AMD CEO Speaks with Jim Cramer About the "Secret Sauce" Behind its Giant-Killing Spree

[First Strike]

The note on where they would gain is proberly the compute area, as AMD cards favor compute related tasks.

Actually, nope. Throughput on paper is one thing, actual throughput in real-world workload is another (cache-miss, warp latency, etc). Run some dgemm test on both cards you will see. Not to mention CUDA and its tool chain are much easier to use.
The only problem is that NV compute card with proper double-precision capabilities is so much more expensive. But for deep learning uses which only require FP32 or lower precision, I haven't seen a single lab that uses AMD card. For enterprise segment, AMD's MI25 hasn't found a single customer yet.

 

[notb]

There are already ray tracing benchmarks that a simple vega 64 beats GV100.

Show me.

Here's some test of AMD's own ProRender from just few months ago. Vega64 lost to Titan Xp and was just 15% faster than 1080Ti.
https://techgage.com/article/performance-testing-amds-radeon-prorender-in-autodesk-3ds-max/
Clearly Vega shines in rendering compared to how it performs in games, but that isn't RTX-level for sure.

Remember, that before RTX came RTRT wasn't really considered as a thing in gaming. It's just way too slow. That's why many people on this forum never heard of it.
However, it's not a new solution outside of gaming.

When earlier this year Nvidia announced that they're working on a new RTRT solution, AMD quickly answered saying that they'll improve RTRT implementation in their ProRender.
But Nvidia wasn't talking about a software approach, but about the ASIC for RTX.
I assume AMD has known about RTX for a while. They don't have a hardware answer yet, so they went for this ProRender thing as a temporary marketing solution.
But I'm pretty sure they're working on a hardware RTX competitor as well. They'll need it to keep up.
Sure, we may have doubts about RTRT popularity in gaming in next few years (well, I sure hope it stays).
But if Nvidia makes an RT accelerator based on this tech, they'll quickly eat whatever AMD still has in the 3D business.

BTW: Nvidia has just announced a dedicated Tensor accelerator for machine learning.
This means that until AMD shows their product (there have been rumors that they're working on a Nv Tensor alternative), everything on page is obsolete:
https://www.amd.com/en/graphics/servers-radeon-instinct-deep-learning

Now I do expect Nvidia to be absurdly more efficient at it from the get go, but amd and brute force are definitely friends of old.

As I said: we can expect AMD to be slightly more efficient in RT on general cores (GCN vs CUDA), but this is nowhere near the jump an ASIC gives.
BTW: I've seen Vega doing RTRT - I doubt it would be enough for Quake II.

 

[lexluthermiester]

Nearly spit up my soda there, so much bullshit there. In some alternate reality, if amd never sold off Adreno she'd be singing a very different tune. The fact is AMD is no longer in the position to invest mobile, it's not a decision when you simply can't.

Incorrect. AMD is in a perfect position to offer mobile products. But would it be wise in an over-saturated market? They made a choice and it is working for them.

 

[btarunr]

Wait...What???
nVidia?? On which front? On most sexy office ladies, or in the bowling championship??

The following is just for desktop discrete graphics cards (does not include APU/IGPs, console SoCs, or even the tons of mobile GPUs AMD is shipping to Apple for its Macbooks and MacPros, sauce):

AMD certainly has a larger slice of the GPU market than its share of the CPU market. It doesn't matter if people are gaming, mining, or doing something kinky with the cards they bought.

So yeah, despite the fact that it lacks a GTX 1080 Ti-competitor, AMD does threaten NVIDIA's bottom-line as of now.

 

[notb]

So yeah, despite the fact that it lacks a GTX 1080 Ti-competitor, AMD does threaten NVIDIA's bottom-line as of now.

And yet you're providing us with this very weird text about how AMD is fighting the "giants". I'll keep calling it advertorial. Or at least: very bad journalism.
Also, you're going on about AMD threatening Nv's bottom-line like if Nv was a huge company and AMD was an ambitious newbie. :-D

As of 2017 Nvidia's revenue is less than twice as much as AMD's (9.7 vs 5.3 bln USD). These are fairly similar companies and they're just competing in a duopoly market - on equal terms.

