Albert Penello said:

At Microsoft, we have a position called a "Technical Fellow" These are engineers across disciplines at Microsoft that are basically at the highest stage of technical knowledge. There are very few across the company, so it's a rare and respected position.

We are lucky to have a small handful working on Xbox.

I've spent several hours over the last few weeks with the Technical Fellow working on our graphics engines. He was also one of the guys that worked most closely with the silicon team developing the actual architecture of our machine, and knows how and why it works better than anyone.

So while I appreciate the technical acumen of folks on this board - you should know that every single thing I posted, I reviewed with him for accuracy. I wanted to make sure I was stating things factually, and accurately.

So if you're saying you can't add bandwidth - you can. If you want to dispute that ESRAM has simultaneous read/write cycles - it does.

I know this forum demands accuracy, which is why I fact checked my points with a guy who helped design the machine.

This is the same guy, by the way, that jumps on a plane when developers want more detail and hands-on review of code and how to extract the maximum performance from our box. He has heard first-hand from developers exactly how our boxes compare, which has only proven our belief that they are nearly the same in real-world situations. If he wasn't coming back smiling, I certainly wouldn't be so bullish dismissing these claims.

I'm going to take his word (we just spoke this AM, so his data is about as fresh as possible) versus statements by developers speaking anonymously, and also potentially from several months ago before we had stable drivers and development environments.

Click to expand...

Albert Penello said:

At Microsoft, we have a position called a "Technical Fellow" These are engineers across disciplines at Microsoft that are basically at the highest stage of technical knowledge. There are very few across the company, so it's a rare and respected position.

We are lucky to have a small handful working on Xbox.

I've spent several hours over the last few weeks with the Technical Fellow working on our graphics engines. He was also one of the guys that worked most closely with the silicon team developing the actual architecture of our machine, and knows how and why it works better than anyone.

So while I appreciate the technical acumen of folks on this board - you should know that every single thing I posted, I reviewed with him for accuracy. I wanted to make sure I was stating things factually, and accurately.

So if you're saying you can't add bandwidth - you can. If you want to dispute that ESRAM has simultaneous read/write cycles - it does.

I know this forum demands accuracy, which is why I fact checked my points with a guy who helped design the machine.

This is the same guy, by the way, that jumps on a plane when developers want more detail and hands-on review of code and how to extract the maximum performance from our box. He has heard first-hand from developers exactly how our boxes compare, which has only proven our belief that they are nearly the same in real-world situations. If he wasn't coming back smiling, I certainly wouldn't be so bullish dismissing these claims.

I'm going to take his word (we just spoke this AM, so his data is about as fresh as possible) versus statements by developers speaking anonymously, and also potentially from several months ago before we had stable drivers and development environments.

Click to expand...

Curufinwe said:

Where did you get Sony's final CPU clock speed from and where did you get the information that the PS4 has a maximum of 10gb/s coherent bandwidth?

Click to expand...

Durante said:

Exactly. Having a unified memory space with (potentially) cache-coherent access is great, but it's not likely to make up for a significant overall CPU/GPU performance discrepancy in a majority of real-world gaming scenarios.

Also, generally when you want to do GPGPU, it's because you have a highly parallel problem with high bandwidth requirements. In such cases you wouldn't want to go through the cache-coherent bus in PS4 anyway (as that only provides 10 GB/s compared to the 176 GB/s of the traditional "GPU-like" bus).

Oh, and PCIe 3.0 tops out at 15.75 GB/s bidirectional, which is actually more than the 10 GB/s bidirectional on the cache-snooping "Onion" bus in PS4. Of course, there should be a pronounced latency difference between the two.

Click to expand...

bishoptl said:

Are you kidding? I mean really - are you kidding?

This is part and parcel of the territory here. You have to answer for your statements, especially if you're here in an official capacity. People get banned for being out of line, but poking holes in the arguments of other posters is well within the rules.

