• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

Speculation: Ryzen 4000 series/Zen 3

Page 190 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

cortexa99

Member
Jul 2, 2018
193
259
106
Really? Show me a 3900XT at 5.1GHz+ that's not being cooled by liquid nitrogen or anything non-exotic
I'm not interested in any debate about older gen CPUs, but well if you all wanted to know......
one month ago in a chinese forum there's a crazy guy doing 5Ghz all core with 3900XT and with chilled water(ice-cooled):


041152cor7wlo7bz659rr9.png

it seems that he sucessfully benched cinebench with that. Reason I didn't post this before it's just because it is not a normal cooling and not be realistic to do daily productivity with this.....so I thought it's not even worth mentioning. But if you guys wanted to know how XT OC then here's an example, and is also an example to speculate how Zen3 OC headroom would be.
 

Zucker2k

Golden Member
Feb 15, 2006
1,528
867
136
I'm well aware that Intel can reach higher clocks than AMD, I've never argued that. My point (and it seems that Mark's as well) was that unlike previous gens, 10XXX series doesn't really have any overclocking room left and is quite similar to Latest Ryzen in that regard (which is good, no perf left on the table).
Mainstream overclocking is all about whatever you can achieve above base frequency using mainstream cooling solutions, ie. air and water. It is not whatever you can get on top of turbo boosting:

Base clock + 1400Mhz = 5,000MHz for the 9900k below is a 38% overclock, and some even go higher. (See screenshot below).

Usually, the pro-AMD members like you argue that MCE is not an Intel feature (and are turned off in reviews), and is in fact, overclocking. So, isn't this akin to eating your cake and wanting to have it too? A 125w 10900k runs around 3.9-4.1GHz in heavy, all core scenarios depending on code. They can be overclocked 1GHz+ on top of that by mainstream gamers and overclockers.

1601478984277.png
 

Thibsie

Senior member
Apr 25, 2017
223
208
116
I'm not interested in any debate about older gen CPUs, but well if you all wanted to know......
one month ago in a chinese forum there's a crazy guy doing 5Ghz all core with 3900XT and with chilled water(ice-cooled):


View attachment 30692

it seems that he sucessfully benched cinebench with that. Reason I didn't post this before it's just because it is not a normal cooling and not be realistic to do daily productivity with this.....so I thought it's not even worth mentioning. But if you guys wanted to know how XT OC then here's an example, and is also an example to speculate how Zen3 OC headroom would be.
It is worth mentioning. For what it is. But there's no reason not to mention it IMO, in its category 'esoteric cooling or something like that.
 
  • Haha
Reactions: Tlh97 and cortexa99

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
21,575
9,645
136
I'm well aware that Intel can reach higher clocks than AMD, I've never argued that. My point (and it seems that Mark's as well) was that unlike previous gens, 10XXX series doesn't really have any overclocking room left and is quite similar to Latest Ryzen in that regard (which is good, no perf left on the table).

I'm also quite certain AMD can't reach 5ghz all-core (though single core will be in the ballpark) but that stat alone is meaningless. Bulldozer did 5Ghz and was still crap.

Zen 2's gaming perfomance deficit is 90% memory interface related (latency) not clock speed. Its perfomance even in lightly threaded apps (other than games) is still often neck and neck with 5.2GHz Intel CPUs, Zen 3 should be all out faster.
Yes the statement was that the Intel overclocking was better than AMD's (in other words) and I was disagreeing with that. Also, yes, AMD may still not go over 5 ghz, and who cares ? Performance <> GHZ, and most of us have figured that out. The earlier AMD's could hot 5 ghz, but they were crap. And yes, latency is AMDs weakness in gaming IMO.

The % Intel can overclock vs AMD both on current generations is very similar.
 
Last edited:

Mopetar

Diamond Member
Jan 31, 2011
6,180
3,012
136
All of that is true, but there's also the human factor to all of this and a lot of consumers still make purchasing decisions around numbers on a product information sheet without understanding them.

The performance will be what it is, but AMD having a 5 GHz all core boost, even if only in the most premium product has some profound mindshare aspects that I think we're all aware of even if we argue against it.
 

CHADBOGA

Platinum Member
Mar 31, 2009
2,045
704
136
so if I tried to boot the system with a Zen 3 CPU with an older BIOS what would happen?...I wouldn't even be able to access the BIOS?
I'm not sure as I have never done this myself, but as long as you have the bios in the USB stick and press the update button on the back of the motherboard, everything should work out.