 

[hat]

And yet you're providing us with this very weird text about how AMD is fighting the "giants". I'll keep calling it advertorial. Or at least: very bad journalism.
Also, you're going on about AMD threatening Nv's bottom-line like if Nv was a huge company and AMD was an ambitious newbie. :-D

As of 2017 Nvidia's revenue is less than twice as much as AMD's (9.7 vs 5.3 bln USD). These are fairly similar companies and they're just competing in a duopoly market - on equal terms.

And how many years has Intel been shitting on AMD? They've made an explosive comeback on all fronts of the CPU market. Their stock price has went up, what, six fold? They're hot on Intel's heels and now Intel is shitting themselves rather than shitting on AMD. After years of what can be accurately be called stagnation, they've rushed 6 and even 8 core chips out the door on the mainstream platform, while pushing clockspeeds up as well, even relying on solder again after years of toothpaste, and they're scrambling at red alert to get something bigger and badder out the door, even relying on TSMC to produce stuff for them, since their own foundries are a mess.

AMD could have very easily been called the little guy for years. Now the little guy is not so little anymore. They've got their big brother sweating like a pig to stay on top. That is neither an advertorial, nor bad journalism. That is simply what is.

 

[las]

And Nvidia sits around 17-18%. Intel rules the whole market with 70%. When comparing only NV and AMD, it's around 31-69%. AMD was never ever at 50% in share. Closest was a 44% in 2010 and their lowest was 18% in 2015 Q2. If we compare with the latest result after exactly 3 years, 31% (nearly double of the 2015 Q2 data) doesn't seem that bad.

Obviously talking about gaming segment here. Nvidia sits close to 80% according to Steam HW Survey.

25% to 30% market GPU share is hardly "lost".

More like 15% when looking at gaming segment. Steam HWS.

I hope 7nm will change this. Have not had an AMD GPU since my 7970, which IMO was their last really good GPU.

Tech aside Lisa Su has done a spectacular job turning AMD around. CEO of the decade!

She didn't do jack. She is cringe. I remember her fucking up the Fury X release saying it was an Overclockers dream. Biggest lie ever.

You can thank Jim Keller instead. The brain behind Ryzen.

Now we just need GPU competition again.

 

[medi01]

Their stock price has went up, what, six fold?

From $1.85 to $30, sixteen times more.

More like 15% when looking...

Like, freaking, really?

 

[notb]

And how many years has Intel been shitting on AMD?

I'm talking about Nvidia, not Intel. Also, what do you mean by "shitting"?

They've made an explosive comeback on all fronts of the CPU market. Their stock price has went up, what, six fold? They're hot on Intel's heels and now Intel is shitting themselves rather than shitting on AMD.

AMD did a nice comeback on product quality. Sales are another thing entirely.

Just a short reminder to be sure we're discussing the same thing. Stock price comes from future earnings (mostly*). Agree?
You're making a product. You're earning 1 cent per item.
You fire your cleaning lady. Now you're earning 6 cents per item.
You didn't improve your product. You didn't improve the sales. Your stock price went up sixfold.

*) there's also some minimal stock price that reflects assets (both physical and intangible)
AMD was unprofitable for a long time and they had financial losses in their forecasts.
Now AMD is making a profit, so the price reflects the expected cashflow.
It doesn't really matter if it went up 6-fold since 2014, since the level you compare to has no meaning profit-wise. What will matter, is what happens in 2018 and 2019, i.e. whether they'll manage to match the forecast profitability or not.

AMD could have very easily been called the little guy for years. Now the little guy is not so little anymore.

Depends how you define "little". The technology behind their products is better, but it's not translating into sales.

That is neither an advertorial, nor bad journalism. That is simply what is.

