AMD Releases Q3 Earnings Results: Significantly Higher Net Income than Previous Quarters

[Prysin]

7 minutes ago, DocSwag said:

Why, were the APUs from 2012-2013 ok?

they suffered from the same issue as the ones from 2014, 2015, 2016, 2017 AND the ones to come for atleast the next 4 years.

[MyName13]

28 minutes ago, Prysin said:

they suffered from the same issue as the ones from 2014, 2015, 2016, 2017 AND the ones to come for atleast the next 4 years.

And that is?

[Prysin]

Just now, MyName13 said:

And that is?

if you have to ask, you need to go read up on how GPUs and CPUs work, basic computer functions, rock bottom basic computer design and last but not least, read in-depth articles from reputable sources.

 

Iv'e long since given up on trying to "educate" people. None of you want study the subject, you just want a specific answer for "A" and nothing surrounding it.

[Eduard the weeb]

AMD  being competitive doesn't compute.

[Devin92]

yet AMD stock shows a 11+% drop today. LOL. Guess investors thought zen will make some miracle.

[MyName13]

17 minutes ago, Prysin said:

if you have to ask, you need to go read up on how GPUs and CPUs work, basic computer functions, rock bottom basic computer design and last but not least, read in-depth articles from reputable sources.

 

Iv'e long since given up on trying to "educate" people. None of you want study the subject, you just want a specific answer for "A" and nothing surrounding it.

Yes, I will certainly study computer science and computer engineering just to find out the answer.

[Prysin]

20 minutes ago, Devin92 said:

yet AMD stock shows a 11+% drop today. LOL. Guess investors thought zen will make some miracle.

no, they have been shorting AMD, thinking they would conjure a loss. Instead AMD posted a profit.

 

net result?

 

"investors" has lost 50+ million dollars on betting that AMD would LOSE money this quarter...  Todays drop is just "salt"... investors got burnt on that stock, and now they want nothing to do with it for a while.

 

you have to remember that investors look more at TRENDS then at actual products. They dont read reviews or dissect the DIY PC community.

Instead they look at graphs of market share and trends of previous quarters.

 

Q1 and Q2 wasnt great, AMD still hasnt gained much market share from Intel nor Nvidia. Thus there is no positive trend... And investors generally dont know how good margins AMD have with their product. Thus they bet on what is likely, which was a loss.

[DocSwag]

27 minutes ago, Prysin said:

if you have to ask, you need to go read up on how GPUs and CPUs work, basic computer functions, rock bottom basic computer design and last but not least, read in-depth articles from reputable sources.

 

Iv'e long since given up on trying to "educate" people. None of you want study the subject, you just want a specific answer for "A" and nothing surrounding it.

Wow, that's really helpful.

 

At the moment electrical engineering is the subject I plan on going into in college, btw. Any information I learn is legitimately helpful.

 

Not sure what speicifcally you're referring to... memory bandwidth?

[Prysin]

2 minutes ago, DocSwag said:

Wow, that's really helpful.

 

At the moment electrical engineering is the subject I plan on going into in college, btw. Any information I learn is legitimately helpful.

 

Not sure what speicifcally you're referring to... memory bandwidth?

ding ding ding. Someone knows how computers work

 

yes, memory bandwidth. It was too low back then, its still far too low with DDR4.

 

To give you a hint; the GPU section in a 7850k, FLOPS wise, should match a R5 250..... The R5 250 had a memory bandwidth of 80 something GB/s... DDR4 dual channel wont get much better then 36-40GB/s (shared with CPU)... meaning you are STILL looking at below HALF the availible memory bandwith of the "dGPU"...

 

and 2012-2013 was even worse, as the defacto DDR3 memory back then was something like 800-1333Mhz... which is abysmally low. Even at DDR3 2400MHz you are STILL memory constrained so badly that the CPUs, even shitty old bulldozer CPUs, would virtually NEVER be the issue.

 

[SpaceGhostC2C]

19 hours ago, emosun said:

I'm not going to pretend intel doesn't take the same amount of time to make something that can be considered TRULY new.

