AMD Vega May Launch with Less Than 20,000 Units Available

[R-T-B]

Fresh from the rumor-mill comes a report that low HBM2 availability may cripple the Vega launch that is expected to happen in the next few weeks, if a report from TweakTown is to be believed. As far as sources, there isn't much other than TweakTown's news report and their article claiming they had been told this by an "exclusive industry source." Apply your usual grain of salt here vigilant reader, but its certainly interesting speculation, if nothing else. It may turn out to be FUD, or it may turn out to be truth. Only the coming weeks will reveal the truth.



View at TechPowerUp Main Site

 

[cryohellinc]

May be true after all, recent leaks suggest that. Unless of course all of them were fake.

 

[G33k2Fr34k]

Wow let me guess... HBM supply issues, or they can't afford to buy and package millions of HBM chips coz they're low on money. The funny thing is, their Vega10 card at 1200MHz is as fast as a GTX1070 that uses regular GDDR5 memory. Even if they have a Vega10 card at 1500MHz, it's gonna be roughly as fast as the GTX1080, which costs way less to produce and is almost a year old. Also, Volta is on the way and is probably gonna be both faster and more efficient than Pascal.

 

[cryohellinc]

Wow let me guess... HBM supply issues, or they can't afford to buy and package millions of HBM chips coz they're low on money. The funny thing is, their Vega10 card at 1200MHz is as fast as a GTX1070 that uses regular GDDR5 memory. Even if they have a Vega10 card at 1500MHz, it's gonna be roughly as fast as the GTX1080, which costs way less to produce and is almost a year old. Also, Volta is on the way, and is probably gonna be faster and more efficient than Pascal.

The way it seems that's the sad truth, and Nvidia will push the prices again on Volta....

 

[cdawall]

Wow let me guess... HBM supply issues, or they can't afford to buy and package millions of HBM chips coz they're low on money. The funny thing is, their Vega10 card at 1200MHz is as fast as a GTX1070 that uses regular GDDR5 memory. Even if they have a Vega10 card at 1500MHz, it's gonna be roughly as fast as the GTX1080, which costs way less to produce and is almost a year old. Also, Volta is on the way, and is probably gonna be faster and more efficient than Pascal.

What gives you that idea? If they have the TDP set super low (exactly how the Nano was setup) the card will perform completely different than a card with a high clockspeed and higher TDP.

Remember the R9 Nano had a GPU clock of 1000mhz, 4096cu's and 4gb of 500mhz HBM, the Fury X had a GPU clock of 1050mhz, 4096cu's and 4gb of 500mhz HBM. 50mhz represented a 15% performance boost? No the 100w of TDP difference did. The nano would flag as a 1000mhz card in 3dmark, but only completed a tiny percent of the tests at that clockspeed, most of the time it spent running 875-950mhz, with dips to 700mhz.

If this SKU fills that same niche, expect a super low (100-150w TDP) and the card to only be running in the 1000-1100mhz range most of the time. There is already rumor of a very similar 3 card range as the Fury lineup.

 

[sergionography]

Wow let me guess... HBM supply issues, or they can't afford to buy and package millions of HBM chips coz they're low on money. The funny thing is, their Vega10 card at 1200MHz is as fast as a GTX1070 that uses regular GDDR5 memory. Even if they have a Vega10 card at 1500MHz, it's gonna be roughly as fast as the GTX1080, which costs way less to produce and is almost a year old. Also, Volta is on the way, and is probably gonna be faster and more efficient than Pascal.

With those quantities there is no way they would even release it unless its atleast 600usd+ or even more ao closer 1000usd.
To sell it at these prices it has to atleast match or even beat a titan or 1080ti.
I dont think we will see a direct 1070-1080 competitor until vega 11 comes out.

 

[evernessince]

Wow let me guess... HBM supply issues, or they can't afford to buy and package millions of HBM chips coz they're low on money. The funny thing is, their Vega10 card at 1200MHz is as fast as a GTX1070 that uses regular GDDR5 memory. Even if they have a Vega10 card at 1500MHz, it's gonna be roughly as fast as the GTX1080, which costs way less to produce and is almost a year old. Also, Volta is on the way, and is probably gonna be faster and more efficient than Pascal.

This is actually not true, AMD actually has a decent amount of cash in the bank right now (especially if you consider Ryzen's success).

Just like to point out, If AMD wanted a card with 1070 performance they would have just rebranded the Fury X. It makes ZERO sense that Vega would be at that performance level unless you are talking about a lower end Vega SKU.

AMD already has a Vega based professional card releasing at 1,500 MHz so we already know they can get much more than 1,200 out of these chips.