This text is clearly one-sided and not really written to a level I'd expect from a news site. Unless we agree this is an opinion site.
Seriously:
"Intel, which is saddled with not just a dated CPU architecture peppered with security holes"

 

[hat]

I'm talking about Nvidia, not Intel. Also, what do you mean by "shitting"?

AMD did a nice comeback on product quality. Sales are another thing entirely.

Just a short reminder to be sure we're discussing the same thing. Stock price comes from future earnings (mostly*). Agree?
You're making a product. You're earning 1 cent per item.
You fire your cleaning lady. Now you're earning 6 cents per item.
You didn't improve your product. You didn't improve the sales. Your stock price went up sixfold.

*) there's also some minimal stock price that reflects assets (both physical and intangible)
AMD was unprofitable for a long time and they had financial losses in their forecasts.
Now AMD is making a profit, so the price reflects the expected cashflow.
It doesn't really matter if it went up 6-fold since 2014, since the level you compare to has no meaning profit-wise. What will matter, is what happens in 2018 and 2019, i.e. whether they'll manage to match the forecast profitability or not.

So AMD made some killer hardware, market share goes up, stock price goes up, but they have no sales. And by Intel "shitting" on AMD, come on... you couldn't even call AMD a competitor until recently with a straight face. Most of us here know how bad their offerings were compared to Intel, until Ryzen showed up.

Depends how you define "little". The technology behind their products is better, but it's not translating into sales.

This text is clearly one-sided and not really written to a level I'd expect from a news site. Unless we agree this is an opinion site.
Seriously:
"Intel, which is saddled with not just a dated CPU architecture peppered with security holes"

By little, I mean basically what I said above... their products were terrible, hence terrible market share/stock price. Intel was sitting pretty for years with minimal effort. Now that they actually have a good product, and Intel is scrambling to maintain their supremacy, that means AMD isn't so much of a little guy anymore. As for your comment about this line:

"Intel, which is saddled with not just a dated CPU architecture peppered with security holes"

I mean, is it not true? Whiskey lake is still basically Skylake. It finally has some security holes plugged in hardware, but not all. Sure, AMD has their problems with security too, as does just about anybody who made a processor in the last 20 years, but based on what we know, Intel's are worse and they have more of them.

 

[lexluthermiester]

More like 15% when looking at gaming segment. Steam HWS.

Steam's stats are not the end-all be-all of hardware info, and even that isn't completely accurate.

 

[StrayKAT]

Steam's stats are not the end-all be-all of hardware info, and even that isn't completely accurate.

Is that all AMD GPUs? Because that's crazy... and lame. These cards aren't even that different and people are treating Radeon like it has AIDs or something. It's stupid.

 

[lexluthermiester]

Is that all AMD GPUs? Because that's crazy... and lame. These cards aren't even that different and people are treating Radeon like it has AIDs or something. It's stupid.

Exactly. I don't get it either. Basically chalking it up to fanboyism from people unable or unwilling to see the big picture.

 

[Totally]

Incorrect. AMD is in a perfect position to offer mobile products. But would it be wise in an over-saturated market? They made a choice and it is working for them.

o rly? What do they have that Nvidia, who has tried to get a toehold but inevitably failed, doesn't?

 

[mtcn77]

o rly? What do they have that Nvidia, who has tried to get a toehold but inevitably failed, doesn't?

Namely, a profitable semi-custom devision. Name one product that flew on Tegra's footing. I'll wait...

 

[Totally]

Namely, a profitable semi-custom devision. Name one product that flew on Tegra's footing. I'll wait...

You're speaking like Nvidia doesn't have one. Keeping waiting until you understand the meaning of word "failed." Name one successful product that AMD currently has in the mobile space. I'll wait...

 

[mtcn77]

You're speaking like Nvidia doesn't have one. Keeping waiting until you understand the meaning of word "failed." Name one successful product that AMD currently has in the mobile space. I'll wait...

If you read my post, you would know semi-custom 'SoCs' are contemporary, but 'mobile' isn't. This is asymmetric core design principle at work, straight from the ARM playbook. All cores aren't equal.