I've had my work here both praised and eviscerated, called out by numerous forum folks both publicly and via PM when I got stuff wrong, and I'm a goddamn admin. Guess what - I wouldn't have it any other way. That is what makes NeoGAF what it is.

There are many, many people who are more than capable of assessing, vetting and debunking technical claims and they have every right to do so. That's the price of doing business here. If we had official Nintendo or Sony reps on board, they would be subject to the same process.

If you're scared, buy a dog.

Click to expand...

NullPointer said:

I think its a matter of giving a poster the benefit of the doubt unless you know otherwise.

In this case Albert has stated two things as facts, one regarding adding bandwidth and other regarding a read/write cycle. Seems very unlike standard PR spin and FUD to hold fast to such statements. And if he's wrong, surely somebody can point that out with similar certainty?

Click to expand...

astraycat said:

While this is great an all, what I (and I presume many others) want are numbers that correspond with the previous numbers given.

109GB/s is easily attained with the following formula:

(1024 bits/cycle) * (1 Byte/8 bits) * (853MHz) = 109GB/s.

However, we have no such formula for the oft-reported 204GB/s number. Assuming simultaneous read writes as a theoretical max for each cycle, we get the following numbers:

(2048 bits/cycle) * (1 Byte/8 bits) * (853MHz) = 218GB/s

This is the main point of contention when talking about ESRAM bandwidth, if you could get it cleared up it would calm the forum wars quite a bit.

Click to expand...

Klocker said:

So we all helped design and engineer the MS silicon in the Xbox One?

Click to expand...

Albert Penello said:

At Microsoft, we have a position called a "Technical Fellow" These are engineers across disciplines at Microsoft that are basically at the highest stage of technical knowledge. There are very few across the company, so it's a rare and respected position.

We are lucky to have a small handful working on Xbox.

I've spent several hours over the last few weeks with the Technical Fellow working on our graphics engines. He was also one of the guys that worked most closely with the silicon team developing the actual architecture of our machine, and knows how and why it works better than anyone.

So while I appreciate the technical acumen of folks on this board - you should know that every single thing I posted, I reviewed with him for accuracy. I wanted to make sure I was stating things factually, and accurately.

So if you're saying you can't add bandwidth - you can. If you want to dispute that ESRAM has simultaneous read/write cycles - it does.

I know this forum demands accuracy, which is why I fact checked my points with a guy who helped design the machine.

This is the same guy, by the way, that jumps on a plane when developers want more detail and hands-on review of code and how to extract the maximum performance from our box. He has heard first-hand from developers exactly how our boxes compare, which has only proven our belief that they are nearly the same in real-world situations. If he wasn't coming back smiling, I certainly wouldn't be so bullish dismissing these claims.

I'm going to take his word (we just spoke this AM, so his data is about as fresh as possible) versus statements by developers speaking anonymously, and also potentially from several months ago before we had stable drivers and development environments.

Click to expand...

Albert Penello said:

I do want to be super clear: I'm not disparaging Sony. I'm not trying to diminish them, or their launch or what they have said. But I do need to draw comparisons since I am trying to explain that the way people are calculating the differences between the two machines isn't completely accurate. I think I've been upfront I have nothing but respect for those guys, but I'm not a fan of the mis-information about our performance.

Click to expand...

• 18 CU's vs. 12 CU's =/= 50% more performance. Multi-core processors have inherent inefficiency with more CU's, so it's simply incorrect to say 50% more GPU.

Click to expand...

• Adding to that, each of our CU's is running 6% faster. It's not simply a 6% clock speed increase overall.

Click to expand...

• We have more memory bandwidth. 176gb/sec is peak on paper for GDDR5. Our peak on paper is 272gb/sec. (68gb/sec DDR3 + 204gb/sec on ESRAM). ESRAM can do read/write cycles simultaneously so I see this number mis-quoted.

Click to expand...

• We have at least 10% more CPU. Not only a faster processor, but a better audio chip also offloading CPU cycles.

Click to expand...

• Speaking of GPGPU - we have 3X the coherent bandwidth for GPGPU at 30gb/sec which significantly improves our ability for the CPU to efficiently read data generated by the GPU.

