AMD NAVI 12 chip intended for RX 5800 ?

by My calculations 5800XT is the card that will compete with 2080-2080S at 599$ . If AMD brings these number to life , NV 2000 series are finished.

 

Wonder if that's going to be the 60 something compute units card then as a upscale of RDNA1 and Navi10 in-between the RDNA2 and Navi20 rumors for next year if that's still happening.
Would be interesting to see how well a larger chip with more of these cores could compare and how well Navi at least for this first version of it will scale though I guess the issues with the Fury and Vega have mostly been resolved other than from what I am reading memory still being something of a bottleneck for the cards.

Clock speeds and voltage will also be a thing if it's going to be 1.2v or higher or a lower clock speed and how much will clock speed affect performance with the additional cores now helping out.
(And what happens when a user just ticks off another -50 to -100mv in Wattman ha ha, so far it looks like that has remained almost unchanged and it mostly just makes a minimal difference to max boost speed but sustains higher average boost.)

EDIT: Hmm I guess Phoronix and other sources also don't have any info yet from the Linux driver other than maybe the bare basics with device ID's but that might give some insight into the rest of the hardware though perhaps AMD will be faster with a full announcement for when this launches.

Still room for a 5900 too although that's entirely up in the air I guess or a 6000 refresh with the next GPU model if they're making additional tweaks to RDNA next and whatever else.


EDIT: Perhaps it could go higher now too but I assume there is some point where it's not as efficient anymore plus the die size for pushing for something like doubling the existing 40 CU's to 80 though it'll be fun to see how the GPU performs and what the specs will be once info is available on these.

 

I suspect that by the time this "Navi 12" hits the market Nvidia will probably be in the process of taping out the 3xxx series. I hope AMD aims their sights a bit higher than the 2080s'

 

For competitiveness, important part is performance per transistor, per clock at price comparable manufacturing node and achievable clock. Navi surely is clocker. And power is actually not bad.

Navi is currently in better position over Turing except total power draw at peak clock. But thing is that it looks "bad" only because it can clock that high. Bigger GPUs will clock lower even while still being capable to reach 2GHz and more due to power targets.

My 5700 XT clocks to 2GHz at 1.075V And I just thrown that value there. Therefore it may reach it even with lower voltage.
I'll look in time for way to enforce static clock and voltage. To know what is minimal voltage required at 1GHz, 1.2GHz, 1.4GHz, ... This will provide further insight into what can be done and where is optimal clock to power draw point.

 

@Fox2232 Are you on a blower style cooler still or have you modded it? And do you play Hell Let Loose?

I've got an Arctic Accelero Xtreme IV coming shortly and will probably buy the 5700XT blower style anyway, so I was wondering about your overall experience.... Thanks in advance

 

No actually the only things that matter are performance per dollar, performance per watt (for heat and noise only) and features. Navi wins the first category, AMD and Nvidia are tied on the second and Nvidia wins features. This is all out of the box however. No one should ever buy a product and expect it to tweak outside of stock. So it’s actually a good time to be a consumer in this space. Things are about to get interesting.

 

Performance per $ (competitiveness) is consequence of "ratio" I wrote about. If card has bad performance per transistor investment at same clock or can't reach higher clock to compensate, it will result in bigger GPU that is more costly to manufacture while delivering same performance as opposition.

@Jagman : I considered Accelero Xtreme IV too. But I am not fond of "perma-glue". When I did it to HD7970 it was race against time. Felt like job for two people.
I still did not push it to limits outside of Path of exile where maxing out With GI+Shadows on Ultra is not really making visual difference, just eats fps

Here is PoE maxed out test in my hideout that is rather full of light sources:

PHP:

Set.Clock   Set.Voltage Actual.Clk  FPS     Power
[MHz]       [mV]        [MHz]               [W]
1200        771         1160        45      76
1400        838         1354        52      92
1600        917         1546        59      115
1800        1011        1735        66      142
2000        1118        1938        72      185
2100        1176        2040        76      220
2150        1215        2098        77      255

 

That does not always translate to the end user though which is why MSRP is more important. You can play your oral gymnastics all you want but the bottom line is cost to end user.

 

No way, Jose

 

It is not any kind of trickery. Imagine that there would be real price war. Those values will tell you if given GPU can be manufactured at profit in times competition is at zero or loses money by making theirs.
It is very close to feasibility.

