Fiji Core overclocking

[mirakul]

I think you forgot about all of the circuitry connecting those logic gates and how logic gates tend to produce heat when they're switching on or off, when resistances is changing. You can run a transistor so it's partially on and it almost acts like a resistor, creating heat all the while. This is why PWM is used to create buck converters because PWM ensures the transistor is fully on or off which reduces the amount of leakage from the transistor because it spends less time in a partially on/off state which has resistance. Also, I think you're forgetting about the wires that connect all of those logic gates together too. Smaller wires also means higher resistance/impedance.

A transistor fully turned "on" has very little resistance, depending on the type of transistor. Some or better than others, but the one thing that's true for all of them is that when they're not fully on or off, their resistance is higher than when it's fully turned on. It's the nature of transistors.

I didn't forget them. The heat produced by them is much less than by the logic gates so I didn't mention it.
And yes, logic gates produce heat when they switch between on and off, their only operation though.
I would recommend you to start doing some research about what happens inside a die, mate.

 

[dorsetknob]

@mirakul


in the 1st post by the OP

Thoughts and technical musings appreciated. This is for tech discussion on Fury (Fiji) only - there is NO need to mention Nvidia in any way shape or form - the topic is about Fiji and Fiji only.

and in your first post you make you include this

Nice
wonder if you can read as well as you post

 

[Aquinus]

I didn't forget them. The heat produced by them is much less than by the logic gates so I didn't mention it.
And yes, logic gates produce heat when they switch between on and off, their only operation though.
I would recommend you to start doing some research about what happens inside a die, mate.

For someone has hypocritical as you as @dorsetknob pointed out, I don't need to prove anything to you. I don't need to learn what I already know and I have a job writing software and a degree in Comp Sci to show for it. How about you go back under whatever bridge you came out from under. Not all logic gates are clocked every cycle. You need to stop acting like a smart ass who knows everything when all you're doing is pissing people off.

 

[mirakul]

@dorsetknob Mate, it's clear that I didn't mention the sacred word or say anything about a comparison. Use your flamebait elsewhere, please.

For someone has hypocritical as you as @dorsetknob pointed out, I don't need to prove anything to you. I don't need to learn what I already know and I have a job writing software and a degree in Comp Sci to show for it. How about you go back under whatever bridge you came out from under. Not all logic gates are clocked every cycle. You need to stop acting like a smart ass who knows everything when all you're doing is pissing people off.

Well, so you are still defending your point that most of the heat in CPU come from wire?
Okaaaaaaaaaaaaaaaaaaay. You should apply for a Nobel prize in Physics.
Another day on TPU.

 

[Aquinus]

Well, so you are still defending your point that most of the heat in CPU come from wire?
Okaaaaaaaaaaaaaaaaaaay. You should apply for a Nobel prize in Physics.

No, I'm saying that it plays a factor and you can't know how much of a factor it is without knowing some very detail information about the wiring in the interposer. You've gotten so far off topic yourself that you've forgotten that we're talking about the interposer which has no logic gates in it...

 

[dorsetknob]

@dorsetknob Mate, it's clear that I didn't mention the sacred word or say anything about a comparison. Use your flamebait elsewhere, please.

NO need to mention Nvidia in any way shape or form
that i believe also covers pictures Nvidia in any way shape or form

If you had the manners to comply with the OP's request then we would not be having this chat.

Edit your no mate of mine i chose my friends and mate's carefully

 

[mirakul]

No, I'm saying that it plays a factor and you can't know how much of a factor it is without knowing some very detail information about the wiring in the interposer. You've gotten so far off topic yourself that you've forgotten that we're talking about the interposer which has no logic gates in it...

Bad on my side it seems.
Back to the topic

This would clear the speculation of 75C wall of HBM and interposer

 

[the54thvoid]

someone on the guru3d board said he was able to get 0- to +100mv over voltage but it was making the card drop to 2d clocks
and unwinder said he suspects it is indeed a hardware limit
if there is a problem it would make sense for AMD to have a fail-safe should the voltage go out of spec either that or the driver power table is broken

Steering the ship back on track - this looks more like the direction the over volt conundrum will take. Bad news if it is at the max. As you said before @OneMoar, if there is a hardware safety limit and it is bypassed, there will be bricked cards.

