News Sony Playstation 4 chip helped AMD avoid bankruptcy — exec recounts how 'Jaguar' chips fueled company's historic turnaround

[Admin]

AMD exec Renato Fragale recalls 'helping AMD avoid bankruptcy' when he managed the team that developed the PlayStation 4 processor.

Sony Playstation 4 chip helped AMD avoid bankruptcy — exec recounts how 'Jaguar' chips fueled company's historic turnaround : Read more

 

[d0x360]

Good times and not just because I bought a bunch of AMD stock when it was sub $2.

Nobody back then would believe how AMD is doing now if you told them about Zen, mi300, or how they are eating away Intel's market share in data center, desktop and now mobile is about to kick off too.

Amazing what some money for r&d combined with good leadership can do

 

[usertests]

RIP in peace Jaguar E-cores, you were too slow for this cruel world.

 

[bit_user]

RIP in peace Jaguar E-cores, you were too slow for this cruel world.

Yes, I was going to point out that these were AMD's E-cores!

They were small enough and low-power enough that you could fit 8 of them + a decent sized GPU on a single die, and that's what counted. After the PS3 and XBox 360, I think game devs were comfortable enough with multithreading that it was probably an acceptable tradeoff to get more slower cores vs. a few faster ones.

Remember, this was right in the middle of Intel's quad-core era. So, to have 8 slower cores wasn't such a bad tradeoff, even though aggregate throughput was still lower. If comparing against a mainstream Intel i7 (at least, from Nehalem [2008] to Kaby Lake [2017]), you had 8 threads either way, thanks to the i7's hyperthreading.

BTW, I think AMD's strategy of going all-in on Zen proved wise. If they had continued developing their E-cores, perhaps it would've unlocked some lower-cost markets for them, but the Zen series cores have generally been smaller than their Intel counterparts, which has helped AMD in partially addressing lower market tiers with it. And now they have the C-series.

 

[jeremyj_83]

Yes, I was going to point out that these were AMD's E-cores!

They were small enough and low-power enough that you could fit 8 of them + a decent sized GPU on a single die, and that's what counted. After the PS3 and XBox 360, I think game devs were comfortable enough with multithreading that it was probably an acceptable tradeoff to get more slower cores vs. a few faster ones.

Remember, this was right in the middle of Intel's quad-core era. So, to have 8 slower cores wasn't such a bad tradeoff, even though aggregate throughput was still lower. If comparing against a mainstream Intel i7 (at least, from Nehalem [2008] to Kaby Lake [2017]), you had 8 threads either way, thanks to the i7's hyperthreading.

BTW, I think AMD's strategy of going all-in on Zen proved wise. If they had continued developing their E-cores, perhaps it would've unlocked some lower-cost markets for them, but the Zen series cores have generally been smaller than their Intel counterparts, which has helped AMD in partially addressing lower market tiers with it. And now they have the C-series.

The AMD E-cores were vastly better than Intel's Atom cores at the time as well.

 

[bit_user]

The AMD E-cores were vastly better than Intel's Atom cores at the time as well.

Heh, the first couple generations of Atoms were in-order and had 2-way SMT. Pretty quickly, they both switched to out-of-order and dropped SMT.

 

[TerryLaze]

The AMD E-cores were vastly better than Intel's Atom cores at the time as well.

Depends on what you consider "at the time" ,the Atom that was released at the same time as the ps4 was much faster but we have no idea if that was already in the cards at the time sony and ms shopped around for a new cpu.

Not that it matters since intel would never sell their cores cheap enough for the console makers.

Anandtech: Intel's new Atom CPU beats AMD's Jaguar in performance

Anandtech have reviewed Intel's new Atom and it looks to finally deliver very competitive performance versus ARM processors at their trademark low power draws. What caught my eye is that it also beat the much higher power draw Jaguar used in the PS4 and XB1 in instructions per cycle (IPC). This...

www.neogaf.com

 

[bit_user]

Depends on what you consider "at the time" ,the Atom that was released at the same time as the ps4 was much faster but we have no idea if that was already in the cards at the time sony and ms shopped around for a new cpu.

