Make your own 6800 Ultra out of a 6800 GT?

[Jeff7181]

Well, recent events and a recent topic in the Video forums have got me curious.

The 6800 GT has a single power connector, it's own BIOS, and a single slot cooling solution.

The 6800 Ultra has two power connectors, it's own BIOS, and a dual slot cooling solution.


My friend recently knocked a capacitor off his 6800 GT and it still works. I'm convinced it's just a capacitor to help even out power coming in the molex connector... there's two of them, one appears to be for the 12V line (rated at 16 V) and I assume the other one is for the 5V line. There's the outline for the 2nd power connector, and two more capacitors. What if you got your hands on the same type of capacitors, and modified a power connector to fit on the board... then flashed the BIOS to a 6800 Ultra?

If the RAM is of the same type, and the GPU is the same... it should work right? Then you could modify maybe a VGA Silencer to fit the 6800.

Then you have a 6800 Ultra for $400 or less.

What do ya think? Is it possible?

 

[imported_Nail]

Capacitors don't control the speed of the GPU.

And you won't have an Ultra. A GT might clock to Ultra speeds, but an Ultra can clock even faster.

 

[Jeff7181]

Originally posted by: Nail
Capacitors don't control the speed of the GPU.

And you won't have an Ultra. A GT might clock to Ultra speeds, but an Ultra can clock even faster.

I don't recall saying capacitors have anything to do with the GPU. Did you even read what I said?

Maybe the reason the Ultra can be overclocked higher is because it's getting more stable power from two power connectors and a BIOS that allows more voltage to the GPU and RAM.

 

[imported_Nail]

Maybe the reason the Ultra can be overclocked higher is because it's getting more stable power from two power connectors and a BIOS that allows more voltage to the GPU and RAM.

Nope. The chip is physically different. Power supply isn't the reason for the price difference.

 

[Jeff7181]

Originally posted by: Nail

Maybe the reason the Ultra can be overclocked higher is because it's getting more stable power from two power connectors and a BIOS that allows more voltage to the GPU and RAM.

Nope. The chip is physically different. Power supply isn't the reason for the price difference.

Call me crazy, but wouldn't it be horribly inefficient and expensivev for them to make physically different chips? I mean... they're both NV40 GPU's... they both have 16 pipelines... they both have all the same features, the Ultra just runs faster and has two power connectors. Similar to how the 9800 Pro based on the R360 core can have the BIOS flashed to effectively make it a 9800XT... and similar to how people took x800 Pro's and enabled the other 4 pipelines and then overclocked it to effectively make it an x800XT.

 

[imported_Nail]

wouldn't it be horribly inefficient and expensivev for them to make physically different chips?

The chips are different due to imperfections introduced in the manufacturing process. The defective chips have pipelines disabled or clock speed reduced in order to operate reliably.

 

[Jeff7181]

Originally posted by: Nail

wouldn't it be horribly inefficient and expensivev for them to make physically different chips?

The chips are different due to imperfections introduced in the manufacturing process. The defective chips have pipelines disabled or clock speed reduced in order to operate reliably.

That's true... however, it's also true that manufacturers are known to "speed bin" chips based on demand. If they have orders for 5000 6800GT's, orders for 1000 6800 Ultras... and they have 2000 chips capable of being a 6800 Ultra, and 4000 chips only capable of being a 6800 GT... then 1000 GT's will have chips capable of running at Ultra speeds as long as the cooling and power requirements are met.

 

[CTho9305]

Originally posted by: Jeff7181

Originally posted by: Nail

wouldn't it be horribly inefficient and expensivev for them to make physically different chips?

The chips are different due to imperfections introduced in the manufacturing process. The defective chips have pipelines disabled or clock speed reduced in order to operate reliably.

That's true... however, it's also true that manufacturers are known to "speed bin" chips based on demand. If they have orders for 5000 6800GT's, orders for 1000 6800 Ultras... and they have 2000 chips capable of being a 6800 Ultra, and 4000 chips only capable of being a 6800 GT... then 1000 GT's will have chips capable of running at Ultra speeds as long as the cooling and power requirements are met.

I've never heard someone I'd consider "reliable" (someone who works in industry) say that. Even so, right now, I hear ATI can't meet demand on their high end parts due to lack of high-speed memory, so it's unlikely they would be downbinning RAM right now (if they ever do...)

 

[Jeff7181]

Originally posted by: CTho9305

Originally posted by: Jeff7181

Originally posted by: Nail

wouldn't it be horribly inefficient and expensivev for them to make physically different chips?

The chips are different due to imperfections introduced in the manufacturing process. The defective chips have pipelines disabled or clock speed reduced in order to operate reliably.

That's true... however, it's also true that manufacturers are known to "speed bin" chips based on demand. If they have orders for 5000 6800GT's, orders for 1000 6800 Ultras... and they have 2000 chips capable of being a 6800 Ultra, and 4000 chips only capable of being a 6800 GT... then 1000 GT's will have chips capable of running at Ultra speeds as long as the cooling and power requirements are met.

