AMD Quietly Made Some Radeon RX 560 Graphics Cards Worse (pcworld.com)

Brad Chacos: When the Radeon RX 560 launched in April it was the only RX 500-series card with a meaningful under-the-hood tech boost compared to the RX 400-series. The graphics processor in the older RX 460 cards packed 14 compute units and 896 stream processors; the upgraded Radeon RX 560 bumped that to 16 CUs and 1,024 SPs. Now, some -- but not all -- of the Radeon RX 560s you'll find online have specs that match the older 460 cards, and sometimes run at lower clock speeds to boot. AMD's Radeon RX 560 page was also quietly altered to include the new configurations at some point, Heise.de discovered. The last snapshot of the page by the Internet Archive's Wayback Machine occurred on July 7 and only lists the full-fat 16 CU version of the card, so the introduction of the nerfed 896 SP model likely occurred some time after that. Sifting through all of the available Radeon RX 560s on Newegg this morning reveals a fairly even split between the two configurations, all of which are being sold under the same RX 560 name. In a statement, AMD acknowledged the existence of 14 Compute Unit (896 stream processors) and 16 Compute Unit (1024 stream processor) versions of the Radeon RX 560. "We introduced the 14CU version this summer to provide AIBs and the market with more RX 500 series options. It's come to our attention that on certain AIB and etail websites there's no clear delineation between the two variants. We're taking immediate steps to remedy this: we're working with all AIB and channel partners to make sure the product descriptions and names clarify the CU count, so that gamers and consumers know exactly what they're buying. We apologize for the confusion this may have caused."

Comments

So... yield problems, in other words.

[Anonymous Coward]

Reading between the lines, one suspects that they have yield problems on the new GPUs and are having to fudge the specs to meet capacity.

Re:So... yield problems, in other words.

[MoarSauce123]

Common practice, partially working chips are sold with the broken components turned off. The level of brokenness varies, in some cases there may only be negligible defects. That was the case with several AMD processors years ago and it was fairly easy to turn the dormant cores back on. Same with the old pencil trick to run the processors at a higher native clock speed without external overclocking. Nevertheless, they should have named the crippled GPUs differently. Rather naive to think that there will be no confusion when selling two different things by the same name.

Re:So... yield problems, in other words.

[DarthVain]

Exactly.

If they really cared they would update their naming scheme rather than depending on consumers to sift though a spec page to see if it is the correct version or not.

Additionally they could abandon their ridiculous naming scheme altogether that they have been following for years and actually name their products something meaningful rather than a bunch of nonsense. That said, when your only competition is nVidia which uses their own equally baffling naming scheme I guess there isn't much pressure to change. They'll both just confuse their customers for PROFIT!

I've bee around computers and hardware my whole life (longer than most, at least insofar as video cards have actually existed), I work in a computer related field, I build my own rigs, and EVEN I when it comes time to pull out the old and buy and new one, or build a new rig pretty much sigh "Oh boy here we go, I'm gonna have to sift through the marketing BS and the rest of the detritus that is the video card industry/market"... Never mind they both have a busted reputation of cheating at all sorts of levels in the past with whatever offering they have. Need some new players to freshen up the place, no more Matrox or 3dx, or anything else out there. It seems the only competition the two current companies have is who can behave worse...

Re:So... yield problems, in other words.

[tlhIngan]

Common practice, partially working chips are sold with the broken components turned off. The level of brokenness varies, in some cases there may only be negligible defects. That was the case with several AMD processors years ago and it was fairly easy to turn the dormant cores back on. Same with the old pencil trick to run the processors at a higher native clock speed without external overclocking. Nevertheless, they should have named the crippled GPUs differently. Rather naive to think that there will be no confusion when selling two different things by the same name.

It's actually a really common practice, and not just for defective GPUs.