Click to expand...

James Sawyer Ford said:

I remember when major nelson pulled the same "adding memory bandwidth" back during the 360 days to tout that system's superiority.

Can't believe it still goes on to this day.

I have no doubt the technical fellow knows his stuff, but that doesn't mean people with experience don't stretch the truth or obfuscate facts in order to make their designs look better. Appealing to authority isn't something I'm going to accept on blind loyalty.

Click to expand...

Toki767 said:

Well we sure as hell probably helped design all the 180s.

Click to expand...

StevieP said:

Durante's post earlier today:
http://www.neogaf.com/forum/showpost.php?p=80713669&postcount=677

Click to expand...

Mark Cerny said:

"First, we added another bus to the GPU that allows it to read directly from system memory or write directly to system memory, bypassing its own L1 and L2 caches. As a result, if the data that's being passed back and forth between CPU and GPU is small, you don't have issues with synchronization between them anymore. And by small, I just mean small in next-gen terms. We can pass almost 20 gigabytes a second down that bus. That's not very small in today’s terms -- it’s larger than the PCIe on most PCs!

Click to expand...

Albert Penello said:

At Microsoft, we have a position called a "Technical Fellow" These are engineers across disciplines at Microsoft that are basically at the highest stage of technical knowledge. There are very few across the company, so it's a rare and respected position.

We are lucky to have a small handful working on Xbox.

I've spent several hours over the last few weeks with the Technical Fellow working on our graphics engines. He was also one of the guys that worked most closely with the silicon team developing the actual architecture of our machine, and knows how and why it works better than anyone.

So while I appreciate the technical acumen of folks on this board - you should know that every single thing I posted, I reviewed with him for accuracy. I wanted to make sure I was stating things factually, and accurately.

So if you're saying you can't add bandwidth - you can. If you want to dispute that ESRAM has simultaneous read/write cycles - it does.

I know this forum demands accuracy, which is why I fact checked my points with a guy who helped design the machine.

This is the same guy, by the way, that jumps on a plane when developers want more detail and hands-on review of code and how to extract the maximum performance from our box. He has heard first-hand from developers exactly how our boxes compare, which has only proven our belief that they are nearly the same in real-world situations. If he wasn't coming back smiling, I certainly wouldn't be so bullish dismissing these claims.

I'm going to take his word (we just spoke this AM, so his data is about as fresh as possible) versus statements by developers speaking anonymously, and also potentially from several months ago before we had stable drivers and development environments.

Click to expand...

TheCloser said:

The math doesn't work and it never will because it's a lie.

Click to expand...

Glorified G said:

Me too. You would think being the designated tech guy and get consistently get called out for your bullshit would be embarrassing enough. The "We CREATED DirectX" was hilarious, but that post took the cake. He really believed he could just add the peak bandwidth and everyone would fall for it.

That is definitely his last post on gaf, guarantee it.

Click to expand...

Albert Penello said:

At Microsoft, we have a position called a "Technical Fellow" These are engineers across disciplines at Microsoft that are basically at the highest stage of technical knowledge. There are very few across the company, so it's a rare and respected position.

We are lucky to have a small handful working on Xbox.

I've spent several hours over the last few weeks with the Technical Fellow working on our graphics engines. He was also one of the guys that worked most closely with the silicon team developing the actual architecture of our machine, and knows how and why it works better than anyone.

So while I appreciate the technical acumen of folks on this board - you should know that every single thing I posted, I reviewed with him for accuracy. I wanted to make sure I was stating things factually, and accurately.

So if you're saying you can't add bandwidth - you can. If you want to dispute that ESRAM has simultaneous read/write cycles - it does.

I know this forum demands accuracy, which is why I fact checked my points with a guy who helped design the machine.

This is the same guy, by the way, that jumps on a plane when developers want more detail and hands-on review of code and how to extract the maximum performance from our box. He has heard first-hand from developers exactly how our boxes compare, which has only proven our belief that they are nearly the same in real-world situations. If he wasn't coming back smiling, I certainly wouldn't be so bullish dismissing these claims.