 

[notb]

So AMD made some killer hardware, market share goes up, stock price goes up, but they have no sales.

OK. So we live surrounded by this idea that AMD market share goes up thanks to Zen. Everyone talks about it. Stock price goes up. AMD does well in "top CPUs" at Amazon and at Mindfactory.de. You've even read something of them getting contracts for Amazon datacenters. So it must be true, right?
Now take a deep breath and remind yourself the last time you've seen sales statistics that confirm this thesis.

And by Intel "shitting" on AMD, come on... you couldn't even call AMD a competitor until recently with a straight face. Most of us here know how bad their offerings were compared to Intel, until Ryzen showed up.

You think too much about subjective product quality and too little about sales. Even if AMD's desktop CPU market share is now 12% instead of 8% or something like that. They're still small compared to Intel. Nothing changed here.
In fact, the limited figures we have don't add up very well. AMD was selling cheap CPUs for the last few years. Ryzen stuff is twice as expensive. It suggests that, while revenue is clearly going up, volume remained on similar level.

As for your comment about this line:
"Intel, which is saddled with not just a dated CPU architecture peppered with security holes"
I mean, is it not true?

I'm talking about ethics and objectivity. About how this is presented in a text.
Intel's architecture is old, but is it dated? Most people here agree that it's still more compelling than Ryzen. Even the hardcore AMD fanboys say that "we're almost there!" or "the difference is very small anyway".
Intel's architecture has security holes, but which isn't? Intel is at least being properly tested (because of market share in servers).

I'm not saying that the author is an AMD supported (not here - in next paragraph). He might be just an enthusiast who admires AMD's work (innovative, very enthusiast/gamer-oriented) and despise Intel's (conservative, enterprise-oriented, somehow dull and slow). But the way he shows this is below the quality that I'd like to see from a journalist. I know this is not The Times, but we're allowed to expect some professionalism, right?

 

[hat]

I'm not really sure why you seem to like to keep repeating the same "AMD has no sales" comments. Clearly they do, else all the positive stats we've seen from AMD would be shit. They're not. Market share and stock prices do not move that much because 20 dudes on TPU bought a Ryzen CPU.

As for the journalism comments... I for one would find the words "old" and "dated" synonymous. And the architecture is old. Whiskey lake is basically Skylake 4. Such comments are not spared for AMD either, with their long history of re-hashing GCN. As for the number of people who find it more compelling than Ryzen? We've got a fair amount on either side. Saying the architecture is old and peppered with security holes is not only true, it's an advantage AMD currently enjoys over Intel. It's likely helped them gain some sales in the server market.

 

[InVasMani]

They don't have a hardware answer yet, so they went for this ProRender thing as a temporary marketing solution.

PCI-E 4.0 just lockout SLI and make a damn quad/octa core GPU 3-slot GPU and you've got x4/x8 the performance of current Vega GPU's in a single card solution. Hell you could CF 2 of them and have x8/x16 the performance. Ray tracing is parallel friendly it scales far better than rasterization despite being slower on similar hardware. I hate to say it, but 4-16 Vega 56 GPU's at 7nm could crush pretty well at real time ray tracing and not be proprietary bs at the same time. If I'm AMD I'm just locking out SLI on PCI-E 4.0 motherboards especially if I plan to go multi-chip approach like Ryzen/Epyc.

BTW: I've seen Vega doing RTRT - I doubt it would be enough for Quake II.

Not sure the resolution, but AMD FirePro W8100 49.3 fps
https://amietia.com/q2pt.html

Vega would defiantly do better than that for the record. All they need is a multi-chip GPU that's like a quad-core VEGA and PCI-E 4.0 x16 and it would crush at Quake 2. Ray tracing is far more parallel friendly than rasterization. AMD should just lock out SLI on PCI-E 4.0 boards and capitalize on ray tracing's parallelism and they'll win that fight pretty easily. I mean besides one less license to pay for.

 

[notb]

I'm not really sure why you seem to like to keep repeating the same "AMD has no sales" comments. Clearly they do, else all the positive stats we've seen from AMD would be shit. They're not.

