ASUS RTX 2080 Turbo (Baseline) Cards Come with TU104 "A" Chips

[btarunr]

The GeForce RTX 2080 Turbo from ASUS is supposed to be a "baseline" RTX 2080 product, which the company can sell at $699, or closest to it. These boards were found to feature the TU104-400A-A1 variant of the TU104 silicon, which NVIDIA allows its add-in card (AIC) partners to ship factory-overclocked speeds with. At this point it's not known if all ASUS RTX 2080 Turbo cards feature the "A" variant TU104 chips, or if it's a lottery. Given that the ASUS RTX 2080 Turbo's PCB is largely based on NVIDIA's reference design, PC Games Hardware (PCGH) has been able to successfully flash the card's BIOS with that of the RTX 2080 Founders Edition cards based on the reference PCB, which have power-limits increased to the tune of 307 W, which facilitates not just higher GPU Boost frequencies, but also better sustainability of elevated boost clock states.

With its "Turing" family of GPUs, NVIDIA created ASIC variants along the lines of chips that board partners are allowed to factory-overclock, and those that they aren't. You can read all about that in our older article. Normally, the TU104-400-A1 silicon is intended for baseline cards such as the ASUS RTX 2080 Turbo, whereas the TU104-400A-A1 goes into factory-overclocked products such as ASUS RTX 2080 ROG Strix. The discovery of TU104-400A-A1 on the ASUS RTX 2080 Turbo makes it the cheapest option for enthusiasts wanting to flash it with BIOS of other reference-PCB based cards that have TU104-400A-A1 chips to increase power limits, and then simply pairing the card with custom liquid cooling, to manually overclock further, thanks to the increased power limits. We're not sure you can flash Founders Edition BIOS on cards that have reference-design PCBs but non-A ASICs.



View at TechPowerUp Main Site

 

[lexluthermiester]

At this point it's not known if all ASUS RTX 2080 Turbo cards feature the "A" variant TU104 chips, or if it's a lottery.

Very likely lottery, unless ASUS specifically selected these.

 

[intelzen]

now one can not even tell what specs he will get by reading fine prints? is it a hardware loot-box system? next step - ddr4 versions mixed in?

 

[londiste]

Card specs say it comes with OC mode which likely means ASUS can only use 400-A chips.
Has it even been shown if A and non-A chips are actually different? OC overhead on RTX GPUs is not that significant to begin with.

 

[ShurikN]

now one can not even tell what specs he will get by reading fine prints? is it a hardware loot-box system? next step - ddr4 versions mixed in?

You will get what you pay for and what is advertised. If Asus is selling and marketing the card with X base clock, Y boost clock and Z mem clock, that is what you get, whether it's 400A, B, C or omega. Everything else is OC related by the customer and therefore not the GPU makers concern.

Card specs say it comes with OC mode which likely means ASUS can only use 400-A chips.
Has it even been shown if A and non-A chips are actually different? OC overhead on RTX GPUs is not that significant to begin with.

GN did a piece on 106-400, A and non A. A has a higher power limit, and can hit (if I remember correctly) 50MHz more when OCing.

 

[silentbogo]

GN did a piece on 106-400, A and non A. A has a higher power limit, and can hit (if I remember correctly) 50MHz more when OCing.

I was just reading that piece. Might be very informative in this context (even though it's related to TU106)
https://www.gamersnexus.net/hwreviews/3386-nvidias-secret-gpu-tu106-400-vs-400a-2070-xc-ultra-review

 

[londiste]

GN did a piece on 106-400, A and non A. A has a higher power limit, and can hit (if I remember correctly) 50MHz more when OCing.

Well, their piece is somewhat crap. Cards GN was playing with were on 2080 and 2070 PCB respectively, VRM temperatures had a noticeable difference and some comparisons were skewed by XC Ultra running at about 100MHz OC out-of-box.

I would really like to see someone put both under water with fullcover block and then lets see what happens, keeping temperatures the same. Cooling matters a lot for Turing cards (as it did for Pascal). Boost 3.0/4.0 scales clock bins all the way down to 30C. And that is just the core.

The only valid difference seems to be the power limit, usefulness of which is a bit suspect. In all the Pascal and Turing cards I have had chance to play with the useful power limit is somehere around +15% (from spec), after that cards will simply run into voltage limit more often than not. Besides, 50MHz is well within normal variance of chips :/

 

[TheLostSwede]

Is any of this even relevant as long as the cards cost what they cost? The cheapest cards in my neck of the woods cost around US$855, with the more expensive "premium" cards going for as much as $985. The Ti's are listed between US$1,260 and and $1,320. Who in their right mind would buy any of these cards? Nvidia has really lost their mind when it comes to the pricing of their products. Note that the high prices are not due to high sales tax, as that's only 5%.

Also, from my understanding, the blower cards are popular in Asia for some reason and are mostly made for that market.

 

[stimpy88]

Yet another disgusting anti-customer strategy from nGreedia.

