In an interview with tech outlet, HotHardware, AMD's Gaming Chief, Frank Azor, talked about the next-gen Ryzen & Radeon products coming to the market in 2022.
AMD's Gaming Chief, Frank Azor, Talks AM5 'Ryzen 7000' Longevity, Why Ryzen 7 5800X3D Is The Only V-Cache Option & How Radeon RX 6500 XT Will Tackle Miners To Become The Most Accessible Gaming GPU
Update: Another interview featuring both Robert Hallock, Director of Technical Marketing at AMD, and Frank Azor has been published by PCWorld which further reveals details for the upcoming products. While we will provide an excerpt of the interview in a bit, the main highlights were the Ryzen 7 5800X3D associated cache design, the switch to LGA socket for AM5, how AM4 coolers will be compatible with the AM5 socket, Ryzen 7000 series maintaining the same dimensions as existing Ryzen CPUs and how Ryzen 6000 improve power efficiency for notebooks.
During the interview, Frank shared some insightful information regarding the next-generation and upcoming products from AMD. We all had several questions after AMD's CES 2022 keynote so here are a few of those answered during the interview:
AMD 3D V-Cache, That Benefits, The Tradeoffs & How It's Helping Make The Best Gaming CPU Ever!
They want to game, They want the best CPU for gaming so 8 cores, tons of cache, it's a match made in heaven!
Modern PC gamers are making a ton of requests to memory, they are memory latency-sensitive than any other category of sensitivity, there's of course frequency sensitivity and core count sensitivity as well but overwhelmingly fast memory access is what's making the performance difference for a lot of these modern pc games especially ones that are still developed on direct x 9 or developed to run on a wide range of hardware like esports titles so when you think about those fast memory accesses and they happen so randomly because a game cannot predict what the user is doing if they want a no scope, that can't be predicted, the textures the models the sounds the animations, they just have to be there or get fetched and so if you can make a local pool of memory substantially larger than any other CPU that's ever been developed, that memory access doesn't have to go to ran anymore it goes to the CPU instead and it's making a massive difference in the performance of pc games. We are seeing uplifts of 20 30 40 percent in titles versus the non-3D stacked cache version of essentially the same CPU. So yes, we're absolutely convinced we do have the fastest gaming chip out there.
Frequency is important, frequency is important but every processor, every game is always a series of tradeoffs, bottleneck mitigations and in our architecture, when you are in the range of four and a half to five GIgahertz, four or five is enough when you put a ton of memory on top you know you are not limited by frequency anymore, you are not giving anything up to target that frequency, its the performance limiters or the performance accelerators move up, move elsewhere in the architecture so we can dial back on the frequency a bit ease up on the thermals, make it easier to cool and drop in a big extra blob of cache on top which is more transistor density, more thermal density so that's a trade-off that was very easy.
What I can say and do know is that there are lots of competing offerings out there within our own stack and from other blue companies and we are very sensitive to the price performance of a part. That's something we model early and often based on a variety of different workloads so we are not, you know, we are not ignorant to what's going on for pricing out there in the market and when we make the final decision, whatever that ends up being we want to come in on the right side.
So many games now are moving to this very latency-sensitive design and a great way to hide latency or work around it is to simply make a memory pool larger. You have eliminated the time that the GPU has to go to a frame buffer or the CPU has to go to the motherboard ram and that can be a huge accelerator so that's the engineering reason for it and the marketing reason for it is that people have started to realize there are non-frequency and non-core count reasons that certain processors certain GPUs are faster than another so the user consciousness has widened, so we are making sure we are continuing to invest in that.
It's a couple of nanoseconds because and time you make an l3 cache larger, it does add slightly more seek time to find the first bit of memory. That's true of any memory type, it could be ram, it could be GPU memory cache so l3 cache latency does go up a few seconds but net you penality of that is vastly outweighed by the benefit of skipping ram altogether. It comes down to the architecture of the chip, so the reason why we want to do an L3 cache is that that's shared amongst all the cores.