I'm going to take his word (we just spoke this AM, so his data is about as fresh as possible) versus statements by developers speaking anonymously, and also potentially from several months ago before we had stable drivers and development environments.

Click to expand...

chadskin said:

Xbox 360 > Xbox One confirmed.

Click to expand...

onanie said:

Some of your points are misleading, or otherwise need clarifying.


Graphics processing is inherently parallel, so 18 vs 12 is indeed 50% more, given the same clock rate.


Each CU being 6% faster still means only 6% speed increase overall compared to YOUR baseline, not Sony's. You can't have it both ways, Albert. Having 50% more CU is not quite 50% more GPU, but having a 6% clock speed increase is more significant than the number implies?

1.8TF > 1.3TF, to the degree that is universally understood about laws of conservation.


As some others may have pointed out, you don't just add the numbers together. At no point in time can the GPU see more than the maximum ESRAM bandwidth. Cold, hard fact.

This leads to my question. Can ESRAM sustain simultaneous read/write cycles ALL the time? If not, then how much of the time?

And please allow me to help out your PR department a little. 204gb/sec, according to your understanding of the number, actually implies the old clock rate of 800mhz. The new number should be 853 * 128 * 2 (simultaneous read/write per cycle) = 218gb/sec. They can thank me later.


Please do tell us more about Sony's audio chip, or do you actually not know like the rest of us?


You are combining read and write bandwidth again for your side, while using the one-way bandwidth for your competition. Not quite being honest there, are you?

Click to expand...

ElTorro said:

That is based on an apparently outdated slide. I can quote Mark Cerny himself.



http://www.gamasutra.com/view/feature/191007/inside_the_playstation_4_with_mark_.php?page=2

Click to expand...

Albert Penello said:

At Microsoft, we have a position called a "Technical Fellow" These are engineers across disciplines at Microsoft that are basically at the highest stage of technical knowledge. There are very few across the company, so it's a rare and respected position.

We are lucky to have a small handful working on Xbox.

I've spent several hours over the last few weeks with the Technical Fellow working on our graphics engines. He was also one of the guys that worked most closely with the silicon team developing the actual architecture of our machine, and knows how and why it works better than anyone.

So while I appreciate the technical acumen of folks on this board - you should know that every single thing I posted, I reviewed with him for accuracy. I wanted to make sure I was stating things factually, and accurately.

So if you're saying you can't add bandwidth - you can. If you want to dispute that ESRAM has simultaneous read/write cycles - it does.

I know this forum demands accuracy, which is why I fact checked my points with a guy who helped design the machine.

This is the same guy, by the way, that jumps on a plane when developers want more detail and hands-on review of code and how to extract the maximum performance from our box. He has heard first-hand from developers exactly how our boxes compare, which has only proven our belief that they are nearly the same in real-world situations. If he wasn't coming back smiling, I certainly wouldn't be so bullish dismissing these claims.

I'm going to take his word (we just spoke this AM, so his data is about as fresh as possible) versus statements by developers speaking anonymously, and also potentially from several months ago before we had stable drivers and development environments.

Click to expand...

Skeff said:

I compared it to the ease of accessibility of the 68gb/s for the DDR3 in the XB1, the maximum theoretical performance of both of those as stated are reasonable representations, I suppose I should have said.

The ESRAM read+write simultaneously is not always applicable.

Click to expand...

StevieP said:

Durante's post earlier today:
http://www.neogaf.com/forum/showpost.php?p=80713669&postcount=677

Click to expand...

onanie said:

Each CU being 6% faster still means only 6% speed increase overall compared to YOUR baseline, not Sony's. You can't have it both ways, Albert. Having 50% more CU is not quite 50% more GPU, but having a 6% clock speed increase is more significant than the number implies?

Click to expand...

