• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Poll: ** The AMD VEGA Thread **

On or off the hype train?

  • (off) Train has derailed

    Votes: 207 39.2%

  • (on) Overcrowding, standing room only

    Votes: 100 18.9%

  • (never ever got on) Chinese escalator

    Votes: 221 41.9%
  • Total voters
    528
Status
Not open for further replies.
From a quick glance there seems to be 12 mosfets and 13 inductors :s don't have the expertise to teardown without spending a lot of time on it.

Not sure if one of them has an unpopulated mosfet space or if its using a different (uprated?) mosfet for that inductor its hard to see in the images.

EDIT: Its one of the inductors that would likely be powering the memory so possibly it is a single phase with an different, maybe uprated, mosfet driver.
 
nashathedog said:
I thought HBM was meant to reduce power draw?
Click to expand...

Mauller said:
yeah, when you are considering the bandwidth and capacity you get, that GDDR5 he is quoting is probably for 6GB clocked at 7000 etc.
Click to expand...

Yup, and theres a chance Hynix are "overclocking" the HBM or they're just being over volted. Considering they only had 1.6Gbps until recently, and FE has 1.89Gbps.
Buildzoid that did that marvelous tear down and look at the PCB saw HBM only has a single phase, and in the video says he's worried its over volted like HBM on Fiji was.

It was his only complaint of the entire FE PCB, stating he had hoped it was 2 phase at least.

If it's true that the HBM2 with two 8GB modules uses that much memory two 4GB ones should really help power usage of normal RX Vega.
 
nashathedog said:
I thought HBM was meant to reduce power draw?
Click to expand...

Another reason HBM is not all its claimed to be lol.

I have never liked HBM from the first time I used it and really think AMD would do better by dropping it on future cards.

Anyone seen DM as it would be interesting to see how he can try to turn this farce into something good.
 
N19h7m4r3 said:
Buildzoid that did that marvelous tear down and look at the PCB saw HBM only has a single phase, and in the video says he's worried its over volted like HBM on Fiji was.
Click to expand...

Ah cool his video has a much clearer shot of the PCB - one of the phases clearly has a different mosfet to the others - gonna watch it see what he says.
 
Kaapstad said:
Another reason HBM is not all its claimed to be lol.

I have never liked HBM from the first time I used it and really think AMD would do better by dropping it on future cards.

Anyone seen DM as it would be interesting to see how he can try to turn this farce into something good.
Click to expand...

The latency and bandwidth with 3-4 modules is great for compute and deep learning. Which is why NVIDIA use it on their top end Teslas and Quadros.

AMD's issue is their "one design" for it all. Vega with GDDR5X would be cheaper and get the job done for gaming; with HBM on their Instinct card.

Take into account the HBM2 supplies issues as well; and RX Vega would have possibly been out by now, cheaper, and have plenty of bandwidth.
 
Kaapstad said:
Another reason HBM is not all its claimed to be lol.

I have never liked HBM from the first time I used it and really think AMD would do better by dropping it on future cards.

Anyone seen DM as it would be interesting to see how he can try to turn this farce into something good.
Click to expand...

Except none of your claims about HBM were backed up with evidence. And had nothing to do with fiji's performance in the end, which was shown by how performance greatly improved over time and in DX12/Vulkan games compared to DX11.

if you had 16GB of GDDR5X running at the same bandwidth, it would likely use more power than this claimed power usage.
 
Kaapstad said:
Another reason HBM is not all its claimed to be lol.

I have never liked HBM from the first time I used it and really think AMD would do better by dropping it on future cards.

Anyone seen DM as it would be interesting to see how he can try to turn this farce into something good.
Click to expand...

I liked HBM1, The only downside I saw was the 4gb limit and 4gb being just 4gb, other than that my experience with the Fury was good, I can see why it'd be an issue running 4k and crossfire with 4gb cards though, If you do end up getting an RX Vega it'd be interesting to see if the additional memory helps it shine better. I agree with the more hassle than it's worth bit though, I'm only guessing but it feels like HBM's teething pains have been tripping RTG up, where as GDDR5x memory has been every bit as good as HBM from a gaming perspective, It's a shame they didn't go with that until HBM matured enough.
 
N19h7m4r3 said:
The latency and bandwidth with 3-4 modules is great for compute and deep learning. Which is why NVIDIA use it on their top end Teslas and Quadros.

AMD's issue is their "one design" for it all. Vega with GDDR5X would be cheaper and get the job done for gaming; with HBM on their Instinct card.

Take into account the HBM2 supplies issues as well; and RX Vega would have possibly been out by now, cheaper, and have plenty of bandwidth.
Click to expand...

Totally agree.

AMDs mistake is not using the right tool for the job.
 
I suppose AMD sticking with HBM is for them to have CPU GPU MEM on one soc ...great for future consoles
 
Mauller said:
Except none of your claims about HBM were backed up with evidence. And had nothing to do with fiji's performance in the end, which was shown by how performance greatly improved over time and in DX12/Vulkan games compared to DX11.

if you had 16GB of GDDR5X running at the same bandwidth, it would likely use more power than this claimed power usage.
Click to expand...

Err what has DX12/Vulkan got to do with HBM?

I did not claim to be an expert on the subject like some people (DM for example), what I did have was access to the same amount of evidence as everyone else "none as HBM was a brand new product". What did it for me was it was poor on the Fiji cards and held back the performance @1080p and lacked the capacity to run higher resolutions with max detail settings. User data is far more relevant than theoretical arguments.
 
Mauller said:
something for people to pick their brains over
Click to expand...

Looking at the Buildzoid video and the datasheets he is almost certainly correct that the HBM is powered by a single phase and the output of that phase is nowhere near 75 watt. At best its capable of around 40 watt sustained load.

Anything extra above the package load and the phase going to the HBM will be VDDCI/VPP - not sure how much wattage of that is connected to the HBM but its pretty low - around 13 watt max.
 
Last edited:
Kaapstad said:
Err what has DX12/Vulkan got to do with HBM?

I did not claim to be an expert on the subject like some people (DM for example), what I did have was access to the same amount of evidence as everyone else "none as HBM was a brand new product". What did it for me was it was poor on the Fiji cards and held back the performance @1080p and lacked the capacity to run higher resolutions with max detail settings. User data is far more relevant than theoretical arguments.
Click to expand...

Thats the point, you kept blaming HBM on the 1080p performance, which using DX12 and Vulkan showed it was not the problem. And the DX11 drivers caught up over time.

Of course i agree that 4GB held it back, but the performance of the HBM itself was not a major reason for fiji's performance issues at lower resolutions.
 
Klo said:
They are limited by money, they couldn't afford to do multiple chips for every use.
Click to expand...

This is the really sad bit.

They tried to be innovative using HBM whereas for gaming GDDR5X would have been better and cheaper. Unfortunately this would not have been so good for their pro cards for the reasons N19h7m4r3 stated above.
 
Status
Not open for further replies.
Back
Top Bottom