A modern chip requires easily $2-3M when you send it to be fabbed - a single mask costs $100,000 by itself, and modern 10+ metal layer ICs require at least 10-20 masks for the metal layers alone, then there's the masks for the transistors, etc, so you're already spending $2-3M on masks alone. It's why chip revisions are like A0, A1, B0, B1, etc. Minor stepping changes (A0->A1, for example) imply only the metal masks changed - masks are so expensive that there are tons of unconnected transistors and gates sitting idle on the silicon. The reason for this is the space is there, so you might as well fab it in (they're free), so if there is a problem, you have spare transistors and gates available to fix it. So the metal layers changed only to accommodate the fix that rewires and potentially uses these spare transistors and gates. Bigger stepping changes, from A3 to B0, mean the entire mask set was changed, including the silicon transistor layers itself (either they ran out of resources, or as is often the case, they redo the design to get more speed out of it too).

Also because of this, one design almost always encompasses many SKUs. It owuld not surprise me if the x60 series all used the same silicon design, just the unneeded parts (or bad parts) are disabled, usually by blowing fuses

Laptop versus 4U graphics cards

[goombah99]

I see GTX 1080s and High end radeons in both laptops and large well ventilated 4U boxes. Are these the same performance? How is this possible given all the fans and larger area of the PCI slot cards? if not how come I don't see some discussions of this?

Re:Laptop versus 4U graphics cards

[viperidaenz]

According to Wikipedia, the 1080's on desktop and mobile have the same specs, except mobile is a TDP of 150W and desktop is 180W.
Mobile doesn't support Boost 3.0, so peak speed is slightly lower.
If I had to guess, they're the same chips with a yield split based on voltage. Lower voltage at the same clock means lower power. Or they're identical and the mobile versions just throttle quicker, due to the lower TDP rating.

False apology

[quonset]

We're taking immediate steps to remedy this: we're working with all AIB and channel partners to make sure the product descriptions and names clarify the CU count, so that gamers and consumers know exactly what they're buying. We apologize for the confusion this may have caused.

AMD is only apologizing because they were found out. It's the same story with any company who tries to pull a fast one on consumers and is found out. Then, and only then, are steps taken to remedy the situation. Not before while they were duping their customers. Only afterwards when they were called out for their shenanigans.

I don't know, seems OK to me

[rsilvergun]

AMD can't police all their partners. There's too many of them. That said, I"m a little shocked to see Asus' name on the bad boy list with Power Color.

Oh, forgot to add

[rsilvergun]

to be fair to Asus, at least on newegg the lower end RX 560s are clearly labeled with the number of stream processors. I've noticed slight variations of specs for years. I compared about 10 different cards when I bought my bro's 1060 6gb before settling on a Gigabyte.

Smells

[Zanderama]

How can AMD blame their partners for not being clear on labelling, after AMD's own website was quietly changed with no reference to the different configurations. Surely if AMD "just" wanted to introduce more 5-series cards they could have called this the 555 or something. I guess they're hoping to release a 660 at some time and then say "wow, look at the improvement over the previous (nerfed) 560".

Re:AMD reinvents Gatrox video cards?

[Anonymous Coward]

FYI, in this context AIB = Add In Board, for those that aren't in the graphics card manufacturing industry...

GPU market now fully in the hand of fraudsters?

[ffkom]

So, nVidia sells you GPUs with part of the memory silently being "low-bandwidth" connected, and AMD silently removes compute units from GPUs of equal naming.
Now that Intel has basically abandoned all GPU manufacturing, the market seems fully in the hand of fraudsters.
Hey Electronic Arts, you should enter this market: Just sell some GPU model (let's call it the "LootGPU"), and make it to have random hardware specifications.

Re:GPU market now fully in the hand of fraudsters?

[goose-incarnated]

So, nVidia sells you GPUs with part of the memory silently being "low-bandwidth" connected, and AMD silently removes compute units from GPUs of equal naming. Now that Intel has basically abandoned all GPU manufacturing, the market seems fully in the hand of fraudsters. Hey Electronic Arts, you should enter this market: Just sell some GPU model (let's call it the "LootGPU"), and make it to have random hardware specifications.