Yeah, Silvermont had a process node advantage and I think is widely regarded as the first "good" Atom-class core. It dates back to when Intel was still playing for the phone/tablet SoC market.

BTW, here's the actual Anandtech article referenced in that forum post:

https://www.anandtech.com/show/7314/intel-baytrail-preview-intel-atom-z3770-tested
​

Also, that post quoted three single-threaded benchmarks. In 7-zip (not specified whether compression or decompression), the Atom was about 25.1% faster, at similar clocks. In Cinebench R11.5 (ST), they basically tied (if you look at the original Anandtech article, they also tie on Cinebench MT!). In the Chrome browser's performance on the Mozilla Kraken benchmark, Atom won by a whopping 41.6%. There are two other CPU benchmarks not quoted in that post, where the Atom's lead was smaller (but still sizeable).

Not that it matters since intel would never sell their cores cheap enough for the console makers.

I'd also like to point out that, until Alder Lake-N, Intel shipped their consumer-focused E-core SoCs in configurations consisting of a max of 4 cores. We can't know if they'd have gone up to 8 cores for these consoles. If not, then Jaguar would still take a multi-threaded performance lead.

 

[TerryLaze]

We can't know if they'd have gone up to 8 cores for these consoles. If not, then Jaguar would still take a multi-threaded performance lead.

That's not even the biggest issue, intel wouldn't be able to incorporate good enough graphics for a console.
They could now though.

 

[thestryker]

BTW, I think AMD's strategy of going all-in on Zen proved wise. If they had continued developing their E-cores, perhaps it would've unlocked some lower-cost markets for them, but the Zen series cores have generally been smaller than their Intel counterparts, which has helped AMD in partially addressing lower market tiers with it. And now they have the C-series.

You couldn't be more right about this one, especially in terms of business health (and market health as we're all benefiting from that choice).

I still think this is my favorite example of how far AMD came overall: https://www.anandtech.com/show/1633...-an-xbox-one-s-apu-the-chuwi-aerobox-review/4

 

[bit_user]

I still think this is my favorite example of how far AMD came overall: https://www.anandtech.com/show/1633...-an-xbox-one-s-apu-the-chuwi-aerobox-review/4

zOMG, look at this 2c/4t Zen 1 APU matching 8c Jaguar!

​

There might be even more stark examples, but I don't know how well-multithreaded most of the other tests are. Obviously, the single-threaded ones are a bloodbath for Jaguar.

 

[jeremyj_83]

Heh, the first couple generations of Atoms were in-order and had 4-way SMT (IIRC). Pretty quickly, they both switched to out-of-order and dropped SMT.

They were just in-order as Atom has never had SMT.

 

[jeremyj_83]

zOMG, look at this 2c/4t Zen 1 APU matching 8c Jaguar!

​

​

There might be even more stark examples, but I don't know how well-multithreaded most of the other tests are. Obviously, the single-threaded ones are a bloodbath for Jaguar.

IIRC the 300GE is Zen+. It is able to run Win 11 where as the 2400G, which is Zen 1, cannot. That said it is crazy how much faster the Zen CPUs are compared to Jaguar. Zen 2 CPUs in the PS5 and Xbox X are something like double to triple the performance of Jaguar clock for clock.

 

[bit_user]

They were just in-order as Atom has never had SMT.

Aw, c'mon man! This is provably false! Don't you ever think to fact-check yourself, before directly contradicting someone?

It Does Multiple Threads Though: The Case for SMT

"Despite being 2-issue, it's not always easy to execute two instructions from a single thread in parallel due to data dependencies between the two. Intel's solution to this problem was to enable SMT (Simultaneous Multi-Threading) on Atom (not all models unfortunately) to allow the concurrent execution of up to two threads. Welcome the return of Hyper Threading.

Remember the rule of thumb for power/performance tradeoffs? Intel's decision to enable SMT on Atom was the perfect example of just that. SMT increased power consumption by less than 20% on Atom, however it also yielded a 30 - 50% increase in performance on the in-order core. The decision couldn't be easier.