You can never guarantee which core is going to let me take it from an application point of view. I am an executable, I've spawned up x number of threads to run my game, maybe physics, audio, animation, as a CPU designer cannot guarantee what core that workload will land on or that thread will land on so if I put on l3 cache, more of that cache is shared amongst all cores evenly within a die so now it doesn't really matter what core that particular thread lands on, I have that fast access all the time geometrically it also makes sense because of Zen architecture, our L3 cache is right in the middle, you've got four cores on the left, big block of l3 cache and then for cores on the right, so it's easy to put that large chunk of cache right in the middle stacked vertically so it makes sense from a geometric and design perspective as well.
Over the last two or three years, in particular, there's been a decline in the general market focus on frequency. I think people are realizing that for example Ryzen can come to the table at 4.6 or 4.7 and credibly beat a CPU that might be running at five or five two and that's a 500-600 MHz spread. You reconcile that the only answer is that maybe frequency doesn't matter.
We as vendors have used Cinebench for a long time simply because it's an alright indicator that runs in a minute or two of single-thread performance but I'll take Zen three in particular as a good example of why that can be deceptive. If we had based the IPC increases of zen 2 to zen 3 on Cinebench only, it would have been nine percent, but we know that the IPC increase across like 25 workloads was on average more like 19 or 20 percent and games were 26 to 30 and its because IPC instructions per cycle really vary depending on the workload and so it's important to take a more holistic view of what the core is doing at a fixed frequency than just a bench. It's not that Cinebench doesn't matter or isn't important but it can give a very limited perspective to what's going on.
AMD Radeon RX 6500 XT Graphics Card - Aiming To Be The Most Accessible Gamer GPU & Hinting at an 8 GB Variant?
Starting with the Radeon RX 6500 XT graphics card, AMD has made some interesting choices for its first 6nm GPU and those include the addition of just 4 GB memory. Frank explains that there are a number of reasons why the decision to go 4 GB was made for this particular SKU. First of all, it is an entry-level SKU with a price point of $199 US and GPU memory is very expensive right now, and we aren't even taking the cost of the new 6nm process in the equation.
It is said that with technologies such as FSR and RSR supplementing the performance of the card, going 4 GB was the right choice but it is also said that consumers shouldn't assume that 4 GB will be the only configuration of the 6500 XT that will ever exist in the market. This points towards a future 8 GB release though we have to wait and see how long that takes to launch.
AMD also confirms that the 4 GB move is to tackle the crypto miners and while software attempts have been unsuccessful at warding off miners, this hard-coded approach (4 GB VRAM) will make it (RX 6500 XT) an unattractive option for miners. The card will also be available with a huge stock at launch so AMD's plan is to make the 6500 XT the most accessible and the most affordable gaming graphics card of the last two years when it launches.
Our main goal was to build Most Accessible GPU with strong 1080p performance.
Our goal was to get to a disruptive price point at a time when GPUs are very expensive and every major GPU that has come out in the last year or two has been I think over 300 dollars right or at least 299 and above. [...] We set the goal of 199 to be able to do that and memory prices have sky-rocketed very honestly and it's extremely challenging if not impossible to get to 199 in today's market with an 8 GB graphics card.
And to be able to design a card that would perform very well with all these modern technologies and achieve this price point and for our partners to be able to build a lot of them and we want them to build a lot of them, we did the performance testing between 8 Gigs and 4 GB, we factor in things like FSR and RSR, and the truth is that with those technologies supplementing the performance of the card and the fact that we are not going after 4k or 1440, we are going after 1080p medium to high settings, we think going with 4 gigs was the right decision.
It's also very unattractive to have four gigabytes if you're a crypto miner so there's a significant benefit of going 4 GB, which we think is going to give the availability of our GPU and market as compared to every other option out in the market our there right now. We will see if that materializes but we are planning to come to market with a strong amount of supply, we think we have done everything right, we hit the right price point. We think people will be really pleased with the performance when they see the real performance of the card versus the perceived performance of four gigabytes when supplemented with FSR and RSR and hopefully will be able to find in the market. I am optimistic that people will appreciate the decisions that we made here and I am very confident in the decision we made but I learn something new in this market every day so we will see what happens.
