• 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.

AMD Locks Down HBM Frequency on Fiji Cards

rtho782 said:
It's still accurate to say that HBM has brought nothing in terms of bandwidth that an 8GT/s 512bit GDDR5 implementation could not have given us.

Both of which are achievable with current hardware.

To hear people tell of it you'd think HBM gave you a 2-3x increase in memory bandwidth.
Click to expand...

Expense. To get more bandwidth out of gddr5 means higher and higher frequnecies which means other issues and problems and power consumption.

HBM is very low power.

Gddr5 has had its day. Even Nvidia agrees. You will only see gddr5 on low end cards now.

AMD has HBM2 out by Xmas and Nividia new cards with their equivalent memory early next year.
 
As a brand new technology I suspect AMD are more worried about bad press from people breaking cards through overclocking than from just blocking it to start with. Given the massively wide buswidth and relatively low clockspeed (compared to current GDDR5) it may well be that HBM just won't play nicely at all with any attempt at overclocking.
 
nails666 said:
As a brand new technology I suspect AMD are more worried about bad press from people breaking cards through overclocking than from just blocking it to start with. Given the massively wide buswidth and relatively low clockspeed (compared to current GDDR5) it may well be that HBM just won't play nicely at all with any attempt at overclocking.
Click to expand...

That's a good point, I can see people trying to get an extra 250mhz on the vram, thinking "That's not massive!", forgetting that it's a 25% OC to HBM.
 
rtho782 said:
It's still accurate to say that HBM has brought nothing in terms of bandwidth that an 8GT/s 512bit GDDR5 implementation could not have given us.

Both of which are achievable with current hardware.

To hear people tell of it you'd think HBM gave you a 2-3x increase in memory bandwidth.
Click to expand...

Accurate to say but not practical which is what is relevant. You might have some trouble convincing AMD and Nvidia to develop a 512bit 8GT/s GPU.

Come next year HBM2 will be leaving gddr5 somewhat in the dust. Ship needed to be jumped sometime.
 
Last edited:
Am OK with that, I don't overclock that much anyways.. Although if amd locked it down in the driver am sure msi afterburner or something will likey find away?
 
Maybe they're concerned people will go into it trying to overclock HBM like GDDR5 and damaging it in the process? Maybe there's just not much headroom for overclocking with the current HBM implementation, so rather have people slate it for being a poor overclocker they'd rather just disable it on their terms.
 
kaku said:
Maybe they're concerned people will go into it trying to overclock HBM like GDDR5 and damaging it in the process? Maybe there's just not much headroom for overclocking with the current HBM implementation, so rather have people slate it for being a poor overclocker they'd rather just disable it on their terms.
Click to expand...

Yes, this is one reason which makes the most sense. Testing for levels of degradation takes time, and memory IC seems to be the most prone to it. It's likely that it's a mixture of little to no head room without the required voltage / sufficient cooling. You may find that access to this will be for select persons only, i.e ones that are known for breaking certain world records.

There is also the same properties when overclocking GDDR5. Current is able to kill any component if the user is neglectful, so it begs the question how very little headroom there might be if they've actively chosen to stop people from overclocking it at all.
 
Silent_Scone said:
Yes, this is one reason which makes the most sense. Testing for levels of degradation takes time, and memory IC seems to be the most prone to it. It's likely that it's a mixture of little to no head room without the required voltage / sufficient cooling. You may find that access to this will be for select persons only, i.e ones that are known for breaking certain world records.

There is also the same properties when overclocking GDDR5. Current is able to kill any component if the user is neglectful, so it begs the question how very little headroom there might be if they've actively chosen to stop people from overclocking it at all.
Click to expand...

Better, insightful, more interesting post. Good work :)

To add to what you are saying HBM is totally new ground. Maybe with heavy OC it could degrade, or result in instability. So why introduce it, i.e when no other current GPU can come close even when overclocked. For now over clocking the HBM just simply isn't needed. That might be the general thinking.

When HBM 2 arrives and there are other competitive HBM cards maybe overclocking will return to the fold. In the meantime I would be perfectly happy with 512GB/s just like I have been with my Titan X's 336GB/s. That's a nice jump in bandwidth.
 
Doesn't really bother me, only got my core overclocked on my 290, memory didn't seem to make much difference, so just put it back to stock and left it.
 
HBM is exciting times, not just for gfx cards, its going to play a part to change how we use our pc's in future
that i dont even see as hype
 
Silent_Scone said:
Hi,

This is entirely subjective though. AMD's own tests show a huge gap between NVIDIA parts in reputably memory intensive game benchmarks.

The bandwidth on Fiji relatively is less than 5% higher than Hawaii at 56.8MBs vs 59.5MBs per GFlop/s.
Click to expand...

It's still way higher than Nvidia have with the Titan X and 980ti
 
TheRealDeal said:
It's still way higher than Nvidia have with the Titan X and 980ti
Click to expand...

This was my point

Boomstick777 said:
Better, insightful, more interesting post. Good work :)

To add to what you are saying HBM is totally new ground. Maybe with heavy OC it could degrade, or result in instability. So why introduce it, i.e when no other current GPU can come close even when overclocked. For now over clocking the HBM just simply isn't needed. That might be the general thinking.

When HBM 2 arrives and there are other competitive HBM cards maybe overclocking will return to the fold. In the meantime I would be perfectly happy with 512GB/s just like I have been with my Titan X's 336GB/s. That's a nice jump in bandwidth.
Click to expand...


It's entirely speculative, but makes the most sense as to why they've decided to close off. It's slightly out of character to put a finer point on it. It does beg the question to the arguably poor choice of words when saying the card is an overclockers dream however. This would imply that they are giving people the right to try uncharted territory, which it appears (at least for the moment) they are not.
 
Last edited:
panyan said:
Seems like a perfectly reasonable reason to me.
Click to expand...

See above


MadMossy said:
It's not just the bandwidth but also the bus width being much much wider than traditional GDDR5 set ups.

4096-bit vs 384/512.

That's a massive chunk of data that can be pushed through up to 8 times faster than GDDR5.
Click to expand...

These two things work in conjunction with one another
 
Last edited:
From AMD's stand point i can see why this has been done but i am a little disappointed as i wanted to see what was capable Bandwidth wise even if it did not make a difference to overall scores.
 
You must log in or register to reply here.
Back
Top Bottom