AMD Ryzen 5000 and Zen 3 on Nov 5th: +19% IPC, Claims Best Gaming CPU (anandtech.com) 57
Dr. Lisa Su, the CEO of AMD, has today announced the company's next generation mainstream Ryzen processor. From a report: The new family, known as the Ryzen 5000 series, includes four parts and supports up to sixteen cores. The key element of the new product is the core design, with AMD's latest Zen 3 microarchitecture, promising a 19% raw increase in performance-per-clock, well above recent generational improvements. The new processors are socket-compatible with existing 500-series motherboards, and will be available at retail from November 5th. AMD is putting a clear marker in the sand, calling one of its halo products as 'The World's Best Gaming CPU.' With the new Ryzen 5000 series, AMD is keeping a similar structure to the previous generation. The first four processors to market will include products in the key Ryzen 5 and Ryzen 7 segments, as well as a pair of high-performance parts with Ryzen 9. These will stretch from six cores to sixteen cores, with increased frequencies and increased performance-per-clock, but with no additional increase in power. The processors are still chiplet-based, with one chiplet having either six or eight cores. Ryzen 5 and Ryzen 7 will have one chiplet, while Ryzen 9 will have two chiplets -- the easy way to identify this is through the amount of L3 cache each processor has. Full details here.
Impressive (Score:5)
Whether you're an AMD fan or not, I think we should all be excited to see some competition back in the market. Intel had grown incredibly stagnant and I have to wonder if they'd still be selling us 4-core CPUs and charging an extra $30 to be able to overclock them.
Re: (Score:2)
Agreed, hopefully this competition will spur Intel into more advancements than AVX-512 and 10++++nm. It will be interesting to see where things go from here.
that's not a real architecture (Score:1)
Re: (Score:2)
Totally. I generally prefer AMD (and have for years) due to their being more system-builder friendly (longer-lived sockets). That said I couldn't bring myself to buy bulldozer. I went right from Athlon II to Ryzen. That was quite an improvement.
Re: (Score:2)
Re:Impressive (Score:5, Informative)
2020-[early]2021 is a particularly poor time to build a "future-resistant" PC. Intel has yet to ship complete systems with PCIe 4.0 support, and both Intel and AMD are expected to deliver DDR5 & USB 4.0 (Thunderbolt 3 compatible, if external I/O is a concern) platforms in late 2021 & early 2022 respectively. AMD is on its last full processor generation for its socket AM4 (perhaps we'll see a Zen 3+ in 2021, which would mean 3 full generations and 2 half-generations on the same socket), and Intel rarely supports a socket for more than 2 generations.
If going with Intel, late 2021 is the best "future-resistant" move, with early 2021 bringing only Rocket Lake-S processors (11th gen, finally moving away from the then almost 6-year-old Skylake architecture, but still on 14nm), with PCIe 4.0 but still no DDR5. Late 2021 will bring Alder Lake-S processors (12th gen & finally on 10nm SF) on a platform that will include PCIe 5.0, DDR5 & USB 4.0.
If going with AMD, early 2022 is the best "future-resistant" move, with Zen 4 processors on the [likely to be long-lived] AM5 platform bringing DDR5, USB 4.0 and likely PCIe 5.0.
Re: (Score:2)
Re: (Score:2)
The AM5 argument is valid, but otherwise, in a desktop form factor USB4 is a solution looking for a problem. It might make sense for expansion on a laptop, but the unified cable and greater bandwidth are mostly pointless on a desktop where plugging in a second+ NVMe, or GPU into the PCIe slot will provide the needed functionality with lower latency/etc. Not only that but USB upgrades in a year or two will probably be fairly expensive as the usual set of vendors release USB4 PCIe boards.
DDR4 vs 5 is another
Re: Impressive (Score:2)
You will learn that it is *always* a bad time to build a future-proof PC. There will *always* be new tech right around the corner.
Re: (Score:2)
Re: (Score:1)
Re:Impressive (Score:4, Interesting)
Whether you're an AMD fan or not, I think we should all be excited to see some competition back in the market. Intel had grown incredibly stagnant and I have to wonder if they'd still be selling us 4-core CPUs and charging an extra $30 to be able to overclock them.
How true. AMD has addressed the two major areas their CPU offerings under performed, power consumption and speed, with this new chip design. They have started to eek back more and more of their market share and put Intel back on their heels. However, don't count Intel out. Intel has incredibly deep pockets, are firmly entrenched in many more markets, and are arguably only marginally out performed by AMD's offerings. Intel has just announced new CPU's of their own as well, increasing clock rates and number of cores. Intel also has yet to benefit from smaller lithography processes, struggling to dial in their fabs processes, but I have a feeling they will succeed eventually and again AMD will be on the retreat without another technical jump.