 

[OneMoar]

*quote from guru3d user "lucky_noob" and unwinder
Doing i2c dump indeed cause black screen crash. But if we scan some specific areas we can get the values:


Scanning GPU VEN_1002&DEV_7300&SUBSYS_0B361002&REV_C8&BUS_3&DEV _0&FN_0...
Scanning I2C bus 0...
Probing device 30...
Invalid device
Scanning I2C bus 1...
Probing device 30...
Invalid device
Scanning I2C bus 2...
Probing device 30...
Invalid device
Scanning I2C bus 3...
Probing device 30...
Invalid device
Scanning I2C bus 4...
Probing device 30...
Invalid device
Scanning I2C bus 5...
Probing device 30...
Invalid device
Scanning I2C bus 6...
Probing device 30...
29 14 1B 0F 16 1E 7C 49 4A 7D 11 66 66 44 FF FF
A2 22 80 10 29 FF 20 00 00 00 00 00 00 00 00 00
00 00 60 60 1C 1E FF 8D 3C 26 A3 31 38 A1 1F 31
38 07 FC 74 6F 05 05 1D 81 A0 60 00 00 66 00 00
A8 90 00 00 00 00 3C 00 14 14 00 40 80 60 60 FF
FF 00 40 88 34 88 44 12 02 02 88 00 20 20 00 3C
3C 03 7D 61 3C 9C 24 88 80 00 00 00 00 FF 06 FF
FF 20 00 00 00 00 00 00 00 00 00 00 15 15 00 00
00 00 3C 00 00 00 00 00 88 88 01 C2 44 00 00 1F
00 05 43 68 28 01 30 70 5E A8 73 A6 02 3C 1E 1F
00 00 00 00 00 00 01 01 08 55 A0 8F CF 00 00 05
04 00 00 00 00 FF 23 C3 00 00 7A 6E 9E 15 3C 00
00 00 00 77 13 00 00 00 00 00 00 03 00 44 00 00
00 00 AD 50 0C 00 10 00 3F 00 00 00 23 C0 F0 1E
00 1F 00 10 01 03 01 88 00 00 00 00 3C 00 00 00
00 00 00 00 00 33 00 00 00 00 00 00 00 00 00 00
Scanning I2C bus 7...
Probing device 30...
29 14 1B 0F 16 1E 7C 49 4A 7D 11 66 66 44 FF FF
A2 22 80 10 29 FF 20 00 00 00 00 00 00 00 00 00
00 00 60 60 1C 1E FF 8D 00 26 A3 31 38 A1 1F 31
38 07 FC 74 6F 05 05 1D 81 A0 60 00 00 66 3C 00
A8 90 00 00 00 00 00 00 14 14 00 40 80 60 60 FF
FF 00 40 88 34 3C 44 12 02 02 88 00 20 20 00 3C
3C 03 7D 61 68 9C 24 88 80 00 00 00 00 FF 06 FF
3C 20 00 00 00 00 00 00 00 00 00 00 15 15 00 00
00 00 00 00 00 00 00 00 88 88 01 3C 44 00 00 1F
00 05 43 68 28 01 F0 70 5E A8 73 A6 02 01 1E 1F
00 00 00 00 00 00 01 01 08 55 A0 8F 3C 00 00 05
06 00 00 00 00 FF 23 C3 00 00 7A 6E A9 15 00 00
00 00 FF 77 13 00 00 00 00 00 00 03 01 3C 00 00
00 00 AD 50 0C 00 10 00 3F 00 00 00 23 C0 F0 00
00 00 00 10 01 03 01 88 00 00 00 00 00 00 3C 00
00 00 00 00 00 33 00 00 00 00 00 00 00 00 00 00



Probing with /ri6, 30, 92 (or ri7,30,92), gives 43 response, so I'm trying to use the third party voltage control method and made msiafterburner.oem2 file consist on these lines:

;OEM
[VEN_1002&DEV_7300&SUBSYS_0B361002&REV_C8]
VDDC_IR3567B_Detection=6:30h
VDDC_IR3567B_Type= 1
VDDC_IR3567B_Output= 0

And the voltage control shows a value that can be controlled up to +100mV *sorry can't post image/link yet*, but every time I alter those voltages, the Card frequencies are sometimes stuck in 300Mhz even with 3D load.

