You might also mention that the 3900X's 32% better single thread performance vs. the 1800X would be a 15% boost clock increase. Therefore, the IPC plus memory bandwidth increase must be 17% (69% total over the previous arch). That's really good for an arch that was already boasting 52% performance over it's predecessor!
We do not yet know the all-core frequency of the 3800X. However it looks like TSMC's 7nm high performance node is oriented more toward power efficiency than performance. Perhaps its optimal voltage - frequency curve is winking more toward mobile processors than desktop processors? That extra +45W TDP for a mere +300 MHz clock of the 3800X was very surprising, though it might hint at the above. Is the optimal base clock per TDP of the 3700X/3800X at 3.3 to 3.4 GHz?
Well, that doesn't mean it's using 105W, just more than 65W. In addition, we are not seeing all the boost frequencies or durations for various combinations of active cores. It may be that the 3800X can boost more cores higher, for longer.
16 cores with 2 memory channels is equivalent to 32 cores with 4 memory channels and 64 cores with 8 memory channels. All three have 1 memory channel per 8 cores. AMD has already done 2 out of 3, so why not do all 3?
My guess would be that yields are still not that great and having the fastest part use only 6 cores gets rid of a lot more die with bad cores. No point in putting out the 16 core until Intel makes the next move.
not released yet because amd has no reason at all to do so.
intel can't even compete with the 12core model at the moment so amd will keep the 16core back leaving room to upgrade when they need to, along with a broader base of useable chips since there will probably be a lot more chiplets with 6 of the 8 cores good enough to clock to 4.6ghz for the 3900x than than there are full 8core chiplets that can do the same, especially this early in the 7nm process.
they even can use faulty chips with 6 great cores and issues on the remaining two in their most expensive (still cheap) consumer chip this way.
plus this way they do not completely cannibalize their threadripper lineupt immediately.
they win on all fronts keeping the 16core away from the market.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
307 Comments
View All Comments
ballsystemlord - Monday, May 27, 2019 - link
You might also mention that the 3900X's 32% better single thread performance vs. the 1800X would be a 15% boost clock increase. Therefore, the IPC plus memory bandwidth increase must be 17% (69% total over the previous arch).That's really good for an arch that was already boasting 52% performance over it's predecessor!
looncraz - Monday, May 27, 2019 - link
The answer is NOT memory bandwidth. It's simply frequency.9900k maintains a high relative all core boost of 4.7GHz. The 3800X very likely drops closer to 4.2Ghz or so... maybe even lower.
Zen has much better multi-threaded scaling at the same clocks than Intel.
Santoval - Monday, May 27, 2019 - link
We do not yet know the all-core frequency of the 3800X. However it looks like TSMC's 7nm high performance node is oriented more toward power efficiency than performance. Perhaps its optimal voltage - frequency curve is winking more toward mobile processors than desktop processors? That extra +45W TDP for a mere +300 MHz clock of the 3800X was very surprising, though it might hint at the above. Is the optimal base clock per TDP of the 3700X/3800X at 3.3 to 3.4 GHz?GreenReaper - Thursday, May 30, 2019 - link
Well, that doesn't mean it's using 105W, just more than 65W. In addition, we are not seeing all the boost frequencies or durations for various combinations of active cores. It may be that the 3800X can boost more cores higher, for longer.Santoval - Monday, May 27, 2019 - link
16 cores with 2 memory channels is equivalent to 32 cores with 4 memory channels and 64 cores with 8 memory channels. All three have 1 memory channel per 8 cores. AMD has already done 2 out of 3, so why not do all 3?peevee - Friday, May 31, 2019 - link
2 and 3 at much lesser frequencies (and thus throughput requirements). 1 has to maintain high frequency to stay above 12C Ryzen 9.Dragonrider - Monday, May 27, 2019 - link
My guess would be that yields are still not that great and having the fastest part use only 6 cores gets rid of a lot more die with bad cores. No point in putting out the 16 core until Intel makes the next move.dustwalker13 - Tuesday, May 28, 2019 - link
where is the 16 core?not released yet because amd has no reason at all to do so.
intel can't even compete with the 12core model at the moment so amd will keep the 16core back leaving room to upgrade when they need to, along with a broader base of useable chips since there will probably be a lot more chiplets with 6 of the 8 cores good enough to clock to 4.6ghz for the 3900x than than there are full 8core chiplets that can do the same, especially this early in the 7nm process.
they even can use faulty chips with 6 great cores and issues on the remaining two in their most expensive (still cheap) consumer chip this way.
plus this way they do not completely cannibalize their threadripper lineupt immediately.
they win on all fronts keeping the 16core away from the market.
smart move.
peevee - Tuesday, May 28, 2019 - link
There is definitely a space left for 3990X, as well as a few Ryzen 3s and/or 3 Gs.svan1971 - Saturday, June 1, 2019 - link
You seem upset, Intel might be your best bet.