-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
Radeon RX 6900XT and 6800/6800XT final specs. 16 GB G6. Up to ~21Tflops(~24 Tflops Boost)
- Thread starter Leonidas
- Start date
SantaC
Member
Amd was waiting for RDNA2 to compete
regawdless
Banned
Doubtful in my opinion. I feel like Nvidia has priced the cards so aggressively, that AMD was caught a little bit off guard.
Let's see how low they can go with the prices.
Guilhermegrg
Member
Considering it's rumors...Mighty strange all models have 16GB. Maybe some feature / approach that AMD is pushing for all models and it requires that amount of memory?
Silver Wattle
Gold Member
Always fun to see Microsoft's marketing effect on people.
regawdless
Banned
Or it was a way to actually save money because they were able to negotiate better prices through larger quantities.
Guilhermegrg
Member
Ok but why spend it on the weaker model anyway? If this is real, there should be a better reason for it I think.
Ascend
Member
How much power did the Vega 56/64 cards consume, and how much does the 5700XT consume?
What die size are the Vega 64 GPUs, or the Radeon VII GPUs, and what size is the 5700XT?
Do you think it was impossible for them to release a larger die Polaris?
Do you think it was impossible for them to release a larger die Navi 10?
KungFucius
King Snowflake
How the hell do AMD's AIB partners put up with this bullshit? Only AMD can sell the highest margin parts! That is super shitty.
regawdless
Banned
That's just an educated guess in my part.
They went for cheaper, widely available RAM. When they negotiate prices with their supplier, only going for one component but guaranteeing huge quantities over some years, can make big differences. To the point that it's actually cheaper to use it for all cards instead of going for multiple configurations. Using the same RAM for all cards makes a lot of things easier and cheaper outside of the purchasing price as well (uncomplicated logistics, faster processes, only one part number in the warehouse etc.).
Again, it can have a totally different reason or be a combination of multiple reasons.
Guilhermegrg
Member
AquaticSquirrel
Member
It's not that every RDNA2 card from AMD will use 16GB, the way it works is AMD/Nvidia start out designing their die with certain number of cores (CUs/SMs), memory config, target clocks, efficiency etc...
So what happens is they design the "full" version of each die, with maximum cores enabled. But just only selling that max version is not cost efficient as there will be faulty silicon where some of the cores fail, don't work properly or have other issues.
The silicon/wafers are expensive, so the companies take faulty versions of their full card and pick some number of cores to disable. This means they can repurpose the faulty ones to create more models which allows them to target more market segments. These are known as "cut down" variants.
This is all planned from the beginning normally to stratify in this way, also clock speeds etc may also be adjusted to fit the market segment or for power draw/thermals.
Anyway, in this case the revealed cards for AMD are their high end cards, these are all actually based on one design called "Navi21", the full die of which has 80 cores with 16GB.
The other two models are cut down versions of this with less cores enabled, so they are all Navi21 variants and it probably makes more sense to keep the memory amount the same as they are essentially the same chip with cores disabled and clock/power draw changes.
xPikYx
Member
If we have a situation in which let's say AMD top card is within the 10% of performance over the 3080 and AMD prices the card at 499 nVidia could drop the price of 50/100 dollars also because there is a lack of cards at the moment and for sure we won't see too many scams again
CuNi
Member
Difference between this guy and Igor:
Igor has industry contacts and very good ties with the general industry.
Igor seems to already know more than most "leakers out there as evident by this quote:
"Out of respect for the launch event planned for 28.10.2020, everyone will of course understand that a website does not want to mutate into a spoilsport. Therefore, some information is intentionally not included in the article and I will not (cannot) make a performance forecast here and now. "
The guy on the other hand says "I don't want to say Igor is wrong" and later "I think the most obvious thing to do is wait", which is a very defensive stance. If he has a "reliable" source, why all this talk like "Well my guy told me XYZ but who knows, let's just wait". He sounds like he is not very confident in his sources.
AquaticSquirrel
Member
You do know Red Gaming Tech was the one who leaked the infinity cache right? Plus most of his info about RDNA2 has been pretty much spot in so far. I would trust him more than some of what I've read from Igor to be honest.
I don't think Paul's only sources are AIBs
CuNi
Member
I'm not here saying he only talks shit or FUD.
Yes he got some things right, I won't deny that. MLID also got some things right.
But when it comes to leakers it's "hit or miss". They do get some things right, but they also get a lot wrong and they live off of "speculations" and rumors.
Igor on the other hand is more or less definitive with his posts.
He got the BIOS from Patrick Schur, who also has a good track record etc.
Igor at least shows exactly how he comes to his claims about power draw etc. and provides evidence which is way more than someone saying "I got someone who told me".
Pagusas
Elden Member
Once all the facts are out we do, we use head to heads in individual games. No one will be talking TFLOPS once all the cards on the table.
wordslaughter
Banned
Other than that the only other, and far greater option, is to wait for benchmarks.
TFLOPS don't matter if they aren't reflected in the benchmarks. The same can be said about RDNA 2's higher clock speed.
Ascend
Member
A regression in performance per clock per CU is highly unlikely for RDNA2.
MarcCppDev
Neo Member
More like Navi 21 and 22 lite
wordslaughter
Banned
It will be very interesting to see the actual clocks that can be sustained.
longdi
Banned
My guess is around 2ghz for their highest binned sku. Anything more is just marketing efforts and unsustainable without breaking an overclock and the efficiency wall.
Md Ray
Member
??
Can you elaborate.
Black_Stride
do not tempt fate do not contrain Wonder Woman's thighs do not do not
I'm talking about stock operations, that's the game clock this time round, won't be higher than specs, unlike rdna1 with its +50-70mhz more headroom.
I guess AIB can go higher than game clocks specs, maybe the very best lightning/rog/Xtreme models can do about 2.15ghz.
Because 7nm is 7nm.
With a 2.5Ghz max, 2.4 should be doable for AIB cards.
