AMD: Volcanic Islands R1100/1200 (8***/9*** series) Speculation/ Rumour Thread

The issue -IMO- is that the silicon is usually designed to work at suboptimal performance levels, after all a power virus is just a code that maximizes silicon operation .
...
The practice of assigning arbitrary so called "TDP" numbers should stop, silicon should be cooled and programmed to it's maximum theoretical performance.

I would say it's more a consequence of the complexity of modern systems. There are so many interacting parts that can affect whether other components can be utilized.
GPUs find it easier because they heavily leverage a cut and paste design methodology that puts many semi-contained pipelines on the same chip, and the architecture is heavily geared to sustained bulk performance with a workload that can happily oblige.

CPUs have a different emphasis at all levels, and their designs frequently stop short of overbuilding because their design parameters weight the costs of overbuilding certain hardware stages and providing the infrastructure to sustain it under load differently. More stringent requirements on the latency of execution and quality of service often directly compromised sustained performance.
Desktop workloads also frequently do not give them enough work to do, because the GPU has snapped them up.


No vsync ftl I guess ;) It is quite expensive and I'm not even really sure why. I guess it just happens to hit a point where most parts of the GPU are near full throughput without any one significant bottleneck.
It's a menu screen that initially had no FPS limit. It's probably a load simple enough that no individual GPU unit, memory, or the PCIe bus has a significant bottleneck problem, while also being so trivial that the GPU unexpectedly loses the protection of being CPU-limited.

How about ATi address the Blackscreen issue people are having with the 290(X)?
It's a SERIOUS issue, and it's not going away.
Pretty pathetic that folks in the know haven't spoken up yet.
The behaviors described don't provide a lot of information as to the root cause, and may have a complex set of physical and system factors that play into it occurring. This might explain why it took an expansion of the device pool and configuration space for it to pop up.
There may be a number of different issues that look similar. There are a few claims that I personally find interesting to think about, but a few forum posts somewhere isn't a lot to say which anecdote falls in which bucket.

If it's as low-level as some think it could be, verifying it could take some time (edit: and might involve some corporate politics).


One random thing I've observed that almost nobody on the Internet can spell Elpida.
 
Is that an option in all games (through drivers)? Genuine question.

AFAIK, AMD cards allow forced SSAA in the drivers and downsampling can be done through a 3rd party application that I think it needs to be updated every time there's a new driver.

nVidia cards allow both forced SSAA and downsampling through the drivers. Downsampling isn't a feature that appears as an option in the drivers, but the driver does let you force-enable a higher resolution than the highest supported by your monitor in Windows and then have the GPU's scaler to downsize the image to a supported resolution.
The higher resolution then appears as an option in games.
 
h work to do, because the GPU has snapped them up.



The behaviors described don't provide a lot of information as to the root cause, and may have a complex set of physical and system factors that play into it occurring. This might explain why it took an expansion of the device pool and configuration space for it to pop up.
There may be a number of different issues that look similar. There are a few claims that I personally find interesting to think about, but a few forum posts somewhere isn't a lot to say which anecdote falls in which bucket.

If it's as low-level as some think it could be, verifying it could take some time (edit: and might involve some corporate politics).


One random thing I've observed that almost nobody on the Internet can spell Elpida.

Did I spell it wrong?

All I know is I've been reading about issues for days, and I'd bet $ that AMD is WELL aware of the issues, yet nothing....some folks think they are ignoring it ( I'm not one of those people).

This is their top shelf/ $ card, this kind of IDIOCY is unacceptable on any level.
 
Did I spell it wrong?

All I know is I've been reading about issues for days, and I'd bet $ that AMD is WELL aware of the issues, yet nothing....some folks think they are ignoring it ( I'm not one of those people).

This is their top shelf/ $ card, this kind of IDIOCY is unacceptable on any level.


If there's such a critical problem with many cards, why aren't we being flooded with news of massive refunds for faulty 290/290X cards?
 
You spelled it correctly. You might be one of the select few that can, going by the posts elsewhere.

Thought so.

This is a serious problem, even more important than the actual performance #'s IMO, as people are buying these top $ products to perform....and they don't even work?

What good are great benches if your card doesn't even function?
 
If there's such a critical problem with many cards, why aren't we being flooded with news of massive refunds for faulty 290/290X cards?



Folks are still trying to dope out the issues, these things weren't mass available either, and some have Hynix chips on them.
How can they " Flood" the channel this early?