I tried the other way using normal voltage control method, adding this line to the hardware profiles using
[Settings]
VDDC_IR3567B_Detection = 30h
VDDC_IR3567B_Output = 0

But the result is the same, the voltage control shows a slider to +100mV but when you set it, the card stuck to 2D Frequencies.

I'll probably do more test when my preordered card is delivered, the one I'm testing with already returned back to AMD.

*/unwinder
That's both positive and negative info. Positive: device 30h in the dump is IR3567B with no doubts, so driver-level I2C access is indeed working and VRM can be accessed on software level. Negative: such reaction on overvolting (graphics card downclocking) smells by hitting some hardware limit, I'm not too optimistic on improving it.

 

[sweet]

why do you think amd shipped the fury x with water cooling AND barred vendors from using there own cooling solutions
I think the interposer simply doesn't tolerate getting over say 75c ... being 65nm that would make sense
when pumps start failing cards are gonna smoke ...

Anandtech have Sapphire OC run at 80C.
http://www.anandtech.com/show/9421/the-amd-radeon-r9-fury-review-feat-sapphire-asus/18

Well, it's not true with silicon. The resistance of a transistor in conduct mode is close to zero. Electronics 101.
Silicon chips produce heat because of logic gates. An interposer with no logic gate will produce close to no heat.

You forgot the wires.

@mirakul


in the 1st post by the OP



and in your first post you make you include this

Nice
wonder if you can read as well as you post

It's hard to find a reviews with AMD only so I don't know why you are so picky, lol

 

[the54thvoid]

It's hard to find a reviews with AMD only so I don't know why you are so picky, lol

Could have censored the 'unwelcome' part! But seriously, it is hard to talk about this without any reference to 'them' but it keeps it more civil when only one person turns up to the fight!

The factors really are:

1 - Is Fiji running at close to it's architectural design core speed/voltage limit?
2 - If so, damn.
3 - If not (preferred option), we need the miracle guys (or gals) to code the software to software over volt.

Unfortunately, the guy who does (Unwinder) hasn't got a Fury yet but the people he is talking too (@OneMoar's post above) aren't giving the best of news back to him.

 

[OneMoar]

oi
a Fury is Not a Fury X so please stop comparing apples to oranges mmkay
if the above statements are true then one could assume that because the Fury is a cut down chip it doesn't draw as much current as the "fury x" thus opening some additional headroom in both the temperature and clock speed department ... O wait thats how it always is ... sigh

 

[TheHunter]

Unwinder said he needs a gpu first (soon apparently), then he will enable support in MSI AB.


So far I saw FuryX up to 1175mhz, usually ~ 1140mhz, Im guessing 1230-1250mhz could be possible with voltage unlock.

 

[ne6togadno]

...
1 - Is Fiji running at close to it's architectural design core speed/voltage limit?
...

i dont think problem is core clocks or hbm clocks. i think interposter and its connection to core and memory is what could cause/cause troubles but i am quite away from electronics to this is just my wild guess.
i'd say lets sit back and relax till ppl with skills (like unwinder) figure out what's going on. interesting weeks are coming.

 

[Ebo]

In its own way I think the Fury is a great card, solid preformance out of the box.

Where I see there can be a problem is that only 2 AIB's have made cards so far, Asus and Sapphire. It still can be because of low volume of GPU's.
People who are in the market for a new highend GFX might consider to buy 1 of those if only they can get their hands on 1, if not the jump to GFX 980 or GFX 980TI sient so hard since they allready are in stock.

Once the smart guys/girls out there find out how to unlock voltage on the card, I dont think we have seen the full potential of R9 Fury line.

 

[the54thvoid]

In its own way I think the Fury is a great card, solid preformance out of the box.

Where I see there can be a problem is that only 2 AIB's have made cards so far, Asus and Sapphire. It still can be because of low volume of GPU's.
People who are in the market for a new highend GFX might consider to buy 1 of those if only they can get their hands on 1, if not the jump to GFX 980 or GFX 980TI sient so hard since they allready are in stock.

Once the smart guys/girls out there find out how to unlock voltage on the card, I dont think we have seen the full potential of R9 Fury line.

No doubt its a great card. And what you say about only 2 partners making cards is true, its a sign of poor supply or worse.
It's not a case though of if the smart guys can figure out how to over clock it. It's if the really smart guys (or the chip designers) can reveal the relationship between core speed, voltage and temperatures.