 

[londiste]

Is any of this even relevant as long as the cards cost what they cost? The cheapest cards in my neck of the woods cost around US$855, with the more expensive "premium" cards going for as much as $985. The Ti's are listed between US$1,260 and and $1,320. Who in their right mind would buy any of these cards? Nvidia has really lost their mind when it comes to the pricing of their products. Note that the high prices are not due to high sales tax, as that's only 5%.

Well, it depends on where you are.
In Europe, cheapest GTX 1080Ti and RTX 2080 are within 30-40€ of each other, aftermarket cooling models are in the same price range. Same story with GTX 1080 and RTX 2070. All with cards in stock.
Situation with RTX 2080Ti is completely fucked up everywhere it appears. In addition to it being in a class of its own and still at 1100+€ there are clearly yield problems and some dying problems.

Regarding the A vs non-A, most manufacturers seem to indicate the chip type clearly in specs. I would put much more value to the cooler slapped on the card rather than the chip variation.

 

[stimpy88]

Well, it depends on where you are.
In Europe, cheapest GTX 1080Ti and RTX 2080 are within 30-40€ of each other, aftermarket cooling models are in the same price range. Same story with GTX 1080 and RTX 2070. All with cards in stock.
Situation with RTX 2080Ti is completely fucked up everywhere it appears. In addition to it being in a class of its own and still at 1100+€ there are clearly yield problems and some dying problems.

Regarding the A vs non-A, most manufacturers seem to indicate the chip type clearly in specs. I would put much more value to the cooler slapped on the card rather than the chip variation.

This is quite the deliberate nGreedia strategy. They are convinced they are Apple. They will not sell their old crap and lose money, hence the steep prices for these old cards. Look for a performance driver for the RTX cards out in the next 2 months, just after the 10 series cards supply dry up.

We know that nGreedia made far to many 10 series chips 6 months ago, and the AIBs didn't want them, as the mining craze had stopped and nobody wanted these cards for sky high money, this situation is exactly what nGreedia wanted. Artificially high prices for 3 year old crap, and a new series of cards with magically no software support, want to bet that situation changes just before the 10 series stock dries up?

 

[TheLostSwede]

This is quite the deliberate nGreedia strategy. They are convinced they are Apple. They will not sell their old crap and lose money, hence the steep prices for these old cards. Look for a performance driver for the RTX cards out in the next 2 months, just after the 10 series cards supply dry up. We know that nGreedia made far to many 10 series chips 6 months ago, and the AIBs didn't want them, as the mining craze had stopped and nobody wanted these cards for sky high money, this situation is exactly what nGreedia wanted. Artificially high prices for 3 year old crap, and a new series of cards with no software support, want to bet that situation changes just before the 10 series stock dries up?

Here all the 1080 variant cards disappeared off the market over night. Not a single card available for purchase. The best you can get is an over priced 1070 Ti. I guess that's also one way to force customers to buy the new cards. I heard that some of the manufacturers are buying up the last stock of 1080 chips, since there's still a demand, so there might be more cards hitting the market in a couple of months, so let's wait and a see.

 

[stimpy88]

Here all the 1080 variant cards disappeared off the market over night. Not a single card available for purchase. The best you can get is an over priced 1070 Ti. I guess that's also one way to force customers to buy the new cards. I heard that some of the manufacturers are buying up the last stock of 1080 chips, since there's still a demand, so there might be more cards hitting the market in a couple of months, so let's wait and a see.

nGreedia pulled them from markets that were not so profitable, and moved them to profitable ones instead. We all had hopes of cheap 10 series cards after the real and disappointing performance numbers and prices of the RTX range were released, nGreedia moved to stop that instantly. This is why there is no RTX software yet, it's to ensure they sell that pile of old expensive GPUs they made far too many of.

 

[londiste]

It is not logical to think Nvidia profits more from 20-series RTX cards than 10-series GTX cards at even remotely similar prices.
Bigger chips, more expensive memory, more complex PCB (as MSI interview revealed).

 

[lexluthermiester]

Yet another disgusting anti-customer strategy from ASUS.

Fixed that for you. If you're going to bash a company, make sure it's the right one.

 

[stimpy88]

Fixed that for you. If you're going to bash a company, make sure it's the right one.

Ahh crap, you're right! Asus must have got the 104 chip design from nGreedia and redesigned it better than nGreedia, printed an A on the chip so that nGreedia wouldn't figure out that Asus are better at designing chips than they are!

Yep, you fixed it for me alright!

 

[lexluthermiester]

Yep, you fixed it for me alright!

ASUS selected and bought the chips from NVidia to install on their own cards. The only role NVidia played was providing the chips. If you're gong to be a troll, at least do so with some measure of competency.

 

[stimpy88]

ASUS selected and bought the chips from NVidia to install on their own cards. The only role NVidia played was providing the chips. If you're gong to be a troll, at least do so with some measure of competency.

OK, why do nGreedia even make an A series chip in the first place? The fact that these chips exist is going to be the thing everyone buying an RTX card will worry about. Do they have one or not? The next thing will be that cards with these A chips will be more expensive, rising the prices even higher than they already are.