Where the battle rages, and AMD has gained ground on Intel in some ways, it looks more like a stalemate than a clear win for either side to me. Yea, it's an improvement for AMD, who was on the ropes until the Ryzen stuff hit the market, but it's not as if AMD is landing very many punches, they just got themselves off the ropes by putting Intel on defense.
Re:Impressive (Score:5, Insightful)
The word is: "eke"
A company like Intel would normally have the next generation ready to roll as soon as a competitor looked like they were going to gain a foothold. The fact that AMD has been ahead suggests that they have no response.
There's also the fact that Intel is still struggling to make 10nm work while AMD is in mass production at 7nm.
All in all it seems to me like AMD is landing a lot of punches right now. Intel can rely on their brand name at the corporate level but AMD is owning the enthisiast market right now.
Re: (Score:2)
Two stories down from this one:
"Intel's 11th-gen Rocket Lake Desktop CPUs Will Debut in Q1 2021"
I thought that one was supposed to catch Intel up to AMD's current slate. And it's not due until well after Zen 3? Oy. They need to get their process fixed.
Re: (Score:2)
Don't get me wrong, AMD has returned from the abyss and narrowly avoided a standing count. It's a miracle that they have the where with all to make Intel step back, but in the furry of blows that AMD may be landing on Intel, isn't going to do much damage. AMD is in a lower weight class than Intel, their opponent is bigger and in better shape, they may land a few body blows on Intel, but there only hope to win the match is a knock out because even if they win this round, they are way behind in the match.
Re: Impressive (Score:2)
Intelâ(TM)s body is their fabs, and we see that 10nm is many years behind schedule (with yields still too poor for desktop chips) and 7nm already starting to see long delays. AMD isnâ(TM)t the real threat to Intel, TSMC is... and TSMCâ(TM)s pockets are probably deeper than Intelâ(TM)s when it comes to pure fab business.
Re: (Score:2)
Yep. Intel's insistence on building their own fabs is their Achilles Heel.
Re: (Score:3)
It is worth remembering that Intel is losing Apple. A business that AMD never had. I just built a new system for the first time in probably two decades and I never even considered Intel for the CPU. Let's just say I wouldn't buy Intel's stock right now.
Re: (Score:2)
the furry of blows that AMD may be landing on Intel, isn't going to do much damage.
IBM was once thought to be unassailable, for the same reasons. Where are they now?
Re:Impressive (Score:5, Informative)
From what I've seen so far, even intel's 14nm+++ is only about 20% less dense than TSMC 7nm in horisontal, and about the same in vertical (transistor width and transistor height). Der8auer of extreme overclocking fame had a very interesting series where he destructively disassembled CPUs under scanning electron microscope to get to see the actual transistors in them:
https://www.youtube.com/watch?... [youtube.com]
You can see intel's 14nm+++ vs TSMC 7nm transistor side by side comparison at the same magnification at 13:17 in the video.TSMC has about the same height, but packs transistors slightly more densely, fitting a bit less than 5 transistors in space where intel fits 4.
Intel's 10nm isn't really a competitor yet because of scaling problems intel has been having with it for a long time. It's why desktop CPUs from intel are still 14nm+++ and not 10nm.
Re: Impressive (Score:1)
Re: (Score:2)
If I had to guess, that's likely what intel thought about 10nm and their upcoming 7nm. But when you're working that close to where laws of quantum physics begin to govern physical interactions, a lot of things we don't really understand yet start to creep in.
Re: Impressive (Score:2)
It's chiplets. As long as Intel clings to not using chiplets, they can claim equal speeds per chip all they like... AMD can just drop some more chiplets into their interconnect and get, what, 1024 cores to make them cry.
Nobody sane writes non-massively-parallelizable code anymore unless algorithmically impossible. The single-threaded argument is weak. (Especially games have no trouble scaling up to one core per pixel, actor, collision, physics force, sound, input, network packet, etc. Ditto for everything t
Re: (Score:3)
Re: (Score:2)
It was a combination of Intel being unwilling to do semi-custom work, but also Intel not having any compelling graphics solution. Even if they'd been willing to do semi-custom work, they'd have to have done off-die graphics. Which admittedly does have some precedent after Intel did make that NUC processor with an Intel CPU and AMD GPU on the same package.
As always (Score:5, Interesting)
They artificially limit the 1-core turbo frequency of the Ryzen 5 to a bit lower than Ryzen 7 and 9, otherwise it would perform just as good in games.