I've never heard someone I'd consider "reliable" (someone who works in industry) say that. Even so, right now, I hear ATI can't meet demand on their high end parts due to lack of high-speed memory, so it's unlikely they would be downbinning RAM right now (if they ever do...)

What happens if yields improve and almost all the TNT2 chips are capable of reaching Ultra speeds, but there isn't enough demand for all the Ultra chips? Simple, you just sell the excess Ultra tested chips as normal TNT2s.

Old, but still holds true. If it's not true, explain why I've had 6 different XP2500's running in my motherboard at 2.2 GHz (or higher) on default voltage.

Anyway... I didn't start this thread to argue about speed binning... if you're saying my theory in the OP won't work, explain why... especially since MANY 6800 GT's are overclocking to 6800 Ultra speeds and in some cases beyond.

 

[CTho9305]

Originally posted by: Jeff7181

Originally posted by: CTho9305
I've never heard someone I'd consider "reliable" (someone who works in industry) say that. Even so, right now, I hear ATI can't meet demand on their high end parts due to lack of high-speed memory, so it's unlikely they would be downbinning RAM right now (if they ever do...)

What happens if yields improve and almost all the TNT2 chips are capable of reaching Ultra speeds, but there isn't enough demand for all the Ultra chips? Simple, you just sell the excess Ultra tested chips as normal TNT2s.

Old, but still holds true.

Where does that author cite someone in industry? He simply states it as fact.

If it's not true, explain why I've had 6 different XP2500's running in my motherboard at 2.2 GHz (or higher) on default voltage.

See my explanation here of what you're doing when you're overclocking.

The datasheet says you can run at .05V below the rated voltage, with drops to .1v below rated voltage for up to 5 microseconds at a time, and at a die temperature of 85C. You're probably providing better power (allows you to clock higher) and running cooler (allows you to clock higher). Also, your "stability tests" aren't nearly as good as the tests Intel and AMD can run, for a couple of reasons.

1) Scan in / out. There is a way to set the state of every storage node in a CPU and run it from there, then check every storage node. This gives nearly 100% fault coverage detection. If you do the test at full speed, it will show any critical paths. The manufacturer knows what values to set where to find the worst paths. As a consumer, you can't even use scan.

2) "At-speed test patterns". Since the manufacturer knows all the circuits in the design, they can come up with "test patterns" (code sequences) to exercise all the worst-case paths. Things like prime95, sandra's burn-in, and 3dmark are often used by overclockers to check stability, but they only hit a FEW paths. If you consider the example of a 32-bit adder, there are 4 billion * 4 billion input combinations. When the manufacturer tests the adder, they'll pick the known-slowest patterns. When you run a stability testing app, you might not hit a slow pattern, and THINK your system is stable even though certain operations still produce incorrect results. Even manufacturer's test patterns don't provide coverage as good as scan in/out.

What it comes down to is, I don't believe some random Firingsquad author who says companies down bin. It seems to me that if I could produce faster CPUs, I'd just drop the prices and make my competitor suffer. If pm or wingznut said Intel downbins, I'd believe it. Even if they do downbin, unless you KNOW what speed the CPU passed tests at, you're running the risk of silent data corruption. Once you overclock, you can no longer be sure if a once-a-month BSOD is a driver issue or a hardware issue. Same for random application crashes. Is xyz game poorly written, or is your CPU calculating data incorrectly?

Anyway... I didn't start this thread to argue about speed binning... if you're saying my theory in the OP won't work, explain why... especially since MANY 6800 GT's are overclocking to 6800 Ultra speeds and in some cases beyond.

I don't know what the difference between the cards is. Nail claims the chips are different. While this may not be true, it's possible they're doing what ATI does - fab 16 pipelines and tell the GPU to disable 4. Either way, I would think it's most likely more than a "better power" change.

 

[DarkSarkas]

Well, the 6800 GT has 16 pipelines. So does the Ultra. Technically, they could run at the same speed, assuming that several different conditions are met:

A. The fabrication process on which they are produced is tried and true. (This means higher yields and a better chance of getting a decent chip as opposed to a "this just barely made the mhz cut" chip).

B. The RAM is overclockable. This is much less likely due to the state of the DRAM market right now.

C. Voltage is added, assuming that A didn't pan out too well.

The two chips themselves are the same. Google will confirm this is you'd like. Chances are, though, because of the problems that Nvidia is having with the process and getting viable 6800 Ultras (hence the high price and low availability) that there won't be many 6800 GTs out there capable of running at Ultra speeds out of the box.

 

[PacFu]

Most of what you are saying is right, Jeff, except the ultra uses a bit faster ram...correct me if I'm wrong but ultra uses 1.7ns and GT is 2.2ns. You may have to trip some switches on the PCB as well. I dunno, good thought though.

 

[Wahsapa]

chances are its not "just missing the capacitors that make it an ultra" but im sure their are other things you need. have you ever made your own video card before? if you think this might work, do it and tell us how it goes.