I suspect there is probably a YouTube video showing the process, out there.
 

jamescox

Senior member
Nov 11, 2009
283
506
136
I feel like there are a few different things going on here. The unified L3 effectively doubles the L3 size as seen by any single process. So, if the game working set has high locality, but spills a bit from a 16MB L3, unifying the L3 to an effective 32MB will be a big help there. The second thing is that, from what rumors we've heard, the IF links between the CCD and the IOD retain the same number of pins. Before, there were two separate pathways between the two chips, one for one CCX, and one for the other. With a unified L3 and a single 8 core CCX, that means that there is a dual ported link between the CCD and the IOD, or, the single link has about twice the bus width. We also hear that the IF speed has increased. What the sum total of all of this means is that, in a situation where there are few heavy threads, such as in most games, there is a massive increase in available bandwidth between the individual core and the memory controller.

Granted, that's all based on rumors that have been passed around, but it certainly speaks to what we're seeing here.

I'd be interested in seeing the memory setup for that 5800x...
I didn't think that it actually used two separate physical links; is this detailed anywhere? I was thinking that both cores just send their request to the IO die over the same physical link and do not do any communication with each other directly.

I don't know if we have any info on the Zen 3 IO die. I was expecting it to be very similar, if not the same, as the Zen2 IO die since it is still in AM4 and doesn't seem to need to change. If that is the case, then any latency improvements would all be in the cpu die. They need to have a higher clock speed or wider IO die for DDR5. I would expect wider doesn't come until Zen 4, so perhaps they have a new IO die with higher clocks and/or other tweaks.
 

jamescox

Senior member
Nov 11, 2009
283
506
136
Yes the statement was that the Intel overclocking was better than AMD's (in other words) and I was disagreeing with that. Also, yes, AMD may still not go over 5 ghz, and who cares ? Performance <> GHZ, and most of us have figured that out. The earlier AMD's could hot 5 ghz, but they were crap. And yes, latency is AMDs weakness in gaming IMO.

The % Intel can overclock vs AMD both on current generations is very similar.
Intel is already pushing the clock speeds to the limits to compete with Zen 2. I don‘t know if that is going to be sufficient to compete with Zen 3. The 32 MB monolithic cache is huge. Intel Xeon line used to top out at 38.5 MB for a $10,000 to $20,000 processor. Now, AMD has forced them to sell a 20MB 10 core processor as a mainstream part for less than $500. Intel is also pushing the clock speeds to the limits. You can buy a lower clocked part and try to push it to stock speeds of the higher stock clocked part, but there isn’t going to be much headroom without extreme cooling.

Edit: going off on a tangent here. I still remember when the Athlon 64 came out (I think it was the Athlon 64; maybe I don’t remember that well) and I was using a 750 MHz HPUX PA-RISC processor at work. For the compile benchmarks I was looking at, the 750 MHz part beat the Athlon at 1200 MHz. It sucks that itanium killed off a bunch of high-end RISC architectures. Anyway, people need to get used to the idea that clock speed is not necessarily relevant across architectures. There was a thread about the worst processor of all time, and in my opinion, that is the abomination that Intel released as the Pentium 4 with SDRAM. It clocked higher than the P3, but without rambus memory it was a huge downgrade. With rambus, it may have been roughly equal. One of my friends actually purchased one of those systems so he had a new system that was slower than the previous system at much higher clock. Rambus was expensive and Intel didn’t want to support DDR because that would have added support to AMD DDR systems. AMD will be in the lead for almost everything and they may be in the lead in gaming, even without hitting 5 GHz. Hyping up 5 GHz is just a possible cause for disappointment, even if AMD still performs better at a lower clock.

There may may be a few outliers, but a lot of games may have a much higher hit rate with the massive cache, improved cache structure, and probably better prefetch. Even if the raw memory latency isn’t improved, the average latency may have gone down significantly. If they have an improved IO die with higher clocks and possibly other tweaks, then the raw latency may be lower also. I believe the low resolution gaming benchmarks are essentially fancy memory latency tests that do not have much real value. Someone just mentioned 720p performance on high end hardware. The cpu in that case is just going through the scene geometry as fast as it can and throwing draw calls at the cpu. While games are probably relatively cacheable, the entire scene geometry isn’t going to fit in cache, so it is just going to cycle data in and out of the cache. AMD may have been hit with issues due to their cache structure (L3 is victim cache) in Zen 2 that Intel may not have hit with a completely different cache architecture. Zen 3 is supposed to have a completely redesigned cache system, so it may all be caching and prefetch with little improvement in raw latency.
 
Last edited:

jamescox

Senior member
Nov 11, 2009
283
506
136
Probably AMD will not have the kind of gaming perfornance some people suggest. Even if Zen3 has around +15% IPC & 6% more clock speed, it doesn't mean +22% more performance in games, at 720p testing, 10900K is around 16% faster than a 3900XT) If I wanted to speculate, I would expect tested with 3090 or Big Navi vs 10900K at 720p, $549 12core model +2%, $449 10core model +1%, $399 8core TOP model same perf, all the way down to $199 6core model same perf as $237 10600KF which is good (this way Intel for the next 4 months at least, is dead at +$200). lol, i see AMD trying an Nvidia with this and launching the first month only the higher tier SKUs and gradually the others (♥$$$)
Is this sarcasm? Are there actually people testing a 3090 with a 10900k at 720p? That is something like 1/36 of the 8k resolution that you are supposed be buying the 3090 for. What kind of frame rates are we talking there? You should not see 720p except in some cheap 13 inch laptop or something.