I will also say that don't assume that four gigabytes will be the only graphics configurations of the 6500 XT that will ever exist.
I think up to now most if not all attempts to try and do things to ward off miners haven't succeeded that much, these software hacks and things have been very quick and easy to overcome. We are all trying to figure it out, this is an attempt, let's see if it works out. This is like a hard-coded thing, this is like a limitation, you really can't even, I don't even think you can reasonably mine Ether with 4 Gigabytes of memory. If it works out, if we are right, this could actually be the perfect gamer and gamer-only graphics card. It's very affordable, it's got all modern technologies, it's got a memory config that hopefully miners will find unappealing, and it's a very small form factor in its shape and very aggressively priced and we hope that there'll be a lot of them on shelves. Every graphics card has been hated in the last two years so I am hoping we nail this one and people will love us for it. We'll see.
AMD Ryzen 7 5800X3D - Only One V-Cache Option But In The Best Ryzen Chip For Gamers
Frank also talks about the newly revealed AMD Ryzen 7 5800X3D and why it is the only CPU option with V-Cache. It is said that the Ryzen 7 5800X is the best chip in the Ryzen 5000 stack that is aimed exclusively at gaming. As such, it made the obvious choice to leverage the gaming performance further with the incorporation of a next-gen technology such as 3D V-Cache. However, there had to be some trade-offs made as the CPU has a power budget of 105W. Due to power and thermal constraints, they had to give in the clock speeds which explains why the CPU is clocked lower but also, more importantly, the benefits of V-Cache outweighed the frequency, for gamers.
With that said, the CPU can still be overclocked to hit those faster clock speeds. As for additional SKUs, and whether we will see V-Cache on future generations, AMD wants to see how well the V-Cache chip is adopted and the initial response from the market and if consumers and of course, AMD itself, see the value in it, then it will be deployed further.
It's a technology that today is benefitting games and gamers so we really wanted to put in on the best gaming processor and that's the 5800.
There aren't many games that get much more uplift out of an 8 core processor and the 5800 is really like the ideal gaming processor of all of them. When you move up the stack, you are getting more cores and it's allowing you to do more with the processor, you can be encoding, decoding, you can be streaming, you can be running a lot of things in the background when you are gaming but there aren't many games that take advantage of 8 or even 4 cores at a given time.
By putting it on a package like the 5800, it allows to give you more performance per core of those eight cores because we are able to focus the technology on those eight cores, we have a TDP limit to fit within so that's able to give us overall better performance per every one of those cores and there's honestly a price factor too. V-Cache is an expensive technology and it's pretty awesome. It's a technology that today is benefitting games and gamers so we really wanted to put in on the best gaming processor and that's the 5800. When you see it come to market, it will be something that folks will be able to find you know accessible to them and they are going to be getting a technology that will benefit what that processor that that processor was exclusively designed for, most efficiently designed for, which is gaming. So those are the main reasons for it.
The other thing is that you're probably wondering why we aren't putting it on every single SKU out there like this cutting edge technology, you know its starting here, we want to see how well it is adopted and if it is adopted and people love it and they see the value in it, then we will probably deploy it further. But new stuff is hard to be very honest, it's really hard and this is really new and there's a reason so many people are excited about it and it's so disruptive to the industry.
You are putting more things on the chip, something's gotta give right, nothing's free and for nothing's free from not just dollars, I mean from a thermal from a power perspective and so on and we are within a 105W so the performance uplift that the V-Cache is providing significantly outweighs the frequency and again, something's gotta give, so that's the high-level explanation I can give.
