-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
AMD Zen 2 (Ryzen 3000) - *** NO COMPETITOR HINTING ***
- Thread starter Jamin280672
- Start date
seeing as how anandtech got the preview and the ryzen2 chip scored less than the 9900k, i think lisa's script would have planned for that eventuality.
(dunno about the crashing bit though
)
It was probably so close run to run that one of two things could have happened
I think they wouldn't have had a live demo if their was a chance of crashing. Remember the Intel F1 "game" VNC video clip demo and the muppet with the wheel pretending to play? 
Associate
- Joined
- 2 Jan 2019
- Posts
- 617
My own calculations suggest 4.1-4.3GHz max for the Ryzen ES, based upon known 2700X scores and the fairly widely predicted 9-13% IPC increase for Zen 2.
1900 for 2700X at 4.3GHz.
2057 for ES at unknown GHz.
1900*1.09= 2061 which implies clock parity.
1900*1.13= 2147 which implies ES was running lower than 4.3GHz. <5% difference in score, implying maximum clock difference of 0.2GHz.
Crude calculations of course, and CB runs do have varying results.
If we look at it from a 9900K perspective, with 2700X supposedly having a 5% MT IPC lead:
2700X at 4.3GHz = 1900
2700X at 4.7GHz = 2077
2700X at 4.3GHz with 9900K IPC = 1806
2700X at 4.7GHz with 9900K IPC = 1973
9900K at 5.0GHz = 2138 as per easyrider's image, and at 2700X IPC it'd be 2245.
In order for the ES to score 2057 at 4.7GHz there would need to be IPC regression from the last gen. That is simply not plausible given the core changes that have been made.
Given that the ES did only equal the 9900K, it could not have been clocked higher than 4.5GHz.
So we have an absolute upper limit for how that ES was clocked, and I very much doubt that it was even near to it.
1900 for 2700X at 4.3GHz.
2057 for ES at unknown GHz.
1900*1.09= 2061 which implies clock parity.
1900*1.13= 2147 which implies ES was running lower than 4.3GHz. <5% difference in score, implying maximum clock difference of 0.2GHz.
Crude calculations of course, and CB runs do have varying results.
If we look at it from a 9900K perspective, with 2700X supposedly having a 5% MT IPC lead:
2700X at 4.3GHz = 1900
2700X at 4.7GHz = 2077
2700X at 4.3GHz with 9900K IPC = 1806
2700X at 4.7GHz with 9900K IPC = 1973
9900K at 5.0GHz = 2138 as per easyrider's image, and at 2700X IPC it'd be 2245.
In order for the ES to score 2057 at 4.7GHz there would need to be IPC regression from the last gen. That is simply not plausible given the core changes that have been made.
Given that the ES did only equal the 9900K, it could not have been clocked higher than 4.5GHz.
So we have an absolute upper limit for how that ES was clocked, and I very much doubt that it was even near to it.
It was very close, maybe even close enough to have gone either way. But i don't think that's what she was worried about, it was going to end at around the same performance as the 9900K and i think that's the point they wanted to make.
She was very insistent that its not at its final clocks so i think this is early finalised silicon still undergoing fine tuning, leaks of the original Ryzen had them as low as 2.8Ghz, i think she was worried about it crashing if she was worried about anything.
I think a few more months of tweaking this thing will have more to give, it will be a monster by todays standard but quickly made mediocre by the rest of them.
Fantastic that we are again living in such times, this defiantly feels like Athlon XP all over again.
Soldato
- Joined
- 30 Aug 2018
- Posts
- 2,572
Changing nothing in hardware or software and doing multiple runs on the same pc with more than adequate cooling, I've seen changes in score of at least 50 points up or down between them. There seems to be no obvious reason for it, even comparing first run vs first run etc.
It can be inconsistent so perhaps that was a worry. Perhaps not.