But at least intel was doing SOMETHING. They were at least standing in the court practicing free throws while they waited for amd to show up. I don't think intel really minded that several year period of no competition as people paid premium prices. 

The thing is, AMD was also doing

19 hours ago, herman mcpootis said:

-take the same thing and add a 1% performance bump

-claim it performs 10% better than last gen

-add/remove a couple pins

-market as new architecture

-profit

as they launched Piledriver, Steamroller, and Excavator.

Not enough when your source is Bulldozer? OK, but that only gets solved with a completely new, from the ground up architecture. So we are back to the 4-year horizon, etc. Bottom line: everyone was "on the court practicing", but when you lag behind there's no way around making a substantial leap. It's not about anonymous internet posters giving AMD one-liner management advice...

 

9 minutes ago, Prysin said:

and 2012-2013 was even worse, as the defacto DDR3 memory back then was something like 800-1333Mhz...

800MHz is DDR2. APU DDR3 support was 1866 and up over the years till 2400, and 1600MHz was pretty much the standard for people buying DDR3 without a second glance already by the time the 3000 series APUs came out.

 

9 minutes ago, Prysin said:

which is abysmally low. Even at DDR3 2400MHz you are STILL memory constrained so badly that the CPUs, even shitty old bulldozer CPUs, would virtually NEVER be the issue.

What are you even talking about? AMD APUs consistently performed better in graphics than non-Iris Intel iGPUs, and always lagged in CPU intensive tasks.

For someone so harsh on the knowledge of others, you seem to be discussing a parallel universe.

[DocSwag]

15 minutes ago, Prysin said:

ding ding ding. Someone knows how computers work

 

yes, memory bandwidth. It was too low back then, its still far too low with DDR4.

 

To give you a hint; the GPU section in a 7850k, FLOPS wise, should match a R5 250..... The R5 250 had a memory bandwidth of 80 something GB/s... DDR4 dual channel wont get much better then 36-40GB/s (shared with CPU)... meaning you are STILL looking at below HALF the availible memory bandwith of the "dGPU"...

 

and 2012-2013 was even worse, as the defacto DDR3 memory back then was something like 800-1333Mhz... which is abysmally low. Even at DDR3 2400MHz you are STILL memory constrained so badly that the CPUs, even shitty old bulldozer CPUs, would virtually NEVER be the issue.

 

You know what would be really cool? If AMD made an APU with some HBM cache on board.

 

THAT, would be really awesome.

 

What is interesting to me though is the HBCC seems to help in memory bandwidth constrained scenarios. It might be that the HBCC will be very useful for APUs...

[Prysin]

1 minute ago, DocSwag said:

You know what would be really cool? If AMD made an APU with some HBM cache on board.

 

THAT, would be really awesome.

 

What is interesting to me though is the HBCC seems to help in memory bandwidth constrained scenarios. It might be that the HBCC will be very useful for APUs...

no it wouldnt. Too much power draw, too high latency... better off waiting for a new gen of stacked DRAM. HBM and HBM2 is too power hungry for APUs, and FAR too costly.

[Prysin]

13 minutes ago, SpaceGhostC2C said:

The thing is, AMD was also doing

as they launched Piledriver, Steamroller, and Excavator.

Not enough when your source is Bulldozer? OK, but that only gets solved with a completely new, from the ground up architecture. So we are back to the 4-year horizon, etc. Bottom line: everyone was "on the court practicing", but when you lag behind there's no way around making a substantial leap. It's not about anonymous internet posters giving AMD one-liner management advice...

 

800MHz is DDR2. APU DDR3 support was 1866 and up over the years till 2400, and 1600MHz was pretty much the standard for people buying DDR3 without a second glance already by the time the 3000 series APUs came out.

 

What are you even talking about? AMD APUs consistently performed better in graphics than non-Iris Intel iGPUs, and always lagged in CPU intensive tasks.

For someone so harsh on the knowledge of others, you seem to be discussing a parallel universe.

oh please,  And even on DIY desktops, APUs are memory constrained.