 

[MrGenius]

I think the interposer simply doesn't tolerate getting over say 75c ... being 65nm that would make sense

No. It doesn't make sense...at all. Quite the opposite. See below.

I doubt that HMB operates at 1.3V

It does in fact operate @ 1.3V. You later proved that. But failed to edit this post. Classic.

its an insulator and an interconnect so what happens when any bit of silicon gets hot .. it starts losing its ability to insulate
see page two of this
https://www.amd.com/Documents/High-Bandwidth-Memory-HBM.pdf
so looking at that image what happens if that interposer gets hot and starts leaking voltage
HMB's interconnect/circuit density is much much tighter then normal GDDR5
and I suspect those TSv's aren't gonna tolerate high temps/more amperage very well at all ..

1. There is no page 2.
2. There's that proof I spoke of though.
3. The interposer is 65nm...so increased leakage due to heat or not would be minimized. Not accentuated. Bigger wires handle higher loads better than smaller wires(given the wires are composed of the same material). That's something someone claiming an understanding of basic electronics, such as yourself, should very well know. More to the point however, we aren't talking about transistors here. So the term leakage is being totally misused. Substitute it with resistance and things clarify themselves. Again, you should know better than to make that mistake too. Read this for proof.

   Smaller transistors have higher leakage currents...

4. I suspect you're wrong about that as well. The TSV part anyway. It's the stacking of the dies and that tighter circuit density that is likely going to cause problems with excessive heat and/or higher amperage/voltage. I'm thinking they probably made the TSVs robust enough to handle more heat, voltage/amperage, and throughput than they're likely to encounter.

In short, the HBM is apparently pretty well tapped out. Not that it matters much. Running it @ stock speed is going to be more than sufficient in most cases. And it seems as though you can get another 50MHz from it quite easily if it weren't. Possibly even another 100Mhz.
http://www.techpowerup.com/forums/threads/overclocked-hbm-its-true-and-its-fast.213875/

oi
a Fury is Not a Fury X so please stop comparing apples to oranges mmkay
if the above statements are true then one could assume that because the Fury is a cut down chip it doesn't draw as much current as the "fury x" thus opening some additional headroom in both the temperature and clock speed department ... O wait thats how it always is ... sigh

Except that isn't how it always is. If it were, then a 7950 would OC better than a 7970. Which isn't the case. What a prime example of putting the "ass" in assume. Nice!

I'm baaaack! Did you miss me?

 

[OneMoar]

No. It doesn't make sense...at all. Quite the opposite. See below.
It does in fact operate @ 1.3V. You later proved that. But failed to edit this post. Classic.

1. There is no page 2.
2. There's that proof I spoke of though.
3. The interposer is 65nm...so increased leakage due to heat or not would be minimized. Not accentuated. Bigger wires handle higher loads better than smaller wires(given the wires are composed of the same material). That's something someone claiming an understanding of basic electronics, such as yourself, should very well know. More to the point however, we aren't talking about transistors here. So the term leakage is being totally misused. Substitute it with resistance and things clarify themselves. Again, you should know better than to make that mistake too.
4. I suspect you're wrong about that as well. The TSV part anyway. It's the stacking of the dies and that tighter circuit density that is likely going to cause problems with excessive heat and/or higher amperage/voltage. I'm thinking they probably made the TSVs robust enough to handle more heat, voltage/amperage, and throughput than they're likely to encounter.

In short, the HBM is apparently pretty well tapped out. Not that it matters much. Running it @ stock speed is going to be more than sufficient in most cases. And it seems as though you can get another 50MHz from it quite easily if it weren't. Possibly even another 100Mhz.
http://www.techpowerup.com/forums/threads/overclocked-hbm-its-true-and-its-fast.213875/
Except that isn't how it always is. If it were, then a 7950 would OC better than a 7970. Which isn't the case. What a prime example of putting the "ass" in assume. Nice!

