Tuesday, December 5th 2017

AMD Officially but Silently Downgrades Radeon RX 560 with an 896 SP Variant

The phenomenon of Radeon RX 560 graphics cards with 896 stream processors is more widespread than earlier thought. It looks like RX 560 cards with 896 stream processors will be more widely available than the previously thought Greater China region; with AMD silently editing the specifications of the SKU to have either 896 or 1,024 stream processors, as opposed to the 1,024 it originally launched with. There are no clear labeling guidelines or SKU names to distinguish cards with 896 stream processors from those with 1,024.

The Radeon RX 560 and the previous-generation RX 460 are based on the 14 nm "Polaris 11" silicon, which physically features 16 GCN compute units (CUs), each packed with 64 stream processors. The RX 560 originally maxed this silicon out, with all 16 CUs being enabled, while the RX 460 has two CUs locked. The decision to change specs of the RX 560 effectively makes it a re-brand of the RX 460, which is slower, and provides fertile grounds for bait-and-switch lawsuits.

Source: Heise.de
Add your own comment

129 Comments on AMD Officially but Silently Downgrades Radeon RX 560 with an 896 SP Variant

#1
Jism
Yeah well then you OC the card a little to compensate that what you have lost. Frankly it's best buy is a RX580/480. Handles WQHD perfectly fine.

This reminds me of the Radeon 800 pro vs Radeon 800XT. You could flash it and unlock a few shaders to get a few hundreds of points for free. But the difference really was marginal, nothing to write about it.
Posted on Reply
#2
Hood
Jism said:
Yeah well then you OC the card a little to compensate that what you have lost
Terrible argument, jism - you can OC the good version and get even more performance, so stop with the grade school comebacks already, you sound like you're 10 years old.
Posted on Reply
#3
EarthDog
Jism said:
Yeah well then you OC the card a little to compensate that what you have lost. Frankly it's best buy is a RX580/480. Handles WQHD perfectly fine.

This reminds me of the Radeon 800 pro vs Radeon 800XT. You could flash it and unlock a few shaders to get a few hundreds of points for free. But the difference really was marginal, nothing to write about it.
lol, you overclock the 560 too.. same performance difference again..

Lol, those cards arent a 560/560d. They are 20% faster. These barely play 1080p on ultra in many titles. Again, fps count more so down low, than it does up high. ;)
Posted on Reply
#4
Jism
Hood said:
Terrible argument, jism - you can OC the good version and get even more performance, so stop with the grade school comebacks already, you sound like you're 10 years old.
Videocard OC'ing is a net result of 5 up to 15%, depending on how well your chip OC's. We're talking pretty much small numbers here. I've overclocked various cards as well, put them under subzero, and all that *fun* stuff many years ago, but the point being from this is that if you want a good performing card you really should stick your money into a card that was designed for it. The RX480/580 is a good middle weight (simular as 1060) and was released back then for 150$ (4GB) and 200$ (8GB). The price difference for a RX560 or 460 is such small that you could simply cut on a expensive memory kit and go for a 580 rather then a 560.

The performance difference will be again, a few percent, and not 'loads' of difference.
Posted on Reply
#5
Midland Dog
Vya Domus said:
About what are you arguing ? It has 4GB of GDDR5 memory without the bandwidth which Nvidia claimed it had. One of the memory chips is connected with a pathetically narrow bus. They obscured that aspect. That was the issue , not that it said 4GB on the box or whatever.

It's funny because most fanboys shit talked the 970 without ever knowing exactly what was the issue with it and mistakenly made the "3.5+0.5" maymays.
no where on the box did it say that it had the full bandwidth, thats just people assuming things, the box never said what bandwidth it had, it just said 4gb, which was true, if the box said 4gb gddr5 @ full bandwidth then yeah but no it said 4gb and the card can use 4gb even if the last bit was knee capped, its still not as bad as cutting the spec of a gpu after its been on the market for so long
Posted on Reply
#6
9700 Pro
It would be cool if those (every card) could be unlocked with a bios update like in the old days. :toast:
Posted on Reply
#7
Tatty_One
Super Moderator
Jism said:
Videocard OC'ing is a net result of 5 up to 15%, depending on how well your chip OC's. We're talking pretty much small numbers here. I've overclocked various cards as well, put them under subzero, and all that *fun* stuff many years ago, but the point being from this is that if you want a good performing card you really should stick your money into a card that was designed for it. The RX480/580 is a good middle weight (simular as 1060) and was released back then for 150$ (4GB) and 200$ (8GB). The price difference for a RX560 or 460 is such small that you could simply cut on a expensive memory kit and go for a 580 rather then a 560.

The performance difference will be again, a few percent, and not 'loads' of difference.
If a guy wants to play his favourite 5 year old game at low res, does not have a lot of cash and therefore does a little research to see what is likely to run that game, comes up with a/the/original RX560 goes and buys it then wonders why the game stutters to crap just to find out he had not bought a full blown 560 because it was not clearly marked as such then for me he has a good reason to be a little upset.

