Vega Rumors

[razor1]

guys, guys, don't worry, performance will be fine. In another month when gaming vega is released, we'll see ledra and razor with crossfire Vegas in their personal rigs.

If thats the only way its going to match a titan xp around 10 % of the time, at least my 1500 watt Corsair will be able to handle em. Might have to remove one of my cpu's or 4 out of the 8 hard drives I have though lol.

Just seeing how AMD showed coming real and people are still saying wait for drivers, wait for RX Vega, guess what AMD showed it, and that came out real, and they also stated RX Vega will be faster in games, I expect it to be faster, but not by much. There is no magic drivers that are going to make anything great from a sub par performance with 300 watts sucking monstrosity.

Just look at Polaris and tell me why was it farther behind Pascal 1060 than Tonga was to the 960 or Hawaii was to 970, and Vega FE is showing worse characteristics than that.

Polaris 150 watts same performance as the 1060 @ 120 watts, that is 25% difference in perf/watt

Vega FE 300 watts lets say = in all aspects to the 1080 performance (ignoring the poor results from Metro which was worse than the 1070 and Hit man which was around the 1070) @ 180 watts. That is a 66% difference in perf/watt. I can only think of one thing even with lets say they get RX Vega to be 10% faster than a gtx 1080 its a step backwards, most likely that core clock is way above the chips ideal envelope.

 

[noko]

So the max clock so far was 1650? Then it was too hot and throttled. So what is the real max clock if uber cooled?

 

[Dayaks]

It'll be fine. The cooler on the Fury X can keep the 275 watt Fury X at 60-65* at 100% load and 50% fan all week long. What's 25 watts more gonna do? 55% fan? It's a nice thick quality rad with plenty of cooling headroom. I've been mining on my Fury X the last week. No concerns with a pair of them in the cosmos 1010 case I have running full tilt. The side panel even stays on!

Right, These 240mm rads are pure overkill, except for people with AMD FX. Pure sales garbage. Just like dual-fan towers.

Unless you're searching for 5+ GHz, or have a date with 8 cores, 120mm rads are plenty.

I don't know, my Fury X pump was making a lot of noise... water pumps don't like heat. I'd honestly get a big air cooled version over water. AMD had all sorts of issues last time.

You guys really think the wattage won't go up from the cherry picked FE cards when they push clocks with the gaming cards?

 

[razor1]

So the max clock so far was 1650? Then it was too hot and throttled. So what is the real max clock if uber cooled?

Even upping the memory the card started throttling, So insufficient cooling, no idea about uber cooled. This guy used 2 fans directly at the intake and exhaust to keep it from throttling.

 

[noko]

Even upping the memory the card started throttling, So insufficient cooling, no idea about uber cooled.

I wonder if AMD design the FE sufficient to pump 400w plus through her? Just think of the power at 1700mhz needed which also means cooling as well. As for the water cooler, think of a cpu at 300w and what kind of cooler that would call for. Biggest difference with the gpu is it is a much larger chip so you have more surface area to allow better cooling meaning much hotter fluid in your loop. Start pumping 350w+ through that radiator and it will turn into a nice furnace. Will be interesting if the AIBs really pump the electrical juice through Vega (bright shinny hot star) and have good VRM design to push it to 400w if Vega can take it. Hot rod.

 

[razor1]

I wonder if AMD design the FE sufficient to pump 400w plus through her? Just think of the power at 1700mhz needed which also means cooling as well. As for the water cooler, think of a cpu at 300w and what kind of cooler that would call for. Biggest difference with the gpu is it is a much larger chip so you have more surface area to allow better cooling meaning much hotter fluid in your loop. Start pumping 350w+ through that radiator and it will turn into a nice furnace. Will be interesting if the AIBs really pump the electrical juice through Vega (bright shinny hot star) and have good VRM design to push it to 400w if Vega can take it. Hot rod.