But in theory the 6 core part should be able to perform better in 1-6 core tasks compared to more expensive 8, 12 and 16 cores parts.
Re: (Score:3)
Not necessarily. It depends on how the chiplets work. Are the 6-core chiplets truly 6 core? Or are they 8-core chiplets that didn't have all 8 pass validation? That is a common move in the semi conductor world given the ways in which chip fabrication can fail. I wouldn't be surprised if the 6 core chiplet is an 8 core that didn't pass grade. If that is the case then it makes sense why the clock speed is lower.
Re: (Score:2)
from what I understand the 12 and 16 cores have 2 chiplets. The 6 and 8 cores have a single one.
Still, even if there are 2 out of 8 deffective cores on the Ryzen 5, there is no reason why the Ryzen 7 should be able to reach an extra 100 MHz on a single threaded task. It should even be the opposite. It's easier to find a processor with 6 cores out of 8 that can reach 4800 MHz than a one with 8 out of 8 capable of reaching that frequency, on a single threaded load.
So not only the base clock, but also the turb
Re:As always (Score:5, Informative)
from what I understand the 12 and 16 cores have 2 chiplets. The 6 and 8 cores have a single one.
That is correct as far as I understand things.
Still, even if there are 2 out of 8 deffective cores on the Ryzen 5, there is no reason why the Ryzen 7 should be able to reach an extra 100 MHz on a single threaded task. It should even be the opposite. It's easier to find a processor with 6 cores out of 8 that can reach 4800 MHz than a one with 8 out of 8 capable of reaching that frequency, on a single threaded load.
Not necessarily. So the reasons behind why the two cores dont work can vary from chip to chip. The way these lowered binned parts work is that the chip maker will find a spec that a large number of failed chips will indeed meet while still generating enough profit. There are several factors that determine what that spec may be. Cost of the fab process, Yields per wafer, nature of common failures, production rates, what prices the market will support etc. This is why up until recently large gains could be made with overclocking and there was a serious silicon lottery.
The specs that chips were sold at were a lower bar that most could meet. Many could go much further. Modern validation and clock boosting actually makes it such that most chips are already boosting to their potential or very close to it without exotic cooling now.
But that binning works with failed chips too. If I have an 8 core die that failed validation it may still work as a 6 core, a 4, 2 maybe even single core. And the same goes for clock speeds too. Perhaps it can only hit half the intended clock speed? AMD will work closely with TSMC to determine what they can work with and if they can invent new SKUs that allow them to use a greater number of dies.
Now that doesn't mean the 6 core chiplet is binned like that. I think it makes sense given how AMD's chiplets have worked so far. I think its unlikely they specifically designed a 6 core chiplet.
Re: (Score:2)
Ryzen 5000 does not use chiplets anymore. Because of that Intel still won on games and high threaded but tied workloads as data between cores had to go to the infinity fabric and 2 different l3 caches.
This architecture shares one big cache and no cross communication between chiplets which is how they got the 19% IPC improvement.
So yes I think it is plausable that the chips that fail quality in the cache or some of the cores and converted to r5s and that would also include frequency boosts as well. Maybe if
Re: (Score:2)
The processors are still chiplet-based, with one chiplet having either six or eight cores.
https://www.anandtech.com/show... [anandtech.com]
Re: (Score:3)
They artificially limit the 1-core turbo frequency of the Ryzen 5 to a bit lower than Ryzen 7 and 9, otherwise it would perform just as good in games. But in theory the 6 core part should be able to perform better in 1-6 core tasks compared to more expensive 8, 12 and 16 cores parts.
I think the correct way to say this is a bit more nuanced, but you are on to something here.
Obviously AMD is keeping their chip die size small and just packing in multiple chips into their CPU's. What usually happens is they will benchmark the individual dies and bin them by performance. They put the lesser dies into the lesser CPU packages, and keep the best dies for use in the high profit margin CPU's. Generally, they know their statistical yields (how many low end parts they get verses the high perfo
Re: (Score:2)
They artificially limit the 1-core turbo frequency of the Ryzen 5
Yes they do. This is a process called binning, turning an otherwise scrap part from a Ryzen 7 into a sellable part at a discount. You're more than welcome to try and run them out of spec, but if its anything like the previous generation then the Ryzen 5 chips will have absolutely abysmal overclocks.
Re: (Score:2)
OK let say you bin for 4800 MHz turbo frequency, and scrap parts will be sold as 4700 MHz parts.
You also bin for 8 cores, and scrap parts will be sold as 6 cores.