 

You are also trying to compare Intel to AMD here, a discussion nobody cares for. Intels iGPU is trash, the ONLY REASON that Iris pro is "faster" is because Intel has a huge L4 victim cache called "crystal well", that acts as a buffer between memory and GPU, and also Intel has FAR superior IMC, allowing for higher effective bandwidth.

 

And unfortunately for you, i OWN APUs, i OWN 2400MHz DDR3 memory and ive used APUs for gaming, ive overclocked them both on CPU and GPU and memory. Heck, i own more then one APU, i also own a FX 8320 and a Athlon X4 845 (Excavator). I know, far better then most, from practical research and testing how these things tick. Both with low and high bandwidth. Tight and loose memory timings.

 

Please educate yourself on actual reality. Just because the parts are availible doesnt mean they are purchased by consumers. We got 6+TB SSDs, on teh consumer market. Still not seeing many people put 6TB SSDs in their gaming rigs.

 

[MoonSpot]

5 minutes ago, Prysin said:

HBM and HBM2 is too power hungry for APUs, and FAR too costly.

In a desktop I'd tend to agree, though I wouldn't mind seeing it happen(Hoping markets of scale kick in somewhere to save us dem monies).  In a notebook, IDK, seems to be a winning combo as long as they're clocked optimally and not balls to the walls like we've been seeing for vega64.  It'd be weaker CPU wise, but can't see it doing anything less than pummeling intel iGPU graphics wise.

[Prysin]

Just now, MoonSpot said:

In a desktop I'd tend to agree, though I wouldn't mind seeing it happen(Hoping markets of scale kick in somewhere to save us dem monies).  In a notebook, IDK, seems to be a winning combo as long as they're clocked optimally and not balls to the walls like we've been seeing for vega64.  It'd be weaker CPU wise, but can't see it doing anything less than pummeling intel iGPU graphics wise.

it is potentially winning combo, but the power draw and cost will ban it from ANY notebook. Only desktop would allow for such retardedly high cost designs to be used.

 

AMD would be better off soldering a single 1GB GDDR5X chip to their notebooks and downclock it to 7-8000MHz. That is FAR more commercially defensible, seeing as a GDDR5X chip like the one found on the 1080Ti costs around 12-15$ a piece... HBM, even gen 1, is even more then that for a single 1GB stack. And 5X will use less power if you downclock it, allowing for a slightly tighter power target.

 

[SpaceGhostC2C]

8 minutes ago, Prysin said:

And unfortunately for you

Really, you need to call reality back asap 

[Prysin]

Just now, SpaceGhostC2C said:

Really, you need to call reality back asap 

You first, ill hold.

[MoonSpot]

12 minutes ago, Prysin said:

it is potentially winning combo, but the power draw and cost will ban it from ANY notebook. Only desktop would allow for such retardedly high cost designs to be used.

 

AMD would be better off soldering a single 1GB GDDR5X chip to their notebooks and downclock it to 7-8000MHz. That is FAR more commercially defensible, seeing as a GDDR5X chip like the one found on the 1080Ti costs around 12-15$ a piece... HBM, even gen 1, is even more then that for a single 1GB stack. And 5X will use less power if you downclock it, allowing for a slightly tighter power target.

 

The VRAM in and of itself isn't the only cost consideration here.  IDC if the VRAM is $75 or $120; its still, generally, considerably less than going with an i7 and 1060/1070 dGPU for laptops.

From what I remember seeing ryzen and vega are actually very efficient, when clocked correctly.  And HBM is far and away much more power efficient than GDDR.  I don't follow you on ether the power or cost arguments here.

 

:edit: That said, it all comes down to if the price/perf is comparable.  Think we're all hoping that HBM on an APU translates into some strength that can be easily leveraged.  But that is pretty much an unknown until they dawn.

[Prysin]

1 minute ago, MoonSpot said:

The VRAM in and of itself isn't the only cost consideration here.  IDC if the VRAM is $75 or $120; its still, generally, considerably less than going with an i7 and 1060/1070 dGPU for laptops.