LOL yeah you are right, but AIB's there is only so much they will do, they still have to stay in PCI-e specs, unless they want to warranty something above and beyond that spec cause if anything happens to other components (doesn't have to be because of the card), everyone will point to the cards fault its not in PCI-e spec. So 300 watts is kinda max I think. But using a better cooler should drop power consumption down a bit so they might be able to up the clocks a little bit out of the box.

 

[noko]

LOL yeah you are right, but AIB's there is only so much they will do, they still have to stay in PCI-e specs, unless they want to warranty something above and beyond that spec cause if anything happens to other components (doesn't have to be because of the card), everyone will point to the cards fault its not in PCI-e spec. So 300 watts is kinda max I think. But using a better cooler should drop power consumption down a bit so they might be able to up the clocks a little bit out of the box.

What they can do is make sure it is 300w or less factory default settings and let the user turn their PC into a blow torch.

 

[razor1]

What they can do is make sure it is 300w or less factory default settings and let the user turn their PC into a blow torch.

or a pinto that got hit on the rear bumper lol.

 

[Archaea]

I don't know, my Fury X pump was making a lot of noise... water pumps don't like heat. I'd honestly get a big air cooled version over water. AMD had all sorts of issues last time.

You guys really think the wattage won't go up from the cherry picked FE cards when they push clocks with the gaming cards?

You had a defective pump. My two Fury x pumps are absolutely silent outside the case. So is my friends's single.

 

[noko]

or a pinto that got hit on the rear bumper lol.

Well it was suppose to be or should I say hinted at 225w 12.5Tflops which would have been a good to very good chip. It looks like 300w at around 1400mhz making it less than 12.5Tflops, peak of 1600mhz looks to be BS since under load it throttles. I agree with you that the top gaming cards probably will be around 10% faster, if anyone really wants to put in what is needed for the power requirements to push it. Just think two of these and a 7900x OC to the max - I think I would need a much larger A/C unit for my room.

 

[Araxie]

Right, These 240mm rads are pure overkill, except for people with AMD FX. Pure sales garbage. Just like dual-fan towers.

Unless you're searching for 5+ GHz, or have a date with 8 cores, 120mm rads are plenty.

Apples to Orange, heat transfer coefficient between CPUs and GPUs have a big difference and it's the fact that GPUS are Direct Die cooled, the CPUs lid even soldered have a big concentration and accumulation of heat there so bigger cooling it's required in order to handle that dissipation.

 

[ChosenUno]

RX Fermi

No wait, it's not even Fermi good. My bad.

 

[noko]

Apples to Orange, heat transfer coefficient between CPUs and GPUs have a big difference and it's the fact that GPUS are Direct Die cooled, the CPUs lid even soldered have a big concentration and accumulation of heat there so bigger cooling it's required in order to handle that dissipation.

What is the max temperature for GPU before thermal throttling? 85c? Meaning the cooling water has to be lower than that temperature to transfer heat. The issue is how much heat in watts can you transfer through a 120mm cooler with cooling water limited to let say a max of 70c? If that GPU and HBM memory is consuming 400w, you will need to transfer 400w heat out into the air keeping the fluid temperature always less than 75c if 10c Delta is what is needed for heat transfer from the GPU die to water. Anyways I think the 7990x2 was design for 400w of heat so should be possible.

 

[Archaea]

Apples to Orange, heat transfer coefficient between CPUs and GPUs have a big difference and it's the fact that GPUS are Direct Die cooled, the CPUs lid even soldered have a big concentration and accumulation of heat there so bigger cooling it's required in order to handle that dissipation.

Nonsense

 

[Dayaks]

What is the max temperature for GPU before thermal throttling? 85c? Meaning the cooling water has to be lower than that temperature to transfer heat. The issue is how much heat in watts can you transfer through a 120mm cooler with cooling water limited to let say a max of 70c? If that GPU and HBM memory is consuming 400w, you will need to transfer 400w heat out into the air keeping the fluid temperature always less than 75c if 10c Delta is what is needed for heat transfer from the GPU die to water. Anyways I think the 7990x2 was design for 400w of heat so should be possible.