I'm baaaack! Did you miss me?

k it seems "idiot" is back in season *loads shotgun*
1. "i doubt that" was in reply to 54ths/Aquri post asking if the core and the HMB ran at the same voltage . they don't.
2. larger process node = lower temp threshold that is ALWAYS the been the case and even more so because the PCH is also 65nm
3. the TSV is the interconnect if anything is going to be a heat related point of failure its going to be there simply because its larger then the rest of the circuit path
4. in relation to stability generally yes a cut down chip is always going to remain more stable at a higher temp then its fully fledged counter pat
when you get down to the nano-meter scale the laws of electromagnetism and Thermodynamics start getting a bit strange
these are
a good read
http://www.amazon.com/dp/354042847X/?tag=tec06d-20
http://www.eng.uc.edu/~beaucag/Classes/Properties/Thermodynamics.pdf
so would you like to keep running you mouth so I can continue to slap it shut ? if so I am game sonny

 

[dorsetknob]

Oh Dear Another Bitchslapping Contest
Can you both wait awhile i need to go out and buy popcorn and Kentucky fried chicken

 

[MrGenius]

We're not talking about transistors. I did say that right? Yep! Sure did.

 

[Aquinus]

doesn't matter the interposer more then anything is a substrate and insulator insulators tend to lose some of there effectiveness the hotter they get
and let me guss you think it doesn't contain any transistors HAHAHAHAAhahahahaha
enough welcome to ignore

While you're correct, the increase in resistance/impedance doesn't increase drastically within operating temperatures.

The bigger point is that we don't know how much current is getting put through the pins and traces intended for power delivery. Those are the ones that will emit the most heat because the current will be the highest (remember, heat created is based on current and logic circuits are pretty low on current but higher on voltage where power delivery tends to be similar voltage, but much higher current because it's feeding all of the logic in the GPU or HBM.

Transistors only make more heat because resistance is changing when transistors switch on or off and as a result, generate heat when switched with current passing through it. The faster it switches states, the less heat will be released. When I say that, I don't mean clock speed, I mean how quickly the transistor actually switches states (as in, how quickly does the channel close or open after being clocked.)

 

[OneMoar]

While you're correct, the increase in resistance/impedance doesn't increase drastically within operating temperatures.

The bigger point is that we don't know how much current is getting put through the pins and traces intended for power delivery. Those are the ones that will emit the most heat because the current will be the highest (remember, heat created is based on current and logic circuits are pretty low on current but higher on voltage where power delivery tends to be similar voltage, but much higher current because it's feeding all of the logic in the GPU or HBM.

Transistors only make more heat because resistance is changing when transistors switch on or off and as a result, generate heat when switched with current passing through it. The faster it switches states, the less the heat will be released.

heres a intresting paper that looks like its dated but relevant especially the bits about the latency and signaling
http://www.cse.psu.edu/~juz138/files/islped209-zhao.pdf
if anything the weak link looks like its gonna be the PCH its lower on the stack and has the least cooling but has the highest transistor density

 

[Aquinus]

heres a intresting paper that looks like its dated but relevant especially the bits about the latency and signaling
http://www.cse.psu.edu/~juz138/files/islped209-zhao.pdf
if anything the weak link looks like its gonna be the PCH its lower on the stack and has the least cooling but has the highest transistor density

An Intel PCH isn't clocked anywhere near as high as a regular CPU. Naturally, clocking any IC slower will reduce power demands. Remember how I said transistors create the most heat when they switch states? Less clock cycles = less state changes = less heat at the same voltage. I suspect that the wider bus and slower memory contributes to this effect. Another big question is how much more heat does the DRAM have to tolerate by being stacked and does that balance out from the heat saved from clocking it slower?

Simply put, there are a lot of good questions and not enough information to really answer many of them.

 

[OneMoar]

An Intel PCH isn't clocked anywhere near as high as a regular CPU. Naturally, clocking any IC slower will reduce power demands. Remember how I said transistors create the most heat when they switch states? Less clock cycles = less state changes = less heat at the same voltage. I suspect that the wider bus and slower memory contributes to this effect. Another big question is how much more heat does the DRAM have to tolerate by being stacked and does that balance out from the heat saved from clocking it slower?

Simply put, there are a lot of good questions and not enough information to really answer many of them.

you are essentially packing more transistor density into a smaller area a by stacking so even with the reduced clock i suspect its going to run pretty toasty

that research paper says 80W per chip stack with a maximum operating temp of 80c
but that paper was written in 2013 so who knows how much the spec changed
anyone with a FURY X wanna volunteer one for some experimentation

 

[OneMoar]

http://forums.anandtech.com/showthread.php?t=2440030&page=1
1.4V and sub-zero cooling and the best he could muster was a extra ~200Mhz
*overclockers dream* ... more like nightmare }:]