Some EVGA RTX 2060 KO models have RTX 2080 guts

onetheycallEric

TS Addict
Staff member

In their review of the EVGA RTX 2060 KO, TechPowerUp were pleasantly surprised to discover the presence of a TU104 die once the card was disassembled. Specifically, a TU104-150.

Nvidia's TU104 silicon is what powers the RTX 2070 Super, RTX 2080, and RTX 2080 Super. Being that a fully equipped TU104 die contains 3,072 CUDA cores, and the RTX 2060 runs 1,920 CUDA cores, we can infer that these TU104 dies are likely those that didn't make the cut for the higher-end cards.

Some yields are better than others, that's just the nature of lithography. Thus, Nvidia has to cherry pick the best chips for high-performing SKUs. However, chips that don't make the cut for more powerful models are free to be recycled for those that don't require a fully capable die. This allows Nvidia to profit from otherwise unused silicon, and it gives AIB partners more flexibility in pricing.

There shouldn't be any meaningful difference in gaming performance between a model endowed with the TU104 die versus one with the TU106 die. However, there are reports of improved workstation performance, such as Blender, if that's of interest. TechPowerUp also reports that the TU104 chip does drink a bit more power, too.

Being that there will be a mix of cards with both TU104 and TU106 dies, there will be no real way for consumers to discern which one they're getting prior to purchase.

Permalink to story.

 

QuantumPhysics

TS Evangelist
The vast majority of consumers won’t care.

They’ll be probably getting a lot of these cards in pre-builds.

2060 is still an excellent choice for a $1000 desktop.
 
  • Like
Reactions: CharmsD

veLa

TS Evangelist
It's common binning practice, of course. It reduces waste and allows companies like Nvidia to create a balanced line up. It's just frustrating when it comes down to the luck of the draw.

One situation I remember from back in the day applied to the original Fermi architecture. At the time, GTX 480/470 yields were so low that Nvidia introduced the GTX 465 with GPU dies that didn't make the cut. They ran hot and Nvidia introduced a cooler and better performing GTX 460 down the line. You could find some great deals on the GTX 465 after that though.
 

benderOS

TS Rookie
Same was with 1060. Why they wont do it to all lineups in the future? Save money on project stage by not doing all those *06/*07 chips. Maybe drop prices. Draws 10W more? Who cares?
 
Nvidia has to cherry pick the best chips
There's no evidence to support an inference that Nvidia does any such thing. As far as we know they've just chosen to stamp TU104-150 in/on a chip as a diversion. Some minor form of industrial espionage protection. There's nothing meaningful or factual in the wild, regarding methods Nvidia or any other company employees to produce lower spec chips, just lard-encrusted fantasy facts - grist feeding a speculation mill.
But hey, publish or perish 'cus it may be a good guess which is rilly rilly a righetous milkenberry Gamers Nexuhisss approach to create some newssssss..
/s Lol
 

yRaz

Nigerian Prince
There's no evidence to support an inference that Nvidia does any such thing. As far as we know they've just chosen to stamp TU104-150 in/on a chip as a diversion. Some minor form of industrial espionage protection. There's nothing meaningful or factual in the wild, regarding methods Nvidia or any other company employees to produce lower spec chips, just lard-encrusted fantasy facts - grist feeding a speculation mill.
But hey, publish or perish 'cus it may be a good guess which is rilly rilly a righetous milkenberry Gamers Nexuhisss approach to create some newssssss..
/s Lol
I find that a conspiracy is far more unlikely than them recycling unusable silicon to save money.
 
  • Like
Reactions: Irata and Godel

Uncle Al

TS Evangelist
It's not exactly deceptive, but it is certainly not full disclosure. As long as the performance is the same it doesn't matter to most users but for the company it's a black mark against their integrity in the market place. Kind of like the year they started putting Chevy engines in Cadillac's. While they are both GM products, the simple fact is they were deceptive and that caused all the problems.
 
Let me ask it this way. Wouldn't it be LESS [edit] deceptive to label the same chips configured differently as, say; TU101 - in series like 102, 103, 104... and so on?

Cause that way, there is nothing anyone can identify in common other than die sizes and the specifications they're given.