The mass flood of RMAS are coming, and it SEEMS that all of them have Elpida chips ( unless the issue is a different one not associated with this issue)..

Am I happy this is happening? Absolutely not, I'm an AMD fan, but it is what it is.
 
The practice of assigning arbitrary so called "TDP" numbers should stop, silicon should be cooled and programmed to it's maximum theoretical performance.
Sorry, I forgot to respond to this earlier.
If you want dozens of SKUs per chip stepping, custom heat sinks for every chip, and wildly varying electrical requirements, then sure.
TDP is the specification that the design uses to make sure there is some kind of consistent and safe target for the physical behavior of the system, since at some point the real physical world does intrude.

Furmark and 4870 was a VRM issue, motherboards still die on prime95, orthos.
Was the card drawing power within spec?
 
He speaks the truth, sadly. Sad bro, muzz is trying to help you folks out and ya'll are telling him to STFU.

The actual specific scientific notation of the equallibrium mass element defect inside the actual physical layer presence of the driver low level communication warrants more samples be tested before actual fault can be ruled directly at VRAM manufacturers or vddci control elements within powerplay modules of ascrupulation. Therefore, HUSH MUZZ!

Did i do it right?
 
Anandtech's numbers show the 290X uber mode to be faster than 780ti

nv-amd.png


Grabbed copies and will mirror them just in case
wow :oops: the hole for AT gets bigger.
 
I would say it's more a consequence of the complexity of modern systems. There are so many interacting parts that can affect whether other components can be utilized.
GPUs find it easier because they heavily leverage a cut and paste design methodology that puts many semi-contained pipelines on the same chip, and the architecture is heavily geared to sustained bulk performance with a workload that can happily oblige.
Good points, but If a GPU is more prone to being pushed to it's limits .. shouldn't it be compensated properly by better thermal management, which would allow it to reach new performance heights? My understanding is that GPUs don't work out all of it's areas simultaneously , there are those who are left idling or at minimum load. A power virus supposedly fires up all areas, requiring more power, only to be throttled down before the chip burns to smithereens!

A power virus is the best practical example for extracted performance from a given silicon. This example should copied and imitated to be used by games and 3d applications, not to be hammered down like it's the new plague!

It's a menu screen that initially had no FPS limit.
I believe it was coupled with a famous driver issue that caused the fans in NVIDIA GPUs to stop working.

Sorry, I forgot to respond to this earlier.
If you want dozens of SKUs per chip stepping, custom heat sinks for every chip, and wildly varying electrical requirements, then sure.
TDP is the specification that the design uses to make sure there is some kind of consistent and safe target for the physical behavior of the system, since at some point the real physical world does intrude.
I see. I still think we can get away with at least more exotic and specific cooling solutions, it will raise the price for sure, but performance should rise too. with that comes a new product segmentation, high-end will become more high-end, middle class will become the new high-end, low-end will become the new middle class .. etc. And lower prices can be set for lower products.
 
There is something completely messed up with their testing methods and procedures...

Who told them to test with open stand and more importantly- WHY? How many people leave their parts like that at home? :devilish:
 
It's good to see a third-party cooler making such a huge difference. Apparently Tom's was a reviewer who had issues with the noise as well since they actually looked for solution.

Wonder what interior case temps are in relation to the "cooler" tests performed?

I know some people on overclock.net are attempting different cooler solutions, so will be interesting to see if this cooler gives the R9 290 series a higher benchmarking capability.
 
Wonder what interior case temps are in relation to the "cooler" tests performed?

Wonder it or not but you should test in environment which closely resembles to what the end-user will see...

What will happen if I tell you that the stock cooler will perform BETTER inside a case with good air-flow?
 
He speaks the truth, sadly. Sad bro, muzz is trying to help you folks out and ya'll are telling him to STFU.

The problem was the delivery.

Try this:

Ratcheted-Down muzz said:
I'm really surprised that this isn't being discussed in this thread but many users are reporting having black-screen issues with 290-series cards. This seems to be especially prevalent on models with Elpida memory. It hasn't been determined whether it is an issue with the chips themselves, a manufacturing issue or a BIOS issue.

I am *VERY* angry about this. I fell that this is something that AMD should have caught before letting these cards end up on their customer's hands.

This gets the same message across and, minus the ranting, would have likely been received much better.
 
What's everyone's opinion about the stock cooler on the 290x for someone who doesn't overclock? Does it do a good enough job or should I wait for non-reference coolers to come out?
 
Back
Top