ElTorro said:

That is based on an apparently outdated slide. I can quote Mark Cerny himself.



http://www.gamasutra.com/view/feature/191007/inside_the_playstation_4_with_mark_.php?page=2

Click to expand...

Jenner wouldn't go into details on the levels of bandwidth available for each bus owing to confidentiality agreements, but based on our information the GPU has full access to the 176GB/s bandwidth of the PS4's GDDR5 via Garlic, while the Onion gets by with a significantly lower amount, somewhere in the 20GB/s region (this ExtremeTech analysis of the PS4 APU is a good read). Whatever the precise figure is for the more constrained CPU area, Jenner would only confirm that it's "enough". Optimising the PS4 version of The Crew once the team did manage to get the code compiling required some serious work in deciding what data would be the best fit for each area of memory.

Click to expand...

onanie said:

As some others may have pointed out, you don't just add the numbers together. At no point in time can the GPU see more than the maximum ESRAM bandwidth. Cold, hard fact.

This leads to my question. Can ESRAM sustain simultaneous read/write cycles ALL the time? If not, then how much of the time?

And please allow me to help out your PR department a little. 204gb/sec, according to your understanding of the number, actually implies the old clock rate of 800mhz. The new number should be 853 * 128 * 2 (simultaneous read/write per cycle) = 218gb/sec. They can thank me later.

Click to expand...

astraycat said:

There's no reason to take such a hard line about it, IMO. It could be this number comes out of some very specific operation like when we use FMA and claim "2 FLOPs" out of it for comparison. It could just be that there's some magic ROP operation specifically to ESRAM that can do a read-modify-write at a very strange rate.

Click to expand...

Albert Penello said:

So if you're saying you can't add bandwidth - you can.

Click to expand...

Albert Penello said:

I'm going to take his word (we just spoke this AM, so his data is about as fresh as possible) versus statements by developers speaking anonymously, and also potentially from several months ago before we had stable drivers and development environments.

Click to expand...

ElTorro said:

Click to expand...

JaggedSac said:

Boom

Click to expand...

npm0925 said:

You are yourself citing an anonymous source in an effort to debunk claims made anonymously. What is the name of this fellow?

Click to expand...

JaggedSac said:

Boom

Click to expand...

astraycat said:

While this is great an all, what I (and I presume many others) want are numbers that correspond with the previous numbers given.

109GB/s is easily attained with the following formula:

(1024 bits/cycle) * (1 Byte/8 bits) * (853MHz) = 109GB/s.

However, we have no such formula for the oft-reported 204GB/s number. Assuming simultaneous read writes as a theoretical max for each cycle, we get the following numbers:

(2048 bits/cycle) * (1 Byte/8 bits) * (853MHz) = 218GB/s

This is the main point of contention when talking about ESRAM bandwidth, if you could get it cleared up it would calm the forum wars quite a bit.

Click to expand...

Ushae said:

Tech deep dives?

Click to expand...

DMPrince said:

Fellow. Technical Fellow.

Click to expand...

Albert Penello said:

I see my statements the other day caused more of a stir than I had intended. I saw threads locking down as fast as they pop up, so I apologize for the delayed response.

I was hoping my comments would lead the discussion to be more about the games (and the fact that games on both systems look great) as a sign of my point about performance, but unfortunately I saw more discussion of my credibility.

So I thought I would add more detail to what I said the other day, that perhaps people can debate those individual merits instead of making personal attacks. This should hopefully dismiss the notion I'm simply creating FUD or spin.

I do want to be super clear: I'm not disparaging Sony. I'm not trying to diminish them, or their launch or what they have said. But I do need to draw comparisons since I am trying to explain that the way people are calculating the differences between the two machines isn't completely accurate. I think I've been upfront I have nothing but respect for those guys, but I'm not a fan of the mis-information about our performance.