"Positive stats" aka financial statements?
Because I believe it's correct - that's why.
I've never said AMD has no sales. But it does seem like they sell more or less the same number of CPUs as before Zen - just at a higher price point. Numbers add up pretty well.

Market share and stock prices do not move that much because 20 dudes on TPU bought a Ryzen CPU.

So have you seen any market share information? Would you mind sharing it? Because I've seen none sensible enough.
But I do have a laugh once in a while looking back at the ecstasy PassMark created last year. Not to look very far:
https://www.techpowerup.com/234864/...gaining-market-share-vs-intel-thanks-to-ryzen

As for stock price... Why not care about it at all? This is not a financial forum and the crypto period shown financial knowledge is weak within the members.
Just a suggestion from me: notice how AMD stock price moves each time a dude from a brokerage firm says "7nm". No new tech, no new financial statement, no sales info. But a "recommendation" arrives and AMD stock jumps 5-10% daily.

Such comments are not spared for AMD either, with their long history of re-hashing GCN.

Are they? On TPU?

Saying the architecture is old and peppered with security holes is not only true, it's an advantage AMD currently enjoys over Intel.

So you believe there are less holes in Ryzen?
Are you aware of the fact that AMD still hasn't officially confirmed that Zen is resistant to Meltdown?
As for "old"... well. I doubt any great redesigning is needed in camp Intel. Their 10nm CPUs will be very similar. 5nm should be as well, if they ever happen.
A good product works for a long time. It just needs some tuning and a refresh once in a while. There's no reason why Skylake wouldn't live for another 3-4 generations.
AMD for some reason can't find a long-time recipe. Once in a while they make a big splash with something new which sometimes sticks (Zen) and sometimes doesn't (Vega). Thing is though, with this kind of approach, you never have enough money or manpower to push the idea further.
Just how long will the Ryzen party keep going? Will 7nm help at all?

I believe some people will keep calling Intel's arch "old" whatever happens.
Few weeks from now we'll have an "old", "dated" architecture 8C/16T doing 5GHz and a "new", "innovative" one bouncing off 4.5GHz.

It's likely helped them gain some sales in the server market.

*alleged* - unless you have any figures

 

[StrayKAT]

AMD for some reason can't find a long-time recipe. Once in a while they make a big splash with something new which sometimes sticks (Zen) and sometimes doesn't (Vega). Thing is though, with this kind of approach, you never have enough money or manpower to push the idea further.
Just how long will the Ryzen party keep going? Will 7nm help at all?

Vega isn't all that new. It's still a "Radeon".

And it isn't so radical that I can't run the same games or something on it.

 

[hat]

"Positive stats" aka financial statements?
Because I believe it's correct - that's why.
I've never said AMD has no sales. But it does seem like they sell more or less the same number of CPUs as before Zen - just at a higher price point. Numbers add up pretty well.

So have you seen any market share information? Would you mind sharing it? Because I've seen none sensible enough.
But I do have a laugh once in a while looking back at the ecstasy PassMark created last year. Not to look very far:
https://www.techpowerup.com/234864/...gaining-market-share-vs-intel-thanks-to-ryzen

As for stock price... Why not care about it at all? This is not a financial forum and the crypto period shown financial knowledge is weak within the members.
Just a suggestion from me: notice how AMD stock price moves each time a dude from a brokerage firm says "7nm". No new tech, no new financial statement, no sales info. But a "recommendation" arrives and AMD stock jumps 5-10% daily.

Are they? On TPU?

Almost seems like we're all going to have to be armed with a set of folders stashed away in the bookmarks full of source links to back up anything that's said these days...