Normally the fluid needs to be under 50C or you damage it.

 

[Michaelius]

Whats it like to be a dyed in the wool fanboy?

I don't know you would have to ask people believing in magic gaming optimisations, drivers , not strong enough PSUs and whatever else theories will be produced over next few weeks. Or make a thread on r/AMD.

But I'll admit I have a flaw in character - I love that crow serving moment when I was told I'm wrong for a year by usual suspects.

 

[defaultluser]

Apples to Orange, heat transfer coefficient between CPUs and GPUs have a big difference and it's the fact that GPUS are Direct Die cooled, the CPUs lid even soldered have a big concentration and accumulation of heat there so bigger cooling it's required in order to handle that dissipation.

Bullshit. The amount of heat transferred is exactly the same as a CPU. A 250w CPU produces the same amount of heat as a 250w GPU. Both have additional heat from ram and VRMs, but those are secondary.

The difference is in the waterblock, not the fucking radiator. And the full-coverage GPU waterblock just means it can cool the ram too, because they run at three times the speed of DDR4. Its still an insignificant portion of power consumed (typically 10-15%)

the same 120MM radiator will result in the same temperatures on a 250w CPU and a 250w (processor plus memory) GPU.

Now can you idiots STOP PRETENDING like you know how thermodynamics work? no, you don't need a monster HSF. The only really undersized HSF is the one Intel ships with their Core i5/i7 processors. And some of the really cheap 3rd-party coolers out thr suck as well. But a simple 120mm tower with heatpipes is plenty for most things.

 

[noko]

Bullshit. The amount of heat transferred is exactly the same as a CPU. A 250w CPU produces the same amount of heat as a 250w GPU. Both have additional heat from ram and VRMs, but those are secondary.

The difference is in the waterblock, not the fucking radiator. And the full-coverage GPU waterblock just means it can cool the ram too, because they run at three times the speed of DDR4. Its still an insignificant portion of power consumed (typically 10-15%)

the same 120MM radiator will result in the same temperatures on a 250w CPU and a 250w (processor plus memory) GPU.

Now can you idiots STOP PRETENDING like you know how thermodynamics work? no, you don't need a monster HSF. The only really undersized HSF is the one Intel ships with their Core i15/7 processors.

Heat maybe the same but the temperature needed to transfer that heat is different due to differences in heat transfer characteristics. HD 7990 x2 was rated for 400w, FE liquid cool 375w. On a CPU you will not transfer that heat rate with that 120mm size cooler with normal ambient air temps and air flow. If so you would not see 240 mm radiators and up. 7900x with Corsair H110 was not able to keep cool at 300w plus. For the CPU the larger radiator is needed for colder fluid temperatures to transfer the heat out of the CPU.

 

[defaultluser]

Heat maybe the same but the temperature needed to transfer that heat is different due to differences in heat transfer characteristics. HD 7990 x2 was rated for 400w, FE liquid cool 375w. On a CPU you will not transfer that heat rate with that 120mm size cooler with normal ambient air temps and air flow. If so you would not see 240 mm radiators and up. 7900x with Corsair H110 was not able to keep cool at 300w plus. For the CPU the larger radiator is needed for colder fluid temperatures to transfer the heat out of the CPU.

Only on the AMD side of the fence, where for some fucking reason the FX series throttles at 62C.

They fixed this major flaw with Ryzen, which throttles at over 80C just like any quality piece of silicon should. So no, now that your throttle points are close enough, the cooler needs are exactly the same.

https://community.amd.com/thread/213058

And yes, AMD's "budget" FX had that one additional issue going for it. That you needed to buy a much bigger cooler than the Intel counterpart to keep it form throttling. Which is why the 240mm fans have become universal, even when they're almost completely unnecessary.