So, here are couple of points about some of the individual parts for people to consider:

• 18 CU's vs. 12 CU's =/= 50% more performance. Multi-core processors have inherent inefficiency with more CU's, so it's simply incorrect to say 50% more GPU.
• Adding to that, each of our CU's is running 6% faster. It's not simply a 6% clock speed increase overall.
• We have more memory bandwidth. 176gb/sec is peak on paper for GDDR5. Our peak on paper is 272gb/sec. (68gb/sec DDR3 + 204gb/sec on ESRAM). ESRAM can do read/write cycles simultaneously so I see this number mis-quoted.
• We have at least 10% more CPU. Not only a faster processor, but a better audio chip also offloading CPU cycles.
• We understand GPGPU and its importance very well. Microsoft invented Direct Compute, and have been using GPGPU in a shipping product since 2010 - it's called Kinect.
• Speaking of GPGPU - we have 3X the coherent bandwidth for GPGPU at 30gb/sec which significantly improves our ability for the CPU to efficiently read data generated by the GPU.

Hopefully with some of those more specific points people will understand where we have reduced bottlenecks in the system. I'm sure this will get debated endlessly but at least you can see I'm backing up my points.

I still I believe that we get little credit for the fact that, as a SW company, the people designing our system are some of the smartest graphics engineers around – they understand how to architect and balance a system for graphics performance. Each company has their strengths, and I feel that our strength is overlooked when evaluating both boxes.

Given this continued belief of a significant gap, we're working with our most senior graphics and silicon engineers to get into more depth on this topic. They will be more credible then I am, and can talk in detail about some of the benchmarking we've done and how we balanced our system.

Thanks again for letting my participate. Hope this gives people more background on my claims.

Click to expand...

• What is this inherent inefficiency you speak of? Can you elaborate? It is not something I've ever heard mentioned.
• Your second point contradicts your first. If 50% more CU performance is viable to inefficiencies, why would 6% extra performance not also be privy to the same thing?
• How did you arrive at the 204gb/s figure for the Esram, can you elaborate? Also you realise this is a very disinginuous claim. YES the bandwidth can be added together in that the DDR3 and Esram can function simultaneously, but this tells only a small part of the full story. The Esram still only accounts for a meagre 32mb of space. The DDR3 ram, which is the bulk of the memory (8GB) is still limited to only 68gb/s, whilst the PS4's GDDR5 ram has an entire 8GB with 176gb/s bandwidth. This is a misleading way to present the argument of bandwidth differences.
• How do you know you have 10% more cpu speed? You said you are unaware of the PS4's final specs, and rumours of a similar upclock have been floating around. It could also be argued that the XO has the more capable audio chip because the systems audio Kinect features are more demanding, something the PS4 does not have to cater to. Add to that, the PS4 does also have a (less capable) audio chip, along with a secondary custom chip (supposedly used for background processing). There's that to consider too.
• That's good that Microsoft understands GPGPU, but that does not take away from the inherent GPGPU customisations afforded to the PS4. The PS4 also has 6 additional compute units, which is a pretty hefty advantage in this field.
• This is factually wrong. With Onion plus Onion+ the PS4 also has 30gb/s bandwidth.

onanie said:

And please allow me to help out your PR department a little. 204gb/sec, according to your understanding of the number, actually implies the old clock rate of 800mhz. The new number should be 853 * 128 * 2 (simultaneous read/write per cycle) = 218gb/sec. They can thank me later.

Click to expand...

ElTorro said:

That is based on an apparently outdated slide. I can quote Mark Cerny himself.



http://www.gamasutra.com/view/feature/191007/inside_the_playstation_4_with_mark_.php?page=2

Click to expand...

xxracerxx said:

Oh yes, the party has arrived.

Sometime it would be better for MS if they just didn't say anything like Sony does.

Click to expand...

Toki767 said:

It amuses me so much that a software company like Microsoft can't think of a more technical title for their cream of the crop engineers than "Technical Fellow"

Click to expand...

Fellow


From Wikipedia, the free encyclopedia

In academia, a fellow is a member of a group of learned people who work together as peers in the pursuit of mutual knowledge or practice. The fellows may include visiting professors, postdoctoral researchers and doctoral researchers.

Click to expand...