Listen, my nGreedia shill, I'm not happy over what your company is doing to PC gaming, and you will not stop me from calling it out when I see it. I'm proud to not be an idiot that enjoys being ripped off, and seeing my hobbies go down the drain far enough that only the rich kids can play it. If that makes me a troll, then it's your problem, not mine.

 

[rtwjunkie]

is it a hardware loot-box system?

Love this line!!

 

[R0H1T]

It is not logical to think Nvidia profits more from 20-series RTX cards than 10-series GTX cards at even remotely similar prices.
Bigger chips, more expensive memory, more complex PCB (as MSI interview revealed).

So do you have a BOM for RTX cards or for that matter profit margins on them?

 

[londiste]

So do you have a BOM for RTX cards or for that matter profit margins on them?

https://www.techpowerup.com/gpu-specs/geforce-gtx-1080.c2839
https://www.techpowerup.com/gpu-specs/geforce-rtx-2070.c3252

1. Dies: 7.2 billion transistors and 314 mm² vs 10.8 billion transistors and 445 mm². 40% bigger chip, more expensive to produce.
2. VRAM: 8GB of GDDR5X on 256-bit bus vs 8GB of GDDR6 on 256-bit bus. GDDR6 is at least the same price as GDDR5X, very likely more.
3. PCB:

RTX 2080 Ti using some 2600 components compared to the GTX 1080 Ti at 1600. For further comparison, it is said the RTX 2080 uses 2400 components and the RTX 2070 some 2200.

GTX 1080 has less complex PCB than GTX 1080Ti and definitely less components. RTX 2070 has more components on it than GTX 1080Ti.

Does not make sense to get a bigger profit margin from this. Memory may be so-so but if MSI was even remotely truthful BOM of card must be lower for 10-series. When it comes to GPUs themselves, even discounting the considerable size increase (and less GPUs per wafer due to that) at these sizes yields start becoming a problem.

 

[R0H1T]

https://www.techpowerup.com/gpu-specs/geforce-gtx-1080.c2839
https://www.techpowerup.com/gpu-specs/geforce-rtx-2070.c3252

1. Dies: 7.2 billion transistors and 314 mm² vs 10.8 billion transistors and 445 mm². 40% bigger chip, more expensive to produce.
2. VRAM: 8GB of GDDR5X on 256-bit bus vs 8GB of GDDR6 on 256-bit bus. GDDR6 is at least the same price as GDDR5X, very likely more.
3. PCB:GTX 1080 has less complex PCB than GTX 1080Ti and definitely less components. RTX 2070 has more components on it than GTX 1080Ti.

Does not make sense to get a bigger profit margin from this. Memory may be so-so but if MSI was even remotely truthful BOM of card must be lower for 10-series. When it comes to GPUs themselves, even discounting the considerable size increase (and less GPUs per wafer due to that) at these sizes yields start becoming a problem.

Smaller node, yields unknown but given the huge jump from 1080Ti (die size) it is reasonable to expect higher cost, how much ~ anyone's guess.

GDDR6 is backed by all 3 leading makers of memory, so no I expect GDDR5X to cost more, given similar yields & lower overall supply for the latter.

You're forgetting the obvious elephant in the room, price ~ how much does the RTX sell for?

 

[londiste]

Smaller node, yields unknown but given the huge jump from 1080Ti (die size) it is reasonable to expect higher cost, how much ~ anyone's guess.
You're forgetting the obvious elephant in the room, price ~ how much does the RTX sell for?

12nm is not a smaller node. It is in the same 16nm/14nm/12nm family. These are all pretty much the same. Even if you look at GTX 1080 and RTX 2070 GPUs, these come out to 23-24 MT/mm².
As I said above, at least in Europe, GTX 1080 and RTX 2070 sell for the same price. Same story with GTX 1080Ti and RTX 2080.

 

[Robcostyle]

I must admit, this is the only good news, among all concerning Turing and ngreedia for the past 3 months

First, because these GPU's are so retardly expensive, I wouldn't even doubt either building custom loop or not anyway.
And the fact, that Asus puts best "A" chips into the cheapest and reference line makes the choice for loop between cards even easier.
Third, my 1080ti strix has an awfull coil sounds, so...

It's just good that I will have a premium option for waterblock, for lower price. (If I'll ever get nuts and buy one for myself)

 

[Vayra86]

I must admit, this is the only good news, among all concerning Turing and ngreedia for the past 3 months

First, because these GPU's are so retardly expensive, I wouldn't even doubt either building custom loop or not anyway.
And the fact, that Asus puts best "A" chips into the cheapest and reference line makes the choice for loop between cards even easier.
Third, my 1080ti strix has an awfull coil sounds, so...

It's just good that I will have a premium option for waterblock, for lower price. (If I'll ever get nuts and buy one for myself)

So what, you're going to 'upgrade' from 1080ti to a 2080? That is a pretty epic waste of time