For the same reason that giant GPU coolers have become popular - because everyone else has them

 

[noko]

Only on the AMD side of the fence, where for some fucking reason the FX series throttles at 62C.

They fixed this major flaw with Ryzen, which throttles at 95C just like any quality piece of silicon should. So no, nmow that your throttle poins are close enough, the cooler needs are exactly the same.

https://community.amd.com/thread/213058

New SkylakeX, KabyLake pushing the clocks - that 120mm cooler on that FE would not fare too well. The CPU Smaller die (less heat transfer area), toothpaste for the TIM (very poor heat transfer characteristics compared to solder) = big ass cooler needed to get the water as cold as possible. One thing that ThreadRipper has going for it, same heat transfer area per die.

 

[defaultluser]

New SkylakeX, KabyLake pushing the clocks - that 120mm cooler on that FE would not fare too well. The CPU Smaller die (less heat transfer area), toothpaste for the TIM (very poor heat transfer characteristics compared to solder) = big ass cooler needed to get the water as cold as possible. One thing that ThreadRipper has going for it, same heat transfer area per die.

Yes, but not on 4 cores or less. And the 6+ core love-fest really has only come in the last 6 months, whereas the 240, and now 280mm radiator lovefest has been brewing for years. Justify it however the fuck you want, but people have routinely been paying $60-80 too much for their cooling needs.

Except for AMD FX. But you people knew what you were getting into.

 

[Dayaks]

Yes, but not on 4 cores or less. And the 6+ core love-fest really has only come in the last 6 months, whereas the 240, and now 280mm radiator lovefest has been brewing for years. Justify it however the fuck you want, but people have routinely been paying $60-80 too much for their cooling needs.

Except for AMD FX. But you people knew what you were getting into.

If you run at stock sure... I run my 5960x at a decent OC and the temps go up fast even with a 280 rad.

Remember nVidia's node jump and how people were surprised by the temps due to the higher thermal density alone? 200 watts is a lot harder to cool on Pascal than it was on Maxwell. AMD is making the same jump. Fury X isn't really a good comparison. Vega will be harder to maintain the same temps due to a higher thermal density.

 

[Captain Newmackwa]

It's not looking good for the gaming version of the RX Vega since a GeForce GTX 1080 can do as much (or more) with so much less. AMD can probably remedy the situation with aggressive pricing but no doubt, that would be cutting a deep hole into their margins considering that Vega has a bigger chip with HBM2 memory and requires more robust power and thermal solutions. By all intents and purposes, a card that's as costly to make as Vega should be at least competitive to the 1080Ti to give AMD the opportunity to earn decent margins.

 

[Dulce Canales]

Is that a fact or just..rumors?

 

[KazeoHin]

Well, you have to wonder, just how much different is the FE from the RX cards? Is it vastly different? How is it binned? Is it using the same number of cores? The same silicon?

These, of course, are trick questions. We know that the FE uses the same silicon and core-count as the RX. Would you think a $1000+ GPU would be binned the same as a ~$500 one? So we are most likely going to get HOTTER and LESS EFFICIENT chips on the RX cards..

It doesn't look good for poor old RTG..

 

[Ieldra]

Well, you have to wonder, just how much different is the FE from the RX cards? Is it vastly different? How is it binned? Is it using the same number of cores? The same silicon?

These, of course, are trick questions. We know that the FE uses the same silicon and core-count as the RX. Would you think a $1000+ GPU would be binned the same as a ~$500 one? So we are most likely going to get HOTTER and LESS EFFICIENT chips on the RX cards..

It doesn't look good for poor old RTG..

Rx Vega will hit 2000mhz no problem with just a passive heatsink and a pleasant summer breeze, no need to worry!

Especially if you're an investor that was most certainly not swindled by a bullshit H1 launch

 

[Boil]

Look, all I want is a RX Vega Nano that I can pair with a Ryzen R7 1700 CPU, place in a Ncase M1 chassis & cool it all with a single 240mm radiator (NO overclocking) in a custom water cooling loop...