The Atom has a 32-entry instruction scheduling queue, but when running with SMT enabled each thread has its own 16-entry queue. The scheduler doesn't have to switch between threads each clock, it can do so intelligently, the only limitation is that it can only dispatch 2 ops per clock (since it is a 2-wide machine). If one thread is waiting on data to complete an instruction, on the next clock tick the scheduler can choose to dispatch an op from a separate thread that will hopefully be able to execute.

Making Atom multithreaded made perfect sense from a logical standpoint. The downside to an in-order core is that if there is an instruction that is waiting on data to begin execution the rest of the pipeline stalls while that dependency is resolved. The chances that you'll have two independent instructions from two independent threads both with misses in cache is highly unlikely."

Source: https://www.anandtech.com/show/2493/10
​

That was written by Anand Lal Shimpi, himself. I found it amusing that he posted it on April 2, 2008 12:05 AM EST, as if waiting until 5 minutes after midnight to keep it from being mistaken as an elaborate April Fools Day joke!

 

[jeremyj_83]

Aw, c'mon man! This is provably false! Don't you ever think to fact-check yourself, before directly contradicting someone?

​

It Does Multiple Threads Though: The Case for SMT​

​

"Despite being 2-issue, it's not always easy to execute two instructions from a single thread in parallel due to data dependencies between the two. Intel's solution to this problem was to enable SMT (Simultaneous Multi-Threading) on Atom (not all models unfortunately) to allow the concurrent execution of up to two threads. Welcome the return of Hyper Threading.​

​

Remember the rule of thumb for power/performance tradeoffs? Intel's decision to enable SMT on Atom was the perfect example of just that. SMT increased power consumption by less than 20% on Atom, however it also yielded a 30 - 50% increase in performance on the in-order core. The decision couldn't be easier.​

​

The Atom has a 32-entry instruction scheduling queue, but when running with SMT enabled each thread has its own 16-entry queue. The scheduler doesn't have to switch between threads each clock, it can do so intelligently, the only limitation is that it can only dispatch 2 ops per clock (since it is a 2-wide machine). If one thread is waiting on data to complete an instruction, on the next clock tick the scheduler can choose to dispatch an op from a separate thread that will hopefully be able to execute.​

​

Making Atom multithreaded made perfect sense from a logical standpoint. The downside to an in-order core is that if there is an instruction that is waiting on data to begin execution the rest of the pipeline stalls while that dependency is resolved. The chances that you'll have two independent instructions from two independent threads both with misses in cache is highly unlikely."​

​

Source: https://www.anandtech.com/show/2493/10​

​

That was written by Anand Lal Shimpi, himself. I found it amusing that he posted it on April 2, 2008 12:05 AM EST, as if waiting until 5 minutes after midnight to keep it from being mistaken as an elaborate April Fools Day joke!

Sorry I forgot that the first 3 or 4 versions had SMT. I didn't go and look because for at least a decade it hasn't had SMT. That said OoO added more performance to Atom than SMT.

 

[wingfinger]

If it wasn't for all do's consoles, we'd all be using intel graphics.

 

[bit_user]

Sorry I forgot that the first 3 or 4 versions had SMT.

Now that I've had occassion to look it up, I can say that Bonnell (2008) and Saltwell (2011) had SMT.

• https://en.wikichip.org/wiki/intel/microarchitectures/bonnell
• https://en.wikichip.org/wiki/intel/microarchitectures/saltwell

After that, Silvermont (2013) dropped it... except for the version of the Silvermont/Airmont core used in Xeon Phi (Knights Landing; 2016) that was modified to include both AVX-512 and SMT-4.

• https://en.wikichip.org/wiki/intel/microarchitectures/silvermont
• https://en.wikipedia.org/wiki/Xeon_Phi#Knights_Landing

That said OoO added more performance to Atom than SMT.

Yes, that's why they made the change.