It was clearly set up so that there was just enough difference to make it pull ahead by a few points. I'd imagine if it scored too highly it would set expectations beyond what will come in the final silcon on the shelves, the opposite if they clocked it too low.
The main take away should be the PPW achieved, which is pretty easy to show the Intel scored 11.3 points per watt, and the AMD 16.8 points per watt, that gives AMD approximately a 49% performance advantage per watt of power used. That is HUGE!
I'm sure she was hoping the Ryzen chip would get a higher score. Due to how close it was and how varying the scores can be, it could've easily gone the other way. In reality the result can only be called a draw.
I also found it funny how she was avoiding mentioning Intel by name for the whole hour (as one often does with competitors) and then at the end said it anyway.
Woudn't you be nervous if you're about to run a live demo on an engineering sample CPU that has every claim to be an industry leader?
Besides, that demo was never intended to beat Intel. It was meticulously crafted to precisely match the 9900K in as like-for-like a comparison as possible, and then reveal that they did it with only half the package.
If they wanted to go big and crush Intel then they could've wheeled out a 16 core package with both chiplets populated, but that then shows your full hand. Intel know what they are facing, and all future Ryzen performance figures drop off from there as you go down the SKUs and the core counts.
Instead, AMD went "well this is 8 cores just like the 9900K and we're tied with test silicon and half the package. Just think what we can do when the silicon is finalised and drop in another chiplet".
Last edited:
As I replied to beany above, and others have also suggested, I think a draw was entirely the point.
Ryzen 1000 was built on a 3Ghz low power Samsung Mobile node, Ryzen 2000 on an improved version of that.
At the time AMD had no choice but to use the node they did, nothing wrong with it other than its designed for mobile parts so its low clocking, makes it very power efficient but if you want to get Desktop like clocks out of it it costs a lot of volts, that's why you need 1.4v on Ryzen 1000 for 3.9Ghz and the same for 4.3Ghz on Ryzen 2000.
Ryzen 3000 is built on a more 'normal' TSMC node, its more like Intel's node tho it being ++++++++++++++++++++++++++++++++++ is more mature, its why i just cannot see Ryzen 3000 clock to 4.7Ghz out of the box, so it must be IPC as well, it may overclock from what may be 4.5Ghz to 4.7Ghz all core but no higher, that will still make it score knocking on for 2300 points in Cinbench.
Hon on EARTH do you know that?
Caporegime
- Joined
- 18 Sep 2009
- Posts
- 30,310
- Location
- Dormanstown.
If we're assuming that AMD can just release a 16 core version, then Intel already know what they're facing. If anything they'd be overestimating how fast AMD could make a 16 core chip run.
AMD always has an issue with delivering a killer blow.
If the CPU produces higher IPC, it does not really matter if it matches clock for clock. That is the way the market is going. Better designs using less power and producing less heat and noise energies to dispose of.
Associate
- Joined
- 2 Jan 2019
- Posts
- 617
We're not talking about easyrider's 9900K anymore.
For a first crack at 7nm, I think that 4.5GHz for the majority of SKUs would be acceptable as an ACT clock. Especially so if it is primarily IPC gains that have brought parity with the 9900K thus far.
We know that 14nm++++++++ has no more legs in the IPC department, so short of more cores at 14nm++++++++ there is little that Intel can offer until they get a properly functioning 10nm process to desktop. Late 2020 at best? By then, AMD will be on 7nm+ with improved IPC and clocks of its own, against a 1st gen Intel 10nm. You might find that AMD holds an IPC and clock advantage by that point.
Intel aren't in a comfortable seat right now.
True, but you're still not going to wade in all guns blazing with your maximum potential at a keynote that's intended purely to introduce the new CPUs.
If those Anandtech numbers are correct then I don't foresee AMD having any issues getting the 2nd chiplet clocking just as fast as we saw, especially with highly-binned dies.
That is such an open statement, you can back that up later any way you wish. For example, even if chips come with 5ghz, you can still later claim to be "right", because lower parts don't come with 4.5.