 

[G33k2Fr34k]

What gives you that idea? If they have the TDP set super low (exactly how the Nano was setup) the card will perform completely different than a card with a high clockspeed and higher TDP.

Remember the R9 Nano had a GPU clock of 1000mhz, 4096cu's and 4gb of 500mhz HBM, the Fury X had a GPU clock of 1050mhz, 4096cu's and 4gb of 500mhz HBM. 50mhz represented a 15% performance boost? No the 100w of TDP difference did. The nano would flag as a 1000mhz card in 3dmark, but only completed a tiny percent of the tests at that clockspeed, most of the time it spent running 875-950mhz, with dips to 700mhz.

If this SKU fills that same niche, expect a super low (100-150w TDP) and the card to only be running in the 1000-1100mhz range most of the time. There is already rumor of a very similar 3 card range as the Fury lineup.

AMD should sell their graphics division to Intel and their CPU division to Nvidia so that we can have some competition in both the CPU and GPU markets.

 

[cdawall]

AMD should sell their graphics division to Intel and their CPU division to Nvidia so that we can have some competition in both the CPU and GPU markets.

AMD already has some IP for GPU's with intel.

 

[G33k2Fr34k]

This is actually not true, AMD actually has a decent amount of cash in the bank right now (especially if you consider Ryzen's success).

Just like to point out, If AMD wanted a card with 1070 performance they would have just rebranded the Fury X. It makes ZERO sense that Vega would be at that performance level unless you are talking about a lower end Vega SKU.

AMD already has a Vega based professional card releasing at 1,500 MHz so we already know they can get much more than 1,200 out of these chips.

Your argument could easily apply to the RX 480 and R9 390. AMD could've just re-branded the R9 390 instead of releasing the RX480, which is generally as fast as the R9 390.
Also, Ryzen is not exactly a success. It's more of a mess that AMD pretends they can fix. Ryzen CPUs blow for gaming due to architectural factors (huge CCS-to-CCX latency) and not because of games needing to be patched.

 

[cdawall]

Your argument could easily apply to the RX 480 and R9 390. AMD could've just re-branded the R9 390 instead of releasing the RX480, which is generally as fast as the R9 390.
Also, Ryzen is not exactly a success. It's more of a mess that AMD pretends they can fix. Ryzen CPUs blow for gaming due to architectural factors (huge CCS-to-CCX latency) and not because of games needing to be patched.

They already rebranded a card to make the 390 it was called the 290. The 480 consumes roughly half the power the 290/390 consume. A rebrand even on a die shrink would not have fixed that. There is a reason the 480 is only 3/4th's of the GPU that was the 290/390.

 

[G33k2Fr34k]

They already rebranded a card to make the 390 it was called the 290. The 480 consumes roughly half the power the 290/390 consume. A rebrand even on a die shrink would not have fixed that. There is a reason the 480 is only 3/4th's of the GPU that was the 290/390.

My argument still applies: This new Vega10 GTX1070 competitor could have 150W TDP, which makes it a more efficient card than the 275W FuryX.

 

[R4E3960FURYX]

Do not hurry up to buy RX VEGA ? 6-9 Month later must be OK .

They need driver optimization a lot. Like AMD previous gen R9 Fury X 1 year later performance gain 30% .

 

[Camm]

Also, Ryzen is not exactly a success. It's more of a mess that AMD pretends they can fix. Ryzen CPUs blow for gaming due to architectural factors (huge CCS-to-CCX latency) and not because of games needing to be patched.

And yet performance says completely otherwise - and in games where performance is odd, we've already seen that can be patched (if you are really that desperate).

Lastly - get used to it. As core counts increase, even Intel will move to multiple complexes because trying to maintain cache coherency across multiple cores chews die space and power.

 

[cdawall]

My argument still applies: This new Vega10 GTX1070 competitor could have 150W TDP, which makes it a more efficient card than the 275W FuryX.

It is going to be more efficient than the fury x. The pipeline structure is different as well. You haven't made any point on that at all.

 

[sergionography]

Your argument could easily apply to the RX 480 and R9 390. AMD could've just re-branded the R9 390 instead of releasing the RX480, which is generally as fast as the R9 390.
Also, Ryzen is not exactly a success. It's more of a mess that AMD pretends they can fix. Ryzen CPUs blow for gaming due to architectural factors (huge CCS-to-CCX latency) and not because of games needing to be patched.

No it doesnt because polaris is inexpensive to make with its modest die size and is more efficient making it suitable for lower tdp areas like mobile. Also polaris was basically designed as a mid point release for multiple reasons. For example The main customer initially for polaris was sony for the ps4 pro. They had the earliest editions of it running at below 1ghz. So in other words polaris was a high volume chip so it was a no brainer to design.