From what I remember seeing ryzen and vega are actually very efficient, when clocked correctly.  And HBM is far and away much more power efficient than GDDR.  Is don't follow you on ether the power or cost arguments here.

actually HBM isnt really that much more effective, not anymore.

It was effective at the time of launch, but GDDR5X, and soon GDDR6 is matching or beating it, both at latency and bandwidth. HBM still has some advantages, but the cost of implementing it, be it directly on die, or as a standalone stack, is too high.

 

Directly on die means bigger dies, lower margins, higher prices and worse yields.

Standalone means a "expensive" interposer and the usualy supply issues we got today.

 

The power cost is still SIGNIFICANT....

a high end notebook today has a power budget (total budget) of around 70-90w, that INCLUDES the motherboard, SSD/HDD, keyboard, USB power, screen, fans and other systems. just adding ONE stack of HBM will cost you 10W of power to gain any reasonable bandwidth out of it. That is 10w you DONT have. seriously, you dont.

GDDR5X will add around 5W if you downclock it, and even at 5w, you are pushing it.

 

The OEMs design their own motherboards, they will NEVER add any feature that they cannot be sure to have 2x margins on.

There is no profit to be had on "high performance APUs", because at that point, you can cut costs, slap in a lower end "dGPU", get the OEM pure Ryzen CPU, sell the whole shit for 50$ more, but with 2x perf of high end APU

 

 

The only "desktop" solution for APUs is quad channel DDR4, and i dont see that happening any time soon

[DocSwag]

44 minutes ago, Prysin said:

no it wouldnt. Too much power draw, too high latency... better off waiting for a new gen of stacked DRAM. HBM and HBM2 is too power hungry for APUs, and FAR too costly.

Cost is definitely a potential issue, but I think it's feasible. Take one stack of HBM2, maybe downclock it to 500 MHz, and you got yourself 2 GBs (or maybe even 4GB) of 128 GB/s cache on board that should still be pretty low latency and not consume too much power (I would assume it'd be under 10w, maybe even 5).

 

It would cost a fair bit, but the APUs are already pretty small so the interposer shouldn't be too large and costly; maybe $30-40, and I'd assume the HBM could be bought for $30-40. Yes, that's still pretty costly, but it's totally feasible and could be something seen in the future.

[emosun]

5 hours ago, cj09beira said:

threadripper perf is totally not old news

no just the price isn't old news , the performance is. 

[AnonymousGuy]

Why the fuck is an entire page of this thread dedicated to back and forth about APU performance?

[MoonSpot]

20 minutes ago, AnonymousGuy said:

Why the fuck is an entire page of this thread dedicated to back and forth about APU performance?

APUs are product that AMD sells.  The thread is about AMD making money.  Miracle APUs could make lots of monies.

Seems fairly on topic for a tech forum, perhaps not if it were a day traders stocks discussion forum though.

How did you want to thread to progress?  You can always add to the discussion or ask any questions you might have to fuel the discussion in the manner of your choosing.

 

Thank you for your patronage, and enjoy your stay!

[AnonymousGuy]

14 minutes ago, MoonSpot said:

APUs are product that AMD sells.  The thread is about AMD making money.  Miracle APUs could make lots of monies.

Seems fairly on topic for a tech forum, perhaps not if it were a day traders stocks discussion forum though.

How did you want to thread to progress?  You can always add to the discussion or ask any questions you might have to fuel the discussion in the manner of your choosing.

 

Thank you for your patronage, and enjoy your stay!

Let's discuss the pandas because AMD exists on the earth and pandas also exist on the earth.  Ergo henceforth, logic!

 

Fucks sake take this bullshit to PM or a separate thread. AMD's stock went down 13% today as a result of their earnings report and none of you fuckwits even noticed.

[ARikozuM]

18 minutes ago, MoonSpot said:

You can always add to the discussion or ask any questions you might have to fuel the discussion in the manner of your choosing.

Agreed. 

 

So how about those Lakers? I hear Ken Griffey, Jr. is being drafted for the defense.