An EA card will be a bare PCB with a voltage regulator and pads for CUs and memory. The actual CUs and memory will be DLC, available on the day of release.

Re:GPU market now fully in the hand of fraudsters?

[Kjella]

So, nVidia sells you GPUs with part of the memory silently being "low-bandwidth" connected

No, the specs nVidia gave reviewers were plain wrong:

The error, as NVIDIA explains it, is that in creating the GTX 970 reviewer's guide, the technical marketing team was unaware of Maxwell's aforementioned and new "partial disable" capabilities when they filled out the GTX 970 specification table. They were aware that the GTX 970 would have the full 256-bit memory bus, and unaware of the ability to independently disable ROPs they assumed that all 64 ROPs and the full 2MB of L2 cache was similarly available and wrote the specification table accordingly. This error then made it into the final copy of the guide, not getting caught even after being shared around various groups at NVIDIA, with that information finally diffused by press such as ourselves.

Basically, they presented it to reviewers as if it had the full 4GB available at full speed because that's what technical marketing believed themselves, both before release and quite some time after. It's only after shit really hit the fan they talked to the engineers again realized that it didn't. So this seems to me like an honest mistake and not underhanded marketing unless you think all this is a cover story and nVidia intentionally gave out false information to commit fraud, which seems rather extreme. How they handled it, like how long it took to discover it, how they addressed it, how they dealt with customers who had based their purchasing decision on it etc. can probably be the topic of a long debate though.

I find this story a much worse case of blatant of blatant and underhanded marketing. They didn't accidentally downgrade these specs. They didn't accidentally introduce two different cards with the same model number. This is pretty clearly someone's conscious decision to sell lower spec cards using a name you'll find for example in reviewer's chart giving a certain performance and now won't, even if the stores are updated. It's bait and switch at its worst, it's worse than the simple rebranding they do to make old cards look like part of a new series. It's the kind of shit that makes me wish marketing would die in a fire.

4850

[mark57]

I had a 4850 for about a year, then killed off driver support. this doesn't surprise me.

Re:AMD reinvents Gatrox video cards?

[Hal_Porter]

Do those shills on Youtube get free hardware in return for good reviews? How about sites like TomsHardware?

Why not give it a different model number?

[jonwil]

Why not give the 14CU lower-powered variant a lower model number. Then there is a clear distinction between the 2 cards and no chance consumers get confused by what the card (or PC) they are buying actually has.

Well, duh!

[Maavin]

Because EVERYONE would suspect that the specs are different?!

Worse than it sounds

[velinion]

This is actually worse than it sounds. The RX 560 is identical to the RX 460, except the RX560 has an extra two compute units and a modest clock bump. Selling a "RX 560 with two compute units disabled" is really just selling RX 460 units with a RX560 name. AMDs alterations to the description of the RX560 on their own website show this to be a deliberate move (check the wayback machine ) My guess? They had some 460s that they couldn't sell, so they bumped the clocks a bit and re-defined what a 560 was in the hopes of moving them as "better" cards.

Re:Worse than it sounds

[drinkypoo]

My guess? They had some 460s that they couldn't sell, so they bumped the clocks a bit and re-defined what a 560 was in the hopes of moving them as "better" cards.

My guess is that their yields are poor because of a design flaw, and now they're redefining the product in order to permit them to use defective parts. This move will temporarily make the bean counters happy because they're not throwing dies away, but it will long-term harm them because this is deceptive douchebaggery.

Re:Worse than it sounds

[Trogre]

So both can drive a 4k display at 60Hz and RGB colourspace?

Re:That explains it!

[higuita]

AMD is supporting their open source drivers, they are the main work force in the radeon mesa drive!!
Care to explain better what you mean by that