So you believe there are less holes in Ryzen?
Are you aware of the fact that AMD still hasn't officially confirmed that Zen is resistant to Meltdown?
As for "old"... well. I doubt any great redesigning is needed in camp Intel. Their 10nm CPUs will be very similar. 5nm should be as well, if they ever happen.
A good product works for a long time. It just needs some tuning and a refresh once in a while. There's no reason why Skylake wouldn't live for another 3-4 generations.
AMD for some reason can't find a long-time recipe. Once in a while they make a big splash with something new which sometimes sticks (Zen) and sometimes doesn't (Vega). Thing is though, with this kind of approach, you never have enough money or manpower to push the idea further.
Just how long will the Ryzen party keep going? Will 7nm help at all?

I believe some people will keep calling Intel's arch "old" whatever happens.
Few weeks from now we'll have an "old", "dated" architecture 8C/16T doing 5GHz and a "new", "innovative" one bouncing off 4.5GHz.

*alleged* - unless you have any figures

AMD probably wouldn't "confirm" that their chips are resistant to meltdown because they can't promise that. Even Intel can't promise that. Their "well tested", as you say, platform only claims to have "mitigations" against such attacks. That means they're saying that they did something about it to reduce the likelihood of a successful attack, but the possibility isn't gone. This crap caught everyone off guard and the entire industry is reeling from it. If these vulnerabilities have shown me anything, it's that I can't expect anything to be secure. These vulnerabilities have existed for 20+ years, only to be unearthed very recently.

As for the old architecture, well, it is old... We're on Skylake 4 now. As far back as my memory goes, we had a few variations of Netburst, Core 2, Nehalem, then we had Sandy Bridge, and Ivy Bridge, which was reworked Sandy, Haswell and Broadwell, then... Skylake, 4 times over. At least with Skylake 1, we got the same little bit of performance increase and efficiency we saw since Sandy. Now we just get more cores lopped in, which, don't get me wrong, is a good thing, but it doesn't serve to increase performance where single thread performance is important. That's my problem with AMD right now (who does have a new architecture), they still haven't beaten Intel there.

I look forward to seeing what Zen 2 does... but I also remain wary of all these security flaws and crappy mitigations... so it's still pretty meh for me right now.

 

[StrayKAT]

Almost seems like we're all going to have to be armed with a set of folders stashed away in the bookmarks full of source links to back up anything that's said these days...



AMD probably wouldn't "confirm" that their chips are resistant to meltdown because they can't promise that. Even Intel can't promise that. Their "well tested", as you say, platform only claims to have "mitigations" against such attacks. That means they're saying that they did something about it to reduce the likelihood of a successful attack, but the possibility isn't gone. This crap caught everyone off guard and the entire industry is reeling from it. If these vulnerabilities have shown me anything, it's that I can't expect anything to be secure. These vulnerabilities have existed for 20+ years, only to be unearthed very recently.

As for the old architecture, well, it is old... We're on Skylake 4 now. As far back as my memory goes, we had a few variations of Netburst, Core 2, Nehalem, then we had Sandy Bridge, and Ivy Bridge, which was reworked Sandy, Haswell and Broadwell, then... Skylake, 4 times over. At least with Skylake 1, we got the same little bit of performance increase and efficiency we saw since Sandy. Now we just get more cores lopped in, which, don't get me wrong, is a good thing, but it doesn't serve to increase performance where single thread performance is important. That's my problem with AMD right now (who does have a new architecture), they still haven't beaten Intel there.

I look forward to seeing what Zen 2 does... but I also remain wary of all these security flaws and crappy mitigations... so it's still pretty meh for me right now.

Skylake is only a few years old. You're acting like it's the umpteenth variation of the Motorola 68k.

But they're not running out of ideas. If you really want something different, it depends on Windows (and it's users) moving forward along with them. Easier said than done.

 

[medi01]

So you believe there are less holes in Ryzen?
Are you aware of the fact that AMD still hasn't officially confirmed that Zen is resistant to Meltdown?

They "didn't" indeed, merely stating "Zero AMD Vulnerability (to meltdown) due to AMD Architecture difference"
AMD Releases response to meltdown and spectre exploits

So you believe there are less holes in Ryzen?

Just when one thought that post cannot get any worse...
That strange connection between FUD: Fear Uncertainty Doubt and amd haters.