I would be more than happy with GTX 1080 performance from said RX Vega Nano, especially if the price is less than a GTX 1080 (the REAL price, NOT the mining craze inflated price)...

Now, if I can get GTX 1080Ti performance at less than GTX 1080 prices, all the better...!

One more month...!

 

[Ieldra]

Look, all I want is a RX Vega Nano that I can pair with a Ryzen R7 1700 CPU, place in a Ncase M1 chassis & cool it all with a single 240mm radiator (NO overclocking) in a custom water cooling loop...

I would be more than happy with GTX 1080 performance from said RX Vega Nano, especially if the price is less than a GTX 1080 (the REAL price, NOT the mining craze inflated price)...

Now, if I can get GTX 1080Ti performance at less than GTX 1080 prices, all the better...!

One more month...!

I'm incredibly confused by your statements right now...

GTX 1080 is 180W. With a custom water loop you are probably looking at the same 180W with balls to the wall OC at 2050MHz and memory at 11GT/s etc. Based on ***ALL*** the benchmarks we have seen so far this will outperform Vega, while using close to half the power. Why would you even consider Vega for a SFF build ?

 

[Ieldra]

This is Raja claiming there is no developer intervention reqiured to make use of the "higher throughput per clock cycle" in terms of geometry. Fascinating claim really considering on the Vega FE spec sheet geometry throughput is literally identically to Fiji's and Polaris'.

I won't lie, I'm highly amused by this launch, if only because of the sheer number of posters who will have to eat crow after over-hyping this damned thing to the Proxima Centauri and back. it is disappointing to see AMD coming up with such a lackluster product after >1 year of waiting and frankly this whole launch is a shambles .

Spoiler: Can't resist

 

[stashix]

#PoorMaxwell

 

[noko]

Perf/$, top gaming performance 1080Ti. It might be fun as well as stupid to push 400w using chill water to cool a Vega. How it is looking, just buy a 1080Ti. I was hoping for a decent upgrade Nano size, that is not looking good either.

 

[5150Joker]

I'm not surprised in the least and those of us who were more pragmatic about what AMD could accomplish already knew they cannot compete with NVIDIA. Frankly, I'm surprised they managed to get it to 1080 levels of performance but that HBM2 + die size is going to destroy their margins.

 

[SnowBeast]

August availability of Vega then? Too late. 2070/80 are coming. Leaks of Nvidia sticking with DDR5X and the newer version of 14nm finfet (12nm supposedly) and building inventory since May looks like another 900 series release coming end of September. Think about it. Smaller process, maybe not Volta just another tweaked Pascal would be enough of a kick in the pants, higher perf/watt@ 900 series pricing. It's about that time. Speculation, sure. It just all adds up. We are going to have Ryzen Rigs with Nvidia Gpu's.

 

[Archaea]

Normally the fluid needs to be under 50C or you damage it.

nonsense again. (Or at least not universal). You can clearly tell on the Fury X that their stock emgineered target temp is about 60-65*. If you raise voltage, temp stays the same and fan speed goes up. If you lower the voltage l, fan speed goes down but temp stays about the same. I've read this in a review and noticed it on my own system in the last week fiddling with mining on the pair of Fury X. If the cooling fluid deteriorates at 50* C then they wouldn't target 60-65* through engineering --- and give the product a standard 3 year warranty.

 

[lolfail9001]

nonsense again. (Or at least not universal). You can clearly tell on the Fury X that their stock emgineered target temp is about 60-65*. If you raise voltage, temp stays the same and fan speed goes up. If you lower the voltage l, fan speed goes down but temp stays about the same. I've read this in a review and noticed it on my own system in the last week fiddling with mining on the pair of Fury X. If the cooling fluid deteriorates at 50* C then they wouldn't target 60-65* through engineering --- and give the product a standard 3 year warranty.

I got confused. Are you talking about Fury X target temp or target temp of it's liquid?