Avoiding a Chevy Cadillac error is simple, might burn in Asus Build, MSI, EVGA, or just a vendor code to indicate which company purchased a chip. But they're never going to do that.
 

Bronxboi

TS Rookie
Same was with 1060. Why they wont do it to all lineups in the future? Save money on project stage by not doing all those *06/*07 chips. Maybe drop prices. Draws 10W more? Who cares?
I think they are concerned with cannibalizing the next in line product. It would actually eat into the sales of both upper and lower products.
 

PEnnn

TS Maniac
It's all irrelevant since:

1 - It seems it's not in stock anywhere.
2 - Its price (now the ludicrous 1 week promotion is over) is way more / close to better cards with similar pricing.
 

Aries Lyon

TS Rookie
It's not exactly deceptive, but it is certainly not full disclosure. As long as the performance is the same it doesn't matter to most users but for the company it's a black mark against their integrity in the market place. Kind of like the year they started putting Chevy engines in Cadillac's. While they are both GM products, the simple fact is they were deceptive and that caused all the problems.
Your analogy isn't really valid because when someone buys a Cadillac they expect a Cadillac engine, so GM was being deceptive by replacing it with a Chevy one (I don't actually know much about this case, just going off the stuff you listed).

Whereas for the 2060, nobody buys it "expecting" to get a TU106 die. They simply expect a certain level of performance, which both the TU104 and TU106 can deliver, within the margin of error. Don't see how this is being deceptive.
 

yRaz

Nigerian Prince
Your analogy isn't really valid because when someone buys a Cadillac they expect a Cadillac engine, so GM was being deceptive by replacing it with a Chevy one (I don't actually know much about this case, just going off the stuff you listed).

Whereas for the 2060, nobody buys it "expecting" to get a TU106 die. They simply expect a certain level of performance, which both the TU104 and TU106 can deliver, within the margin of error. Don't see how this is being deceptive.
You have no idea what you're talking about. A starter on a Cadillac engine could be manufactured by several different factories to the same specs. Everything from piston rings, head gaskets, rubber seals, ect, are cookie cutter parts. They don't change the part number just because they change the rubber seals 3 months down the road.

You have NO IDEA how common this across all industries. Different packaging, serial numbers all adds to the cost of a product. Companies are not going to spend MORE money on something that is intended as a cost savings measure. All that's required is that the product they are selling performs as the product they are selling.
 

Aries Lyon

TS Rookie
You have no idea what you're talking about. A starter on a Cadillac engine could be manufactured by several different factories to the same specs. Everything from piston rings, head gaskets, rubber seals, ect, are cookie cutter parts. They don't change the part number just because they change the rubber seals 3 months down the road.

You have NO IDEA how common this across all industries. Different packaging, serial numbers all adds to the cost of a product. Companies are not going to spend MORE money on something that is intended as a cost savings measure. All that's required is that the product they are selling performs as the product they are selling.
You misunderstand; we have the same viewpoint. I was trying to refute the claim that Nvidia was somehow being "deceptive" by using cut down TU104 dies instead of TU106. Like I said, I don't know anything about how the engine situation is like, I was just pointing out the logical inconsistency in OP's analogy. I personally don't really care what's going on with Chevy/Cadillac, so chill out
 

yRaz

Nigerian Prince
You misunderstand; we have the same viewpoint. I was trying to refute the claim that Nvidia was somehow being "deceptive" by using cut down TU104 dies instead of TU106. Like I said, I don't know anything about how the engine situation is like, I was just pointing out the logical inconsistency in OP's analogy. I personally don't really care what's going on with Chevy/Cadillac, so chill out
A lot of context is lost in text, so my apologies. It is frustrating to me how little people understand how much their devices change without serial number changes. You usually can only find numbers in build dates or production line numbers that only the manufacturer has access to detailed information on.

You can get boards with the same serial number that has multiple different variants because suppliers for transistors and capacitors can change same day on the production line.
 
The vast majority of consumers won’t care.

They’ll be probably getting a lot of these cards in pre-builds.

2060 is still an excellent choice for a $1000 desktop.
This particular card perhaps not as it seems to use sub par components - 1660 PCB and HSF, cheaper voltage regulation / power delivery, higher power consumption and high noise vs standard 2060 cards...

