- Mar 3, 2017
- 1,747
- 6,598
- 136
That's nonsense, Genoa was a massive improvement over Milan.
Sebastian sits in on some interview of an engineer of questionable competence and the internet has to defend AMD. What the heck, Sebastian???That's nonsense, Genoa was a massive improvement over Milan.
Zen 5 | Zen 4 | Zen 3 | Zen 2 | |
Transistor Count | 8.315B | 6.5B | 4.15B | 3.9B |
Caches. L2 got twice as big. I think uop cache got larger, not sure about BTBs and so on. And don't forget about register files [not only more entries but for SIMD they doubled in size] and tweaks to make it clock higherThing I just noticed re-reading the Zen 5 Info thread:
6.5 / 4.15 = 56% increase
Zen 5 Zen 4 Zen 3 Zen 2 Transistor Count 8.315B 6.5B 4.15B 3.9B
8.315 / 6.5 = 27.9% increase
How did Zen 4, which is supposedly more or less a tweak on Zen 3, have this much more transistor count? Obviously the node leap was massive, but...that's still huge. 56% more.
Zen 5 is a massive rework with a completely new decode, and they still only added 28% in comparison.
L2 doubled, AVX512 and actually the higher clocks need a big Transistor budget. The Core itself also has enough changes, of course not the big obvious ones that ZEN5 has but the 13% IPC of ZEN4 isn't that much below ZEN5s 16%.How did Zen 4, which is supposedly more or less a tweak on Zen 3, have this much more transistor count? Obviously the node leap was massive, but...that's still huge. 56% more.
Cache, double pumped AVX-512 & frequency increase. AMD has stated in the past, I think it was for RDNA 2, that designing for higher frequency requires higher transistor count, as counterintuitive as it seems.Thing I just noticed re-reading the Zen 5 Info thread:
6.5 / 4.15 = 56% increase
Zen 5 Zen 4 Zen 3 Zen 2 Transistor Count 8.315B 6.5B 4.15B 3.9B
8.315 / 6.5 = 27.9% increase
How did Zen 4, which is supposedly more or less a tweak on Zen 3, have this much more transistor count? Obviously the node leap was massive, but...that's still huge. 56% more.
Zen 5 is a massive rework with a completely new decode, and they still only added 28% in comparison.
Typical AMD APU launch.
View attachment 106474
Z3 to Z4, got a onboard GPU for the first time since Z1 and avx-512. Z4 to Z5, avx512 doubled in capacity and others have mentioned several things. THose were the most noticeable things though.Thing I just noticed re-reading the Zen 5 Info thread:
6.5 / 4.15 = 56% increase
Zen 5 Zen 4 Zen 3 Zen 2 Transistor Count 8.315B 6.5B 4.15B 3.9B
8.315 / 6.5 = 27.9% increase
How did Zen 4, which is supposedly more or less a tweak on Zen 3, have this much more transistor count? Obviously the node leap was massive, but...that's still huge. 56% more.
Zen 5 is a massive rework with a completely new decode, and they still only added 28% in comparison.
Anybody giving credence to to someone commenting on server performance of Genoa over Milan who does not think it was an immense improvement is frankly a moron. in addition to the mhz which is like one ghz more at full load of all cores, plus the addition of avx-512, which IS a real server thing, proves they don't know what they are talking about. Genoa to Turin, I will know soon, as I have a Turin in my hand. But doubling the avx-512 alone is a big change. I don't expect that many other noticeable changes, but we will see.
How do you have a Turin in hand? ES? What is stopping you from running it now?Anybody giving credence to to someone commenting on server performance of Genoa over Milan who does not think it was an immense improvement is frankly a moron. in addition to the mhz which is like one ghz more at full load of all cores, plus the addition of avx-512, which IS a real server thing, proves they don't know what they are talking about. Genoa to Turin, I will know soon, as I have a Turin in my hand. But doubling the avx-512 alone is a big change. I don't expect that many other noticeable changes, but we will see.
These are the transistor counts for the CCD, so core and CCX changes only - iGPU, I/O, IMC, IOD changes are not reflected here.Z3 to Z4, got a onboard GPU for the first time since Z1 and avx-512. Z4 to Z5, avx512 doubled in capacity and others have mentioned several things. THose were the most noticeable things though.
Probably lack of bios support on his motherboard. The server OEMs aren't just spamming out prerelease bioses like the consumer boards do for upcoming new releases.How do you have a Turin in hand? ES? What is stopping you from running it now?
Gee, first there is no motherboard or vendor that has Turin CPU BIOS out there , so if I put my $2600 aquisition in, its a risk. Yes, its an ES. I have to decided to take the risk.How do you have a Turin in hand? ES? What is stopping you from running it now?
True. If a studio doesn't test on AMD APUs, not AMD's fault. They are just ignoring a potentially large percentage of sales, especially in second/third world countries.If 26 games out of 30 work flawlessly it means that the 4 remaining ones have half broken code, i wont abound on that matter but FI google "Resident Evil crash", and do the same with F1 24, it s just so much easy to blame AMD for the incompetence of some games developpers.
The idea of jumping with joy upon seeing the chip boot has probably got something to do with taking that riskI have to decided to take the risk.
I mean I'd heard of downloading more RAM before but downloading more IPC is a new thing for sure. I'll take it.Sweet, I guess we all got a nice performance bump without even purchasing any new AMD products. Thanks AMD!
It's.... a Conroe Jump?
We'll never see another Conroe jump as long as x86 is still produced on silicon.It's.... a Conroe Jump?
Hopefully a team of engineers reads your post and accepts the challengeWe'll never see another Conroe jump as long as x86 is still produced on silicon.