 

[Archaea]

I got confused. Are you talking about Fury X target temp or target temp of it's liquid?

The card itself. That's what's measured by wattman. This in the context of someone saying the same Fury X cooler wouldn't be sufficient for 300 watt Vega with its 25 watts more draw. And then someone else saying cooling solution deteriorates above 50*c.

Now as far as the 375 watt version and an OC on top of that? I don't know. It might seem the fan speeds would ramp up high enough to be a bother with 100 + more watts, but it could cool it to below throttle temp im sure because I've never seen the Fury X pair temps higher than about 65 and fan speeed of 55* and they are in crossfire (or not for mining) in a case with poor airflow (cosmos 1010) with the side panel on and thats with an overclock and tdp raise allowance of 25% (not sure how much it actually uses of that allownace).

Maybe for Vega, they deepen the rad and up the pump RPM, and thicken the fan mm for more pressure at low RPM and it could still be silent like fury X with the increased power draw.

 

[fuzzylogik]

so.. who games with a professional graphics card? besides the cost i think there's other reasons why people don't buy them for a great gaming experience? Isn't the "nice" thing about this card is that it's possible to do both on one workstation? Sure, it's not the "best" for gaming but good enough to simplify the creation process while still being pretty decent for workstation use without being crazy expensive? Sounds like these might be a steal for small indy developers or the like *shrugs* I'll wait for the actual gaming card reveal and a [H] review before actual judgement - and then go for whatever is the best bang for my buck.

 

[Dayaks]

nonsense again. (Or at least not universal). You can clearly tell on the Fury X that their stock emgineered target temp is about 60-65*. If you raise voltage, temp stays the same and fan speed goes up. If you lower the voltage l, fan speed goes down but temp stays about the same. I've read this in a review and noticed it on my own system in the last week fiddling with mining on the pair of Fury X. If the cooling fluid deteriorates at 50* C then they wouldn't target 60-65* through engineering --- and give the product a standard 3 year warranty.

The card itself. That's what's measured by wattman. This in the context of someone saying the same Fury X cooler wouldn't be sufficient for 300 watt Vega with its 25 watts more draw. And then someone else saying cooling solution deteriorates above 50*c.

Now as far as the 375 watt version and an OC on top of that? I don't know. It might seem the fan speeds would ramp up high enough to be a bother with 100 + more watts, but it could cool it to below throttle temp im sure because I've never seen the Fury X pair temps higher than about 65 and fan speeed of 55* and they are in crossfire (or not for mining) in a case with poor airflow (cosmos 1010) with the side panel on and thats with an overclock and tdp raise allowance of 25% (not sure how much it actually uses of that allownace).

Maybe for Vega, they deepen the rad and up the pump RPM, and thicken the fan mm for more pressure at low RPM and it could still be silent like fury X with the increased power draw.

I was talking about the fluid itself, not the GPU chip temp. And again there's no way a faster, cheaper GPU than the FE will have the same tdp.

You also ignore the thermal density of Vega is higher. It'll be significantly worse off than Fury x by that alone.

 

[razor1]

I was talking about the fluid itself, not the GPU chip temp. And again there's no way a faster, cheaper GPU than the FE will have the same tdp.

You also ignore the thermal density of Vega is higher. It'll be significantly worse off than Fury x by that alone.

I agree it looks like the thermal density is higher in Vega, but with Fury X since its got a water cooler hard to do a straight out comparison. Also agree the liquid needs to be around 50 to 60 C, any more it looses efficiency fast.

Having said that though, if anyone with a Fury X would like to take that AIO off and test it out, I would think it will hit 300 watts at default settings. So Vega and Fiji might actually end up close to each other when it comes to thermal density. But it all depends on where the clock ranges are.

 

[Hameeeedo]

Rys from RTG says Vega FE has no gimped gaming drivers compared to RX Vega, the driver is just older.
https://forum.beyond3d.com/posts/1989522/.