I mean if you do not mind all this and getting a leftover / landfill edition, go for it.

If a 2060 is to your liking, maybe the extra few $ are better spent on a regular / FE 2060.
 
Last edited:
It's all irrelevant since:

1 - It seems it's not in stock anywhere.
2 - Its price (now the ludicrous 1 week promotion is over) is way more / close to better cards with similar pricing.
Just checked and Newegg has it in stock - for $299 +3.99 shipping. That's still a couple of Dollars cheaper than the next 2060.

Actually, they have a Gigabyte RTX 2060 for $299.99 after $30.00 rebate card with free shipping. But this one has many complaints of being hot and noisy.

Anyhow, it seems like the low price may have been a 5600XT release stunt, I.e. "look here, the 2060 is cheaper, so don't bother looking at the competition".
 
So at least with RTX 2060 KO people can get more than advertised performance, the same can't be said about 5600XT since some models will not get vBios updates. Furthermore AMD gave specific guidance to reviewer on which model to review (only ones with vbios updates). At least GamersNexus and Hardware Unboxed gave their honest opinion on 5600XT launch:

 

Evernessince

地獄らしい人間動物園
So at least with RTX 2060 KO people can get more than advertised performance, the same can't be said about 5600XT since some models will not get vBios updates. Furthermore AMD gave specific guidance to reviewer on which model to review (only ones with vbios updates). At least GamersNexus and Hardware Unboxed gave their honest opinion on 5600XT launch:

1. There are certainly going to be 5600 XTs that go past stock clocks so in fact you can get better then stock performance. Do you honestly think there won't be vendors who push the clocks higher? It always happens with every card.

2. I don't see how a single 2060 model applies to all 2060s. What you are doing here is cherry picking. In addition, there isn't even any reports of card with improved gaming performance beyond that which the silicon lottery already allows. You are basing your opinion not on fact or actual accounts but an assumption.

3. If by honest you mean giving it an 85 / 100 then yes. I don't know where you were going with this point. The same person who does HWUB's reviews also does TechSpot's. The title for the 5600 XT review is copied verbatim. Somehow though I imagine you weren't angling for a positive review despite facts being the contrary.
 
  • Like
Reactions: Irata
1. There are certainly going to be 5600 XTs that go past stock clocks so in fact you can get better then stock performance. Do you honestly think there won't be vendors who push the clocks higher? It always happens with every card.

2. I don't see how a single 2060 model applies to all 2060s. What you are doing here is cherry picking. In addition, there isn't even any reports of card with improved gaming performance beyond that which the silicon lottery already allows. You are basing your opinion not on fact or actual accounts but an assumption.

3. If by honest you mean giving it an 85 / 100 then yes. I don't know where you were going with this point. The same person who does HWUB's reviews also does TechSpot's. The title for the 5600 XT review is copied verbatim. Somehow though I imagine you weren't angling for a positive review despite facts being the contrary.

1. Here is how Tim thought about 5600XT launch, so some models from Asrock, MSI and Asus will not received vbios that increase the memory speed, only the core clock speed (Msi Gaming X and Asus Tuf will remain 12gbps even when flashed with new bios). The performance delta between the ones with 14gbps bios to the ones with 12gbps can be up to 13%. Guess which model AMD specified reviewers to benched at this point ? (hint: only the ones that have the new 14gbps vbios). The unlucky or uninformed buyers could get 5600XT that is 10% slower than what they expect, that is beyond any simple silicon lottery or inadequate cooling.

2. I was talking about how some 2060 KO have TU 104 chip and that have improved workstation performance (up to +47%), so if you are lucky to get them, that's good, if not you have a normal 2060.

3. How do you feel if the Sapphire Pulse model is not available in your nearest local store and you bought the Gigabyte Windforce, Msi Gaming X, Asus Tuf 5600XT model that are up to 10% slower ? Even if you are lucky to get the Sapphire Pulse, there are higher chances they might fail, remember that manufacturers did not validated their early batches with increased memory clock speed even if they have 14gbps memory. All 1660 Ti have 14gbps memory but run at 12gbps, 2080Super have 16gbps memory running at 15.5gbps (check TPU pcb shots).
 
Last edited: