AMD Radeon Pro Vega Frontier Edition TDP and Pricing Revealed

[HD64G]

Not good card. I hoped in same performance but lower tdp (250 max). This time i think i must change from Amd to Nvidia for GPU...
... or simply they need stability, so "tdp" is for 300/ 375 W, but power used is much lower.

As you said, max power of the board is 300/375W in order for the owner to be able to oc it. Not the actual power consumption figure. It should be very clear to anyone related to computer HW for a few years even as a hobby.

 

[efikkan]

Professional cards might have a few percent headroom in TDP for stability, but they certainly don't contain a OC headroom. That would be a first…

 

[Deleted member 172152]

Professional cards might have a few percent headroom in TDP for stability, but they certainly don't contain a OC headroom. That would be a first…

The air-cooled one could oc a bit if you don't mind listening to your miniature jet engine all the time.

 

[EarthDog]

Professional cards might have a few percent headroom in TDP for stability, but they certainly don't contain a OC headroom. That would be a first…

Have you tried? Seen it? Are we able to? Why would you think that?

My thought is that they are binned like anything else. Just needs to hit X clocks with X voltage and power use and it fits a bin. Some have headroom some don't. I don't know. I never saw it, honestly and why I am asking.

 

[notb]

Vega Frontier Edition is a premium card for corporate customers. I suspect they went balls to the wall with the thermals/power because they could. This card is a really poor metric to measure the consumer cards by.

You assume that corporate clients care less about high power consumption, while most of the time it's the other way round.

Should've waited if you don't have a g-sync monitor.

But why? As he said: 1080Ti is fast enough for pretty much every traditional gaming these days (<=4K).
Anyway, as AMD seems to understand, the real money is to be made in the mainstream category, not the top-end gaming. That's why they don't upgrade the flagship cards very often and the top RX card is somewhere in the middle of NVIDIA offer.

 

[efikkan]

AMD is barely participating in the midrange where the real money in the consumer market is made. There might be volumes in low-end, but the margins are also very small. At the moment they only have RX 580 to "compete" with GTX 1060, even though it has the power consumption of a GTX 1080.

It's over a year since the launch of Pascal, Pascal is now past midway in its life cycle, and AMD is still struggling to prepare its competitor. GV104 will arrive early next year, while Navi is pushed to the end of 2018…

AMD should have dropped HBM so they at least could have mass produced their chips. They need to compete with GTX 1080, GTX 1070 and GTX 1060, that's where the revenue is. Perhaps even put a model between GTX 1060 and GTX 1070, since there is a huge performance and price gap there.

 

[FordGT90Concept]

You assume that corporate clients care less about high power consumption, while most of the time it's the other way round.

They generally don't. Most of the customers buying these cards are going to use them for that 26 TFLOP FP16 performance. They're really not going to care how much energy it uses because to get the same level of performance from NVIDIA (lower power consumption), they'd have to spend a lot more money on the hardware itself. Convincing a board to spend $1 million on a super computer versus $3 million is easy. The 30% higher energy costs on the backend is just details.

 

[notb]

They generally don't. Most of the customers buying these cards are going to use them for that 26 TFLOP FP16 performance. They're really not going to care how much energy it uses because to get the same level of performance from NVIDIA (lower power consumption), they'd have to spend a lot more money on the hardware itself. Convincing a board to spend $1 million on a super computer versus $3 million is easy. The 30% higher energy costs on the backend is just details.

But now you're concentrating on heavy calculation usage for large servers, datacenters etc. In that case you're right, because here power consumption is just about the cost.

However, in general power consumption is not just money.
These cards will land not only in RACK servers, but also in desktop workstations designed for office infrastructure. As such, it is important how much energy they draw (because there's a limit) and how much heat/noise they produce (obvious).
A TDP of 375W is beyond anything we've seen lately and the FP16 performance is just slightly better than Pascal's (TDP=250W).

 

[kruk]

AMD is barely participating in the midrange where the real money in the consumer market is made. There might be volumes in low-end, but the margins are also very small. At the moment they only have RX 580 to "compete" with GTX 1060, even though it has the power consumption of a GTX 1080..

The difference in power consumption is important only if you game 24/7, otherwise it's negligible.

Here, enjoy some nice numbers from computerbase.de:
The difference between 1060 FE and RX 580 in idle is 4 Watts. The difference at gaming is 77 Watts. The difference in Youtube video is 2 Watts. If you leave the RX 580 idling with monitor off, it can turn off into ZeroCore Power modus which consumes virtually no power, but I'm putting 1 W there just to be fair. Both cards perform virtually identical in games.

If you leave the computer running 24/7 and it idles 12 hours, you game 6 hours, watch a Youtube for 2 hours and do other things non graphical intensive things for 4 hours the 1060 FE will consume: (12h*7W + 6h*111W + 2h*29W+4h*7W)/24h = 35W per hour and the RX 580 will consume (12h*1W + 6h*188W + 2h*31W+4h*11W)/24h = 50W per hour.

15 Watts/hour difference. Better have a fire extinguisher near your RX 580 system or it might catch fire! /s

 

[FordGT90Concept]

A TDP of 375W is beyond anything we've seen lately and the FP16 FP32 performance is just slightly better than Pascal's (TDP=250W).

FTFY

I can't see anyone buying Vega over Pascal for FP32.

 

[efikkan]

The difference in power consumption is important only if you game 24/7, otherwise it's negligible…

You know very well the cost of energy consumption is the least concern.
AMD can't really compete when their contender consumes 185W and the competition consumes 120W, their competition becomes the superior choice here. Energy consumption matters because of heat and noise. AMD needs comparable efficiency to become relevant again.

With Vega FE with a (claimed) TDP of 300W, and the comparable product Quadro GP100 with a TDP of 235W, it becomes clear that Vega will struggle to compete. In the professional market they probably can sell a few by selling them a bit cheaper, but in the consumer market the margins are a lot smaller.

 

[kruk]

You know very well the cost of energy consumption is the least concern.
AMD can't really compete when their contender consumes 185W and the competition consumes 120W, their competition becomes the superior choice here. Energy consumption matters because of heat and noise. AMD needs comparable efficiency to become relevant again.

Oh, come on. You didn't even click the link and read the full test . Here, noise and temps: again, negligible differences. You are blowing things out of proportion ...

As for RX Vega: when the card comes out and if its noisier, hotter and consumes 15 Watts/hour more per my calculation vs. 1080Ti FE then you won and AMD messed up. Happy now?

 

[notb]

FTFY

I can't see anyone buying Vega over Pascal for FP32.

Nope. I meant GP100 with 20.7 TFLOPS in FP16. But I made a mistake nevertheless: Pascal is rated at 235W, not 250W.

 

[FordGT90Concept]

GP100 is the primary competitor of the Vega Frontier Edition.

 

[EarthDog]

Have you tried? Seen it? Are we able to? Why would you think that?

My thought is that they are binned like anything else. Just needs to hit X clocks with X voltage and power use and it fits a bin. Some have headroom some don't. I don't know. I never saw it, honestly and why I am asking.

@efikkan

 

[Prince Valiant]

Nope. I meant GP100 with 20.7 TFLOPS in FP16. But I made a mistake nevertheless: Pascal is rated at 235W, not 250W.

I'm well out of my depth here, please correct me if I'm wrong.

Isn't $1,700 for 20%(ish) more flops a much more attractive prospect than $6,000 for GP100 regardless of heat or noise unless those are absolutely critical issues?

 

[Deleted member 172152]

I'm well out of my depth here, please correct me if I'm wrong.

Isn't $1,700 for 20%(ish) more flops a much more attractive prospect than $6,000 for GP100 regardless of heat or noise unless those are absolutely critical issues?

Depends really. Wx7100 and p4000 get close to or beat vega frontier air-cooled in specviewperf, so depending on your applucations it can be better or worse than 6000 dollar nvidia gpu's. How pro are you?

Don't worry if you are an über-pro, AMD will make a wx vega or something as well probably that's more suited for quadro style applications at some point and for servers there will be the MI25.

I'm well out of my depth here, please correct me if I'm wrong.

Isn't $1,700 for 20%(ish) more flops a much more attractive prospect than $6,000 for GP100 regardless of heat or noise unless those are absolutely critical issues?

Depends really. Wx7100 and p4000 get close to or beat vega frontier air-cooled in specviewperf, so depending on your applucations it can be better or worse than 6000 dollar nvidia gpu's. How pro are you?

Don't worry if you are an über-pro, AMD will make a wx vega or something as well probably that's more suited for quadro style applications at some point and for servers there will be the MI25 accelerator.

 

[Prince Valiant]

I'm just a regular dude but I was curious about it .

I hope they manage to get Vega out the door sometime soon. I don't think they can afford more delays in the high end GPU segment and they need something new to trickle down across the range.

 

[Deleted member 172152]

I'm just a regular dude but I was curious about it .

I hope they manage to get Vega out the door sometime soon. I don't think they can afford more delays in the high end GPU segment and they need something new to trickle down across the range.

FE arrives near the end of the month and the rx cards will arrive late july apparently. Probably because of drivers and mainly getting stock to retailers.

 

[notb]

I'm well out of my depth here, please correct me if I'm wrong.

Isn't $1,700 for 20%(ish) more flops a much more attractive prospect than $6,000 for GP100 regardless of heat or noise unless those are absolutely critical issues?

Well... GP100 is supposed to be an enterprise-grade hardware prepared for heavy 24/7 load (yet offering normal GPU functionality).
It's a bit to early to say how well built and spec'd this new Radeon Pro Vega FE is going to be, but the pricing suggests it'll just be an expensive consumer/workstation model, not a calculation workhorse. Just by raw numbers we see in specification, it is going to outperform AMD's own Pro alternatives costing over twice as much.

Plus, NVIDIA can always ask more because of CUDA...

 

[efikkan]

Have you tried? Seen it? Are we able to? Why would you think that?

My thought is that they are binned like anything else. Just needs to hit X clocks with X voltage and power use and it fits a bin. Some have headroom some don't. I don't know. I never saw it, honestly and why I am asking.

Please read the post before mine, and you'll see what I'm talking about.
I've never seen a product incorporate OC headroom inside their TDP, and certainly not a professional product. If the final TDP really is 300/375 W, then the actual max consumption + a smal safety margin will equal this number, as with every previous product.

 

[djemergenc]

What is NVIDIA doing over there? The only job they had to do is work on their GPUs and mobile processors, but here comes AMD, killing the GPU game while manufacturing CPUs and chipsets on the side.

 

[notb]

What is NVIDIA doing over there? The only job they had to do is work on their GPUs and mobile processors, but here comes AMD, killing the GPU game while manufacturing CPUs and chipsets on the side.

Here comes AMD doing WHAT? :-D
BTW: NVIDIA is doing way more things than AMD. And they're making a profit. A lot of it.

 

[evernessince]

If those TDP numbers are real this is looking like R600 (HD 2900 XTX) all over again... I hope that the performance is at least monstrous...

Every time, without fail there is always someone who thinks TDP = power consumption. It is not and companies will calculate TDP differently.