That 10% matters when you are near the "cusp" of smooth gameplay.
Posted on Reply
#8
Jism
I think the 10% of taken out of proportion, it's stated only 4% and in most extreme cases (Which none in here could provide) 10% ...
Posted on Reply
#9
Tatty_One
Super Moderator
I think mathematically it would be difficult to be that low, the 560 has 14% more shaders and is clocked higher but I agree there seems to be little "real world" evidence currently.
Posted on Reply
#10
Apocalypsee
There is no replacement for displacement, period. I say it quite a number of times. In FireStrike to get similar amount of performance as fully unlocked core requires over 100MHz core clock for the 128sp and 8 TMU deficit, and framerate is 10% difference in both game test. Refer to my post in Your PC ATM thread. I done more testing just now to double confirm and yes I get similar result throughout.
Posted on Reply
#11
Jism
It's like 2 to 3 frames a second difference.
Posted on Reply
#12
Vya Domus
Midland Dog said:
no where on the box did it say that it had the full bandwidth, thats just people assuming things, the box never said what bandwidth it had, it just said 4gb, which was true, if the box said 4gb gddr5 @ full bandwidth then yeah but no it said 4gb and the card can use 4gb even if the last bit was knee capped, its still not as bad as cutting the spec of a gpu after its been on the market for so long
You are stuck on such meaningless details who gives a damn what was written on the box and what wasn't. On Nvidia's website and everywhere else it was meant to have '224 GB/s of memory bandwidth' like the 980. And it didn't , that's all there is to this. After so much time this matter still mystifies people.

The point is , that was different from this situation. In this case we have a product that has two separate entries and the issue is that the consumer doesn't know exactly which is which. However the existence of these specs was not concealed , just simply not made clear.

In the case of the 970 it was concealed , everywhere by everyone. Nvidia and the AIBs. The reason why only Nvidia got blamed is because they said that they distributed incorrect marketing material "by mistake".

So to get back to your initial comment , your observation simply isn't relevant to this discussion. I have no clue as to what are you even trying to prove.
Posted on Reply
#13
Apocalypsee
Jism said:
It's like 2 to 3 frames a second difference.
You missing the point. That few frames is stand between semi playable to playable, and you know FireStrike is known to be tough on low end cards. On games you can see it will made a larger difference, you can check out reviews for that.
Posted on Reply
#14
bug
Vya Domus said:
You are stuck on such meaningless details who gives a damn what was written on the box and what wasn't.
Apparently AMD does. According to their statement, this is exactly what they're working on fixing together with AIBs.

Vya Domus said:
On Nvidia's website and everywhere else it was meant to have '224 GB/s of memory bandwidth' like the 980. And it didn't , that's all there is to this. After so much time this matter still mystifies people.
Yes and in the middle ages we used to burn people to the stake (though we were pretty open about it, so that may have been ok in your book). What does ancient history have to do with questionable labelling of a product? If some made a bigger blunder in the past, all future, lesser blunders are ok? I don't understand.
Posted on Reply
#16
bug
Assimilator said:
Is Jism the official TPU AMD apologist?
I'd say him and Vya Domus are in Nvidia's pocket, because their attitude certainly doesn't make AMD any favours.
Posted on Reply
#17
newtekie1
Semi-Retired Folder
Vya Domus said:
You are stuck on such meaningless details who gives a damn what was written on the box and what wasn't. On Nvidia's website and everywhere else it was meant to have '224 GB/s of memory bandwidth' like the 980. And it didn't , that's all there is to this. After so much time this matter still mystifies people.

The point is , that was different from this situation. In this case we have a product that has two separate entries and the issue is that the consumer doesn't know exactly which is which. However the existence of these specs was not concealed , just simply not made clear.

In the case of the 970 it was concealed , everywhere by everyone. Nvidia and the AIBs. The reason why only Nvidia got blamed is because they said that they distributed incorrect marketing material "by mistake".

So to get back to your initial comment , your observation simply isn't relevant to this discussion. I have no clue as to what are you even trying to prove.
The other important factor with the 970 issue, that people seem to like to ignore, is that nVidia never changed the performance. The GTX970 cards they sent to reviewers were exactly the same as what people were buying at retail. This meant that the reviews were not giving misleading performance information. And most people look to reviews for performance information. I don't really think anyone buys a card based on the number of ROPs it has, or the memory bandwidth, or at least they shouldn't be. They buy cards based on how they perform(in more ways than just framerates too).

However, this is not the first time AMD has tried to directly pull a switch on people, and released cards that will perform worse than what was reviewed. People going out to buy a graphics card today, trying to do proper research before they buy, are going to be reading reviews and they might be looking at buying the RX 560. They will see performance data for the full RX 560, because lets face it we are far enough along in the product cycle that no major review site is going to go back and re-review the RX 560 896SP Edition, and AMD and AIBs aren't sending out review samples either I'm betting. This is where the real problem is, and what makes what AMD did much worse than what nVidia did.

Also, to be fair, I don't think the AIBs even knew about the memory issue on the GTX970. As far as they were concerned, they were designing PCBs to connect to a GPU that had a 256-Bit memory bus. The memory mapping partition and memory bus being divided was handled all inside the GPU.
Posted on Reply
#18
Vya Domus
Nvidia did nothing wrong or at least no where near anything that would be an issue. The lawsuit was to congratulate them about how honest and consumer friendly they are. I just misinterpreted things.