 

[RejZoR]

Your argument could easily apply to the RX 480 and R9 390. AMD could've just re-branded the R9 390 instead of releasing the RX480, which is generally as fast as the R9 390.
Also, Ryzen is not exactly a success. It's more of a mess that AMD pretends they can fix. Ryzen CPUs blow for gaming due to architectural factors (huge CCS-to-CCX latency) and not because of games needing to be patched.

And this is why Ryzen's won't sell as they could. Idiots blowing "Ryzen can't run games well" out of all possible and impossible proportions. Which is a total lie. Apart from 3 games that bottleneck on any CPU anyway compard to others, Ryzen CPU's can EASILY be gaming CPU's. Same as drama around Windows Vista, just 100x worse because it's even more unjustified bashing of a great product. Only thing I acknowledge are the launch issues, but quite frankly, it has been dramatized again even though same problems also appear om Intel platform...

 

[ZoneDymo]

Coincidentally, both Vega and their last couple of iterations of GCN chips were entirely designed by their Chinese team. That's what you get for going cheap AMD. They should fire that retard Raja Kadouri immediately, along with their Chinese design team, and then hire a North American graphics IP design team.

Wow, reported for that bunch of hate spewing ableism/racist garbage.

 

[G33k2Fr34k]

And this is why Ryzen's won't sell as they could. Idiots blowing "Ryzen can't run games well" out of all possible and impossible proportions. Which is a total lie. Apart from 3 games that bottleneck on any CPU anyway compard to others, Ryzen CPU's can EASILY be gaming CPU's. Same as drama around Windows Vista, just 100x worse because it's even more unjustified bashing of a great product. Only thing I acknowledge are the launch issues, but quite frankly, it has been dramatized again even though same problems also appear om Intel platform...

I'm not bashing it, I'm stating the facts. Ryzen is actually pretty competitive for non-gaming tasks. The reason it sucks for gaming is because game workloads can be made up of thousands of simultaneous interdependent threads that share data and migrate across CPU cores all the time. Having a cross CCX latency that is 3 times as high as typical core-to-core latency causes significant slowdowns, which is what's happening with Ryzen.

Here, check for yourself gaming vs non-gaming 1500X at 3.5GHz vs 7700K at 3.5GHz.
https://arstechnica.com/gadgets/2017/05/amd-ryzen-5-review-1600x/2/

 

[G33k2Fr34k]

Wow, reported for that bunch of hate spewing ableism/racist garbage.

Why is that racist and what happened to free speech?

 

[G33k2Fr34k]

And yet performance says completely otherwise - and in games where performance is odd, we've already seen that can be patched (if you are really that desperate).

Lastly - get used to it. As core counts increase, even Intel will move to multiple complexes because trying to maintain cache coherency across multiple cores chews die space and power.

If it can be patched so easily, then why are Ryzen CPUs still under-performing badly in most games? It's not simple you know. The CPUs have been out for 3 months and they still haven't fixed their bad gaming performance problem. The problem is not software, it's the Ryzen CPUs. More specifically, the problem is the cross CCX latency in all current Ryzen CPUs.

 

[the54thvoid]

Why is that racist and what happened to free speech?

Calling Raja a retard and saying the American design team are superior to the Chinese. Not racist really, just a bit silly.

 

[RejZoR]

I'm not bashing it, I'm stating the facts. Ryzen is actually pretty competitive for non-gaming tasks. The reason it sucks for gaming is because game workloads can be made up of thousands of simultaneous interdependent threads that share data and migrate across CPU cores all the time. Having a cross CCX latency that is 3 times as high as typical core-to-core latency causes significant slowdowns, which is what's happening with Ryzen.

Here, check for yourself gaming vs non-gaming 1500X at 3.5GHz vs 7700K at 3.5GHz.
https://arstechnica.com/gadgets/2017/05/amd-ryzen-5-review-1600x/2/

And which entirely disappears if you run a fast RAM. And by fast I mean 3600 MHz or more. Which is fast, but certainly not the fastest available (or the most expensive). And even without this, it's still plenty fast.

It's so funny how people constantly have 60fps as some sort of performance level bar, but when games on Ryzen are hitting "just" 120 instead of 140fps on 7700k, the world instantly implodes. It's really so bizarre...

You're again blowing the CCX thing way out of proportions. Not to mention AMD CPU actually has lower latency within each CCX than Intel has across all cores. But no one seems to mention that because dramatizing around out-of-CCX is more interesting...

 

[Octopuss]

Why is that racist and what happened to free speech?

You shouldn't be allowed any free speech with so dumb posts and super retarded name.

 

[G33k2Fr34k]

You shouldn't be allowed any free speech with so dumb posts and super retarded name.

TRIGGERED