 

There is one "small" thing that many of us seem to be unable to understand.
We can criticize Nvidia how much we can, me as the first one, but at least they have released all their new cards this time by letting the customers decide which one to buy.
AMD instead is making some dirty strategies here, they are releasing their products one after another sparse in time by trying to get much money as they can.
At day zero they release the 5700 xt (crap) then they go with the 5800 and then with 5900, 5950 and who knows wich others.
Not so much intelligent customers that ve bought the 5700 will surely start to think on an immediate upgrade.
AMD are not the good guys remember. They are doing business and they are doing it well thanks to subtle strategies and people that appreciate such moves.

 

I don't think there's many people who think AMD is the good guy. And those who think that can just be ignored. All those companies, AMD, Intel, nVidia, Google, Microsoft, Sony, Apple, ... care about one thing. It's called money. Don't ever be fooled thinking otherwise. It's all about money for those big publicly traded companies. And as a consumer you should also care about one thing. It's also called ******* money. nVidia, AMD or Intel personally i don't give a flying ****. I'll buy whatever is best for my budget. Right now i try to not spend more than give or take 300-400$ CAD on the CPU and 500-600$ CAD on the GPU (1k total max for the two). If i would be upgrading now the RX 5700XT would be a very interesting card depending on the price of the non blower models.

 

It's not that simple. AMD is a cash strapped company, every product launch has a massive upfront cost associated with it - from engineering, to inventory, to marketing. That's not to mention that 7nm yields were not great and the cost per transistor was higher than 12/14nm. The cost scaling of a larger Navi chip might not work against Nvidia's higher end chips until EUV or yield increase.

They are a business and these are business decisions but its not like AMD is sitting there going "lol we're not going to give our customers a choice" its more like "this is the best we can do given a bunch of constraints"

 

Both Nvidia and AMD have done this several times in the past and it´s perfectly normal! It´s very rare for any of them to release the entire productt stack at the same time. Not to mention everyone knows that there´s going to be something faster then the 5700xt... and something slower too!

 

Oh, I'm sorry. You said you like paying $1000 for *80 ti? What's next gen? $1500 for *80 ti and $1000 for *80 non-ti? That's what happens with no competition. Or you end up with the same product refreshed year after year with only tiny increase in performance and an increasing price tag, because the company has no reason to innovate. Why make something better if there's nothing else threatening your top-dog position in the market?

Be glad AMD is releasing anything at all. We would not have gotten the "super" release without it (which IMO is a stupid release anyways)

 

So maybe 5800/XT's in time for Christmas?? (Navi12)

5900/XT first half of 2020? (Navi20)

These cards should compete with Nvidia 3000's I would think.

 

Denial, H83, Ssateneth,
It could be. It may be that AMD could not do anything more this time but I am of the opinion that hw manufacturers can not be considered on par like software houses.
I ve always been much more inclined to believe into a software house stating that they did what they could and nothing less instead of believing to the same words told by a hw company.
Can we be so naive after that AMD did release the Radeon 7 gpu? It has passed only half a year since the official release and the numbers say that they could make a bigger gpu.
Radeon 7 does bring 13.2 billion transistors while the 5700xt 10.3 b.
Once I was much more romantic and inclined to believe in fairy tales, years have passed and I learned my lessons.

 

The problem i have right now is what is considered mid range and high end. I have no trouble with nVidia or AMD asking a premium for Titan and 80Ti level cards. It makes perfect sense. But mid range used to be 70 cards. I've been a 70 owner for the past 12 years. Went from Amd 5870 to 670, 770, 970 and now 1070. For the most part the price always made sense. It went up but never significantly outside of normal inflation and i was always able to sell the old card for a decent price and upgrade for a respectable amount of money.

I paid my 1070 Strix 570 CAD almost two years ago. Inflation is usually in the range of 2-3% a year. Canadian $ back then was at 0.78 and it's now at 0.76 so not much difference there. So a 70 card now should be roughly 600$ CAD. The cheapest (non open box) 3 fans 2070 on sale right now are around 680$ CAD which is 80$ over what it should technically be.

So now you have the 2060 as mid range and a bunch of cards as high end? Doesn't make any sense to me. Mid range is supposed to be in the middle not at the bottom.