I'd guess the 95W and 105W chips have more overclocking potential than the lower power versions. In the same way that we make fun of Intel for rating their chips based on the base speed, AMD bases power on all-core boost speeds. If boosting power allows for higher speeds beyond the official boost speeds, AMD may still be able to hit 5GHz with this generation(we can't know until testing is done).
They will need it as AMD just attacked them on practically every front and apparently has the support from board partners with tech features being ahead of Intel boards for the first time in over 15 years.. Their also releasing a ton of different motherboards which is a good indicator that companies like Asus/Msi are expecting these to sell extremely well.
2006 isn't the early 2000s. Before Core 2 the Pentium 4 and P3s couldn't beat an Athlon/64 at the same clock speed. The P4 had to be almost 1GHz faster to beat an Athlon 64. It wasn't until 1st Gen Core i that they were faster in Enterprise. While Core 2 was fast in a single socket, the shared FSB made the Opteron faster in 2+ sockets.
Adding to your point SC. I built a fair number of i7 920s and PhenomII systems. Yeah you took a hit but it wasn't really noticeable for most. They were behind in launch times but it wasn't really until the FX line came out that they really took a massive hit as their single threaded mediocrity was quite noticeable by then as with their A8s/10s etc. Great graphics sure but you could notice the tradeoff.
I would expect a 5 year old process like Intel’s 14 nm to be tweaked to get very high clock speed at this point. Getting higher clock is going to be difficult on smaller process sizes so much of the improvements will be from IPC style improvements and system architecture improvements.
What intel is missing is the density and the chiplet architecture. The 12-core Zen 2 part has 64 MB L3. Intel 14 nm parts top out at 38.5 MB L3 and most of those are Xeons that cost thousands of dollars. AMD will have 64 MB L3 on a $500 part and 32 MB L3 on a $200 part. Cache density scales very well with smaller process.
Intel did make a 56 core Xeon, but I don’t know who would ever want to buy it. It doesn’t have any of the power optimizations allowed by designing the chiplets to be in an MCM from the start and it is 14 nm still. The TDP is 400 Watts. It is just two standard 28 core parts placed on a single package. They are close to 700 square mm each, so they are not cheap. The yields wouldn’t be great even at 14 nm++. The yield of processes under 14 nm isn’t going to be too good so using tiny cpu chiplets is a big win for yields. The 64-core Epyc 2 might be around 1000 square mm of silicon total. The 56 core Xeon (2 die) would be close to 1400 square mm with much less cache; 77 MB L3 total. The 64 core Epyc 2 will have 256 MB L3. That is a massive difference in cache density.
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
Targon - Monday, May 27, 2019 - link
I'd guess the 95W and 105W chips have more overclocking potential than the lower power versions. In the same way that we make fun of Intel for rating their chips based on the base speed, AMD bases power on all-core boost speeds. If boosting power allows for higher speeds beyond the official boost speeds, AMD may still be able to hit 5GHz with this generation(we can't know until testing is done).Sychonut - Monday, May 27, 2019 - link
Looking forward to Intel's 14++++.III-V - Monday, May 27, 2019 - link
They don't need it. They have the best performing process in the industry, and have held that crown since the early 2000s at the latest.What they do need, desperately, is an updated architecture. They have security holes in need of patching, and their IPC is lacking.
just4U - Monday, May 27, 2019 - link
They will need it as AMD just attacked them on practically every front and apparently has the support from board partners with tech features being ahead of Intel boards for the first time in over 15 years.. Their also releasing a ton of different motherboards which is a good indicator that companies like Asus/Msi are expecting these to sell extremely well.schujj07 - Monday, May 27, 2019 - link
2006 isn't the early 2000s. Before Core 2 the Pentium 4 and P3s couldn't beat an Athlon/64 at the same clock speed. The P4 had to be almost 1GHz faster to beat an Athlon 64. It wasn't until 1st Gen Core i that they were faster in Enterprise. While Core 2 was fast in a single socket, the shared FSB made the Opteron faster in 2+ sockets.just4U - Monday, May 27, 2019 - link
Adding to your point SC. I built a fair number of i7 920s and PhenomII systems. Yeah you took a hit but it wasn't really noticeable for most. They were behind in launch times but it wasn't really until the FX line came out that they really took a massive hit as their single threaded mediocrity was quite noticeable by then as with their A8s/10s etc. Great graphics sure but you could notice the tradeoff.jamescox - Tuesday, May 28, 2019 - link
I would expect a 5 year old process like Intel’s 14 nm to be tweaked to get very high clock speed at this point. Getting higher clock is going to be difficult on smaller process sizes so much of the improvements will be from IPC style improvements and system architecture improvements.What intel is missing is the density and the chiplet architecture. The 12-core Zen 2 part has 64 MB L3. Intel 14 nm parts top out at 38.5 MB L3 and most of those are Xeons that cost thousands of dollars. AMD will have 64 MB L3 on a $500 part and 32 MB L3 on a $200 part. Cache density scales very well with smaller process.
Intel did make a 56 core Xeon, but I don’t know who would ever want to buy it. It doesn’t have any of the power optimizations allowed by designing the chiplets to be in an MCM from the start and it is 14 nm still. The TDP is 400 Watts. It is just two standard 28 core parts placed on a single package. They are close to 700 square mm each, so they are not cheap. The yields wouldn’t be great even at 14 nm++. The yield of processes under 14 nm isn’t going to be too good so using tiny cpu chiplets is a big win for yields. The 64-core Epyc 2 might be around 1000 square mm of silicon total. The 56 core Xeon (2 die) would be close to 1400 square mm with much less cache; 77 MB L3 total. The 64 core Epyc 2 will have 256 MB L3. That is a massive difference in cache density.
webdoctors - Monday, May 27, 2019 - link
Hopefully good sales for the 3600X, maybe some microcenter cpu/mobo combos.Machinus - Monday, May 27, 2019 - link
So...does Intel even sell CPUs anymore???sorten - Monday, May 27, 2019 - link
Significantly more than AMD, but that's likely to change.