I am of the opinion that such low resolution test are testing raw memory latency, and that is about it. You might as well run some synthetic memory testers. Although, there is a possibility that it is partially the different cache architectures between AMD and Intel. Running through the scene data may cause the cache to cycle more in Zen 2. I think there was something about a single core only being able to access 8 MB of the 16 MB before the latency increased; I don’t have time to look that up now. AMD will certainly have worked on this in Zen 3. Even if Zen 3‘s raw latency is still higher, which I expect it will be, Zen 3 average latency due to massive cache improvements may be lower. AMD would probably be able to achieve similar raw latency with a monolithic APU, but not an MCM.
 

Hitman928

Diamond Member
Apr 15, 2012
3,654
4,102
136
Edit: going off on a tangent here. I still remember when the Athlon 64 came out (I think it was the Athlon 64; maybe I don’t remember that well) and I was using a 750 MHz HPUX PA-RISC processor at work. For the compile benchmarks I was looking at, the 750 MHz part beat the Athlon at 1200 MHz.
If it was at 1.2 GHz, it would have been an Athlon "Thunderbird" core which was followed by the AthlonXP line which was then followed by Athlon64. Overclocking was so much more fun (and frustrating) back then :)
 

TitusTroy

Senior member
Dec 17, 2005
314
38
91
AMD Ryzen 9 5900X 12 Core & Ryzen 7 5800X 8 Core Zen 3 CPUs Could Potentially Launch As Early As October 20th

The potential launch date of AMD's next-generation Ryzen 9 5900X and Ryzen 7 5800X Vermeer Zen 3 CPUs may have been unveiled and the Ryzen 5000 CPU series could hit the market even before the introduction of AMD's RDNA 2 based Radeon RX 6000 series graphics cards...as for the launch date, both sources reported at least one day that matches and that's the 20th of October...

 

Zstream

Diamond Member
Oct 24, 2005
3,396
277
136
Might be a little off-topic, but what's the best ram that peeps decided on for 4000 speeds?
 

uzzi38

Golden Member
Oct 16, 2019
1,817
3,654
116
I don't think this had been posted yet, but everyone in the Twitter rumorverse is alleging that this is the R9 5900X 12C/24T Zen 3 part. All the usual tech blogs are running with it, for dem clicks.

Yeah, think it's the 4.9GHz boost sample.

AMD is out for blood.

EDIT: Yep, I knew I recognised the number. It is.

 

Saylick

Golden Member
Sep 10, 2012
1,059
969
136
Yeah, think it's the 4.9GHz boost sample.

AMD is out for blood.

EDIT: Yep, I knew I recognised the number. It is.

I wonder what the two-digit number after the last hyphen means... That Igor's Lab link has 100-000000065-04_46/36_Y and 100-000000061-06_49/37_Y when the processor in that CPU-Z shows 100-000000061-08. Just a newer sample?
 

french toast

Senior member
Feb 22, 2017
988
824
136
Probably AMD will not have the kind of gaming perfornance some people suggest. Even if Zen3 has around +15% IPC & 6% more clock speed, it doesn't mean +22% more performance in games, at 720p testing, 10900K is around 16% faster than a 3900XT) If I wanted to speculate, I would expect tested with 3090 or Big Navi vs 10900K at 720p, $549 12core model +2%, $449 10core model +1%, $399 8core TOP model same perf, all the way down to $199 6core model same perf as $237 10600KF which is good (this way Intel for the next 4 months at least, is dead at +$200). lol, i see AMD trying an Nvidia with this and launching the first month only the higher tier SKUs and gradually the others (♥$$$)
Nobody tests $700+ GPUs at 720p mate...i have not seen Techspot cracking out Quake 3 Arena for their RTX 3090 review.
Intel is around 5% ahead at 1440p and negligible at 4k due to the gpu bottleneck, there are some outliers which show a much bigger gap however.

Latency is the most important metric for improving Ryzen's gaming performance at this moment and Zen 3 should fix that.
 

cytg111

Lifer
Mar 17, 2008
16,858
6,627
136
Ok, thats confirmation of the previous leak, thats awesome! Next some gaming benches, if someone were to leak some of those that would be great ;).
 

Hans de Vries

Senior member
May 2, 2008
273
666
136
www.chip-architect.com
I don't think this had been posted yet, but everyone in the Twitter rumorverse is alleging that this is the R9 5900X 12C/24T Zen 3 part. All the usual tech blogs are running with it, for dem clicks.

The CPU-Z leak in perspective:

Screenshot_20201001-094237_Chrome.jpg
 

ASK THE COMMUNITY