AMD AM5 Platform To Follow AM4's Legacy, Not Just Designed With Ryzen 7000 'Zen 4' In Mind But Future Generations Too
Lastly, we have a bit more insight on the longevity plan of the AMD AM5 platform which launches alongside the AMD Ryzen 7000 'Zen 4' Desktop CPUs in the second half of 2022. It is said that AM4 has been a transformative platform for AMD and that it's truly amazing for what it accomplished in its almost 5+ year life cycle. From Ryzen 1000 to Ryzen 50003D, there's nothing like it and the cost of ownership versus the competition has been truly a game-changer in the consumer DIY segment.
While Frank isn't committing to anything, he's hopeful that AM5 would follow the same legacy as AM4, lasting four, five, or even more years in the DIY segment.
The other guys give you a socket for like a year, maybe two years, so the total cost of ownership of AM4 AM5, there's just nothing that compares to it.
When you look at what we've accomplished with AM4, it's unprecedented for any other company other than AM4. We have remained on the AM4 socket for four generations, maybe more on that, like five or six years, it's remarkable, there's no other x86 PC platform that has accomplished that.
If you try to jam something into a socket that you planned seven years ago right because we have to plan this stuff 2+ years before you ever see it, you would just end up being uncompetitive. So what AM5 gives us is the ability to put not just everything you're going to see in Zen 4 but it's going to give us the ability to put the future generations in there as well so why go to AM5?
Well look if you are sitting on a Ryzen 1000, 2000, or even a 3000 processor and you are thinking about maybe moving to a 5000 or even to a 3D V-Cache one, we are putting this out there so that you can understand that hey, something else is coming in a few quarters. That should be welcomed as hey guys thanks for giving us heads instead of not telling me anything I go in and I buy into something you know with an AM4 and then six months later, it's potentially outdated or obsolete.
So the fact that we shared this thing and this information are because we're really trying to do right as well as showing off our goods and that we are making progress on our roadmap, that's the reason for AM5. AM4 is a 7-year-old socket now if not older than that and for everything that we have planned for the next 7, 5 years, you can't deliver than in the AM4 socket. For that reason, we have to go AM5, it is going to be essential to enable all the technology/innovation that we have planned for you over the next five years.
That's just in the next couple of years, we are looking at things three four-plus years from now, I can't give you a comment on the longevity of the socket, please I'm not giving anybody a commitment, all I am telling you is that you get to a point where you hit a brick wall as to how far you can go and AM4 is amazing, it's amazing! how far we have been able to bring it.
We got a socket for four-five years, we will hopefully be able to accomplish something similar with AM5.The other guys give you a socket for like a year, maybe two years, so the total cost of ownership of AM4 AM5, there's just nothing that compares to it and we know transitions are always tough, we understand that but I hope people can maintain perspective appreciate what we are accomplishing
AMD Ryzen 7000 Zen 4 LGA Socket Considerations & AM4 To AM5 Cooler Compatibility
Choosing LGA over PGA feels contentious to the public if you look at the chatter online but it really comes down to pin density, how many pins do you need on the bottom of the CPU to pin out the features that are in the chip when you can do it with LGA you just do it with LGA because you took or you do it with PGA because the CPU manufacturer can take on the complexity of pinning things our and that complexity is not on the motherboards, you are not shifting complexity to all the different motherboard in the market but when you reach a certain limit, maybe you are going PCIe Gen 5, DDR5, you've got more features, well you need more pins now and PGA might not be the right choice anymore so we are switching to LGA, it's 1718 pins and its just density you know how much is in the chip and what do we need to put down on the board we needed more pins.
A lot has been said about the shape of the CPU lid and those cutouts are making room for capacitors, surface mounted capacitors, that would have been on the underside of the CPU in a PGA equivalent part but you need all the space you can get on the bottom of the chip so we moved things up top and made those cutouts and that allowed us to keep the same package size as AM4. It is not pin-compatible but those CPUs are the same length, width, height, and that allows us to maintain cooler compatibility with the socket. (About whether the LID is sealed or not), It is sealed and you're not going to fill it up (With Thermal paste).
Overall, Frank & Robert gave some really good information during the interviews, and kudos to both HotHardware and PCWorld for hosting these interviews. You can see the full interview below:
