here.

 

[blackened23]

And don't give me an excuse that AMD doesn't profess they don't recommend Hynix in their reference boards or nVidia don't insinuate they use Samsung vRam because that's BS.

[citation required] Show us the AMD or Nvidia white paper that states this.

This is aside from the fact that you're not providing any evidence that Samsung is better in terms of OC'ing. Maybe it is. I don't know, i'm not convinced. But if you're telling us it is, where's the proof? We need a large sample size of 20-50 or more cards for proof positive evidence for what you're saying - as far as I can tell, you're not providing that. And you're not providing proof that nvidia or AMD recommend a specific brand of VRAM. I'm near certain that isn't the case. They CANNOT do that. GDDR5 brand has always been based upon availability and are interchangeable as long as the requirements for the vendor are met.

Either way this is a moot point. Let's say Samsung GDDR5 *does* overclock better. It doesn't matter at all because you get what you get based on availability, that's just all there is to it really from what i've read @ EVGA and some other boards. Samsung is hard to obtain and scarce. Sucks for the guys who want Samsung GDDR5, but you can't really do anything about it. Maybe you could complain to Samsung about their GDDR5 availability, but I doubt you'd make much headway there.