Chances are more 6 core parts will be able to reach 4800 MHz than 8 core parts.
Marketing wins. That's why Intel often disable (or limit) turbo on Core i3 CPUs. Otherwise they would be as good as Core i7 in gaming.
Re: (Score:2)
Chances are more 6 core parts will be able to reach 4800 MHz than 8 core parts.
Based on what? You're assuming that the top end of the spec is the norm and everything else is a defect. The reality is the opposite. The bottom end of the spec is the norm and everything else is considered a benefit.
But in any case the overclocking numbers in the end don't lie. There are actual performance limitations, not just some software limits to give users the middle finger.
Re: (Score:2)
well statistically, you have more chances to find 6 good cores out of 8 capable of reaching 4800 MHz than 8 good cores out of 8, isn't it?
Re: (Score:2)
Proper Shared L3 Cache? (Score:3)
It reads as if the Zen3 architecture now runs more like Intel's 'Smart Cache' were all cores have equal access to the L3 cache, potentially allowing a single thread to use the full L3 entirely, is this right? I believe with Zen2, the 16MB L3 cache was actually 4x 4MB private caches for each core.
Re:Proper Shared L3 Cache? (Score:5, Informative)
Almost right. With Zen 2, each chiplet has two clusters of four cores. Those four cores can each access 16 MB of L3 cache that is shared by the four cores. With Zen 3, the chiplet has one cluster of eight cores, with 32 MB of L3 cache accessible to each core.
Obviously, the 12- and 16-core variants still split their total L3 cache, but not as much as in Zen 2. Zen 3 has a leg up on Intel's larger processors -- for example, the Core i9-10980XE is an 18-core processor with 24.75 MB of L3 cache. So each Intel core can access less L3 cache, and also has less L3 cache averaged across cores.
Re: (Score:2)
Those four cores can each access 16 MB of L3 cache that is shared by the four cores. With Zen 3, the chiplet has one cluster of eight cores, with 32 MB of L3 cache accessible to each core.
This is an interesting change. It makes me wonder by how much the latency has increased. The signals are traveling a longer distance so latency will increase with the 8-way shared cache. I suppose the L2 cache prevents this from being too much of an issue. The improvement gained from 32 MB of cache must outweigh any increase to the latency.
You're welcome. (Score:4, Funny)
I was kidding about this with a friend just yesterday. Oh well, I am happy with the new machine, and I doubt I would have waited until next month to build anyway.
Re: (Score:2)
Thanks. We were waiting for you to take one for the team :-)
Seriously though it was pretty much certain that Zen 3 was going to ship this year, even after Covid hit AMD said they still had plans to release this year.
Re: (Score:2)
Thanks. We were waiting for you to take one for the team :-)
Seriously though it was pretty much certain that Zen 3 was going to ship this year, even after Covid hit AMD said they still had plans to release this year.
I knew it was going to happen. After all newer/faster hardware always comes out. But one week, seems like someone is messing with me. :)
Re: (Score:1)
Re: (Score:2)
AMD publicly set the date of this announcement a month ago, and everything they were going to announce had already leaked in advance other than the exact pricing and availability, so it wasn't a surprise that new CPUs were coming. Besides, you're likely still within your return period for your new gaming PC, you can always just return it.
Re: (Score:2)
AMD had set a date for Zen3 weeks ago. It's good that you're not pissed, because if you were it could have been completely avoided by paying attention =D
Re: You're welcome. (Score:2)
Welcome to the 2000s/90s, kid!
That was the norm, back then. It never was the right time.
I remember buying CPUs for half a grand, that got slashed by half in price, only a week later.
Good work! (Score:2)
Nice to see AMD continuing to improve. I got a 3800X recently, so I probably won't be upgrading anytime soon, but that 5800X looks wonderful.
Great on desktop side, not yet server side (Score:2)
If I can, I hope to get a Ryzen 5 Pro 4650U laptop before the end of the year, (6 cores, 12 threads). None of this Intel garbage o
Re: (Score:2)
Thats where you may be a little out of date. AMD ARE making huge inroads on the server market with the Epyc chips. the big scale datacentres, the ones that buy in bulk, Amazon, etc, are going for the Epics in greater numbers than ever. And its not just for the price.
Re: (Score:2)
Guess I have spent too much time in small to medium corporate data centers.
On a side note, Intel's CPU line up has been confusing. In the past, for server CPUs, it was Xeon. Then the specific line depended on number of sockets, amount of cores and max memory supported. But, we now have various SoCs, multiple competing different lines and of course different generations. I am waiting