AMD on the other hand is pure evil , tryna steal our money with their lowest end cards and minuscule market share using cutting edge scamming techniques.

It's OK , we know the story. Seriously , no need for a wall of text to explain that..
Posted on Reply
#19
newtekie1
Semi-Retired Folder
Vya Domus said:
Nvidia did nothing wrong or at least no where near anything that would be an issue. AMD on the other hand is evil , tryna steal our money with their lowest end cards and minuscule market share.

The lawsuit was to congratulate them about how honest and consumer friendly they are. I just misinterpreted things.

It's OK , we know the story. Seriously , no need for a wall of text to explain that..
I never said nVidia did nothing wrong, the AIBs did nothing wrong, but nVidia did do wrong. However, AMD's wrong doings are worse, and they've been doing them since long before the GTX970 fiasco.
Posted on Reply
#20
Vya Domus
newtekie1 said:
AMD's wrong doings are worse, and they've been doing them since long before the GTX970 fiasco.
Such as ? If there are so many of them, how do you quantity them as being worse ? And if there's clear cut answer to that how did AMD manage to avoid so many lawsuits ? I mean the 970 was a small thing according to you , yet a shitstorm took place.

Legit questions , I am curios as you seem 100% convinced that's how things are.
Posted on Reply
#21
bogmali
Looks at thread title and want to ask:

How did it transition to Nvidia :confused::wtf:

Let's not make this into another AMD v Nvidia drama:banghead: otherwise I will issue thread bans or vacations to those who continue to troll/derail this thread.
Posted on Reply
#22
newtekie1
Semi-Retired Folder
Vya Domus said:
Such as ? If there are so many of them, how do you quantity them as being worse ? And if there's clear cut answer to that how did AMD manage to avoid so many lawsuits ? I mean the 970 was a small thing according to you , yet a shitstorm took place.

Legit questions , I am curios as you seem 100% convinced that's how things are.
In the HD4000 days the HD4850 was reviewed and was a pretty strong performer. It was an extremely good value, arguably the best value at the time for a high end card. But a few months after the launch, they found out that their crappy single slot reference cooler that they recycled from the HD3850 wasn't good enough, and the VRMs on reference cards were prematurely failing on reference cards under extended load. So they added code to their drivers to detect high load and throttle performance. This lowered the cards performance. But it was long enough after the reviews that no-one went back to update their reviews with the new performance numbers and no new HD4850 reviews were coming out.

That's the first time I remember them pulling a stunt like this.
Posted on Reply
#23
Vya Domus
newtekie1 said:
In the HD4000 days the HD4850 was reviewed and was a pretty strong performer. It was an extremely good value, arguably the best value at the time for a high end card. But a few months after the launch, they found out that their crappy single slot reference cooler that they recycled from the HD3850 wasn't good enough, and the VRMs on reference cards were prematurely failing on reference cards under extended load. So they added code to their drivers to detect high load and throttle performance. This lowered the cards performance. But it was long enough after the reviews that no-one went back to update their reviews with the new performance numbers and no new HD4850 reviews were coming out.

That's the first time I remember them pulling a stunt like this.
Can't seem to find any source confirming this event you mentioned but I'll take your word for it. Still, what you described could hardly be called "a stunt" whose purpose was to deceive and scam people but rather it's about having a faulty design on one of their products. This is something which is certainly not unique to AMD and it doesn't help backing up your claim that their doings have been so much worse. I can point out to many other similar instances of poor design behind competing products which proved detrimental (worse than just reduced performance) as time went by.

But most importantly it's totally unrelated to things such as false adverting and having multiple specifications of the same product which is what this is mainly about. There is a pretty significant difference between those things and examples of bad engineering.
Posted on Reply
#24
newtekie1
Semi-Retired Folder
The reasoning behind doing it doesn't make it better. It doesn't change the fact that they released a card, let it be reviewed, and then limited its performance after. And they didn't just limit the reference cards, they limited all HD4850 cards via the driver.

Instead of just replacing the faulty cards with a fixed version, they decided to just limit performance.
Posted on Reply
#25
Hood
newtekie1 said:
The reasoning behind doing it doesn't make it better. It doesn't change the fact that they released a card, let it be reviewed, and then limited its performance after. And they didn't just limit the reference cards, they limited all HD4850 cards via the driver.

Instead of just replacing the faulty cards with a fixed version, they decided to just limit performance.
We can't really expect a small company like AMD to do the right thing - that might have cost them serious money, and what would the stockholders say? This is why companies use questionable tactics, and all of them do at times, because they have to answer to those stockholders. Humans all make mistakes, damage control is a make or break proposition at times. That's why I prefer to deal with larger companies, like NVIDIA and Intel, they have more resources to do the right kind of damage control, instead of the shady kind. As for AMD, I can think of a dozen examples of their shady tactics just in the Ryzen and Vega releases this year. I know NVIDIA and Intel have done the same things in the past, but AMD is the champ, in my perception.
Posted on Reply
Add your own comment