Also, I'd point out that mobile-oriented cores uniformly seem to avoid SMT. Evidence is that it improves performance-per-area (PPA) but not performance per Watt. As such, it really appears to be increasingly relegated to servers, at this point. Intel's recent Lions Cove announcement underscores this distinction, as they said they completely removed SMT from the client cores but kept it in the server version.

I'd be remiss if I didn't mention GPUs - they're big into SMT. That's also consistent with Intel re-adding it to Silvermont/Airmont for Knights Landing.

 

[AgentJ24]

AMD exec Renato Fragale recalls 'helping AMD avoid bankruptcy' when he managed the team that developed the PlayStation 4 processor.

Sony Playstation 4 chip helped AMD avoid bankruptcy — exec recounts how 'Jaguar' chips fueled company's historic turnaround : Read more

This guy is overselling his help. Playstation did not save AMD.

The Rise, Fall and Revival of AMD

AMD is one of the oldest designers of large scale microprocessors and has been the subject of polarizing debate among technology enthusiasts for nearly 50 years. Its...

www.techspot.com

"By the end of 2016, the company's balance sheet had taken a loss for 4 consecutive years (2012's financials were battered by a $700 million GlobalFoundries final write off). Debt was still high, even with the sale of its foundries and other branches, and not even the success of the system package in the Xbox and PlayStation provided enough help."

 

[bit_user]

This guy is overselling his help. Playstation did not save AMD.

The Rise, Fall and Revival of AMD

AMD is one of the oldest designers of large scale microprocessors and has been the subject of polarizing debate among technology enthusiasts for nearly 50 years. Its...

www.techspot.com

"By the end of 2016, the company's balance sheet had taken a loss for 4 consecutive years (2012's financials were battered by a $700 million GlobalFoundries final write off). Debt was still high, even with the sale of its foundries and other branches, and not even the success of the system package in the Xbox and PlayStation provided enough help."

While it still took a Zen licensing deal with Zhaoxin to tide AMD over to the point where they started getting revenue from Zen, I think it's clear they wouldn't have lasted even that long, without that revenue from the consoles.

https://en.wikipedia.org/wiki/AMD–Chinese_joint_venture
​

 

[spongiemaster]

Good times and not just because I bought a bunch of AMD stock when it was sub $2.

Nobody back then would believe how AMD is doing now if you told them about Zen, mi300, or how they are eating away Intel's market share in data center, desktop and now mobile is about to kick off too.

Amazing what some money for r&d combined with good leadership can do

You left out the most important factor in AMD's turnaround. Intel getting overly aggressive with 10nm and screwing it up for 4 years. If Intel had stayed the course and continued hitting their tic tok roadmap, Zen would have been an interesting industry footnote like the Voodoo 6000 and AMD would not have survived.

 

[spongiemaster]

This guy is overselling his help. Playstation did not save AMD.

The Rise, Fall and Revival of AMD

AMD is one of the oldest designers of large scale microprocessors and has been the subject of polarizing debate among technology enthusiasts for nearly 50 years. Its...

www.techspot.com

"By the end of 2016, the company's balance sheet had taken a loss for 4 consecutive years (2012's financials were battered by a $700 million GlobalFoundries final write off). Debt was still high, even with the sale of its foundries and other branches, and not even the success of the system package in the Xbox and PlayStation provided enough help."

The quote contradicts your own point. They were losing money with Playstation and Xbox sales. Without the significant revenue from them, AMD doesn't make it to Zen.

 

[arnoL]

Thanks for this article. I always wondered how much the PS 4 helped keeping AMD afloat.Very instructive article Thanks again.

 

[greenreaper]

I had one of those Bobcats, in the X120e netbook I got for taking convention notes. It ended up being my main system for several years after my desktop tower died. Truly anaemic CPU performance with dual-core 1.6Ghz and minimal cache, but the iGPU and video decode was sufficient to keep me going until it died (even ran it for a while without the terrible screen once it broke, using the HDMI output and a DisplayLink dongle). Not bad for $440+extra RAM.