Intel i9 7900x CPUs

AMD’s Threadripper will face more powerful competition that we first thought. A 18-core processor called Core i9-7980XE will be Intel’s flagship model of Core-X series.
https://videocardz.com/69900/exclusive-intel-to-launch-18-core-core-i9-7980xe-cpu
Good news. AMDs Ryzen and Threadripper have finally freed us from the 4-core base i7 & 8/10-core HEDT i7 limit. I never thought we would see 18 cores in non-Xeon processor this year, or the next 3 years for that matter.

That 4.5 GHz turbo clock of Core i9-7900X is going to be problematic to AMD. It will beat every single existing CPU that has under 12 cores (both AMD and Intel). And will beat most existing 16-core Xeons as well (as the clock advantage is so big).
 
There's no indication yet that Intel are changing anything in their plans or pricing until their new line is released. However, actually naming the high end i9 is almost certainly a reaction to Ryzen.
 
I thought the i9 name was around before AMD announced their R9?
It was, shortly before the Threadripper official announcement. R9 is not a thing, never was. I'm mainly referring to the i# naming scheme Intel reserves for consumer parts which they've never bothered with but suddenly it's a thing.
 
That 4.5 GHz turbo clock of Core i9-7900X is going to be problematic to AMD.
That's seems to be Intel's response, improve peak GHz of a few select cores; The Turbo Boost 3.0 frequency can only be reached by one or two "good" cores on the die, the rest are stuck at Turbo Boost 2.0's 4.3GHz . That'll win them all single threaded benchmarks.

But in the end, the base frequency only improved to 3.3GHz, - a 10% improvement on 6950x when running all cores flat out. And this at a TDP of 140W instead of the 95W of the Ryzen eight core products.

I think AMD has several options wrt. positioning a product to compete with this: More TDP and/or more cores and/or lower price.

Cheers
 
That's seems to be Intel's response, improve peak GHz of a few select cores; The Turbo Boost 3.0 frequency can only be reached by one or two "good" cores on the die, the rest are stuck at Turbo Boost 2.0's 4.3GHz . That'll win them all single threaded benchmarks.

But in the end, the base frequency only improved to 3.3GHz, - a 10% improvement on 6950x when running all cores flat out. And this at a TDP of 140W instead of the 95W of the Ryzen eight core products.
Skylake-X also has 4x larger L2 caches, and that's going to improve IPC in most applications (including games). Could be close to 15% higher IPC in total (if we assume bigger L2 cache gives 4.54%). Add that to current Intel IPC lead -> 3.3 GHz 10-core i9 should definitely beat all 12 core Threadripper models in all benchmarks.

And with that 4.5 GHz turbo, it will beat all Intel and AMD CPUs (including i7 7700K) with <12 cores in all applications and games. i9-7900X seems to be a very nice no-compromise enthusiast CPU. Obviously the 16-core Threadripper will compile code slightly faster and CPU render slightly faster, but the 10-core i9 will beat it in all games and productivity applications (including video encoding, since AVX-512 will excel in that).
 
Very nice pricing on the Skylake-X for the consumer, competition is truly a wonderful thing to have again! Motherboards will be rather expensive though. Goes to show just how amazing the Ryzen architecture is, how far they have come back into a comparable product. I wonder how disruptive AMD are going to be with Threadripper 16c/32t part.
 
Afaik Turbo Boost 2.0 is only for 1 core and Turbo Boost 3.0 is only for the two best cores. At least that is how it worked up until now (TB 3.0 is new though). I'll be impressed if they manage to run the 10 core part at 4.5 on all cores without delidding, as we already know how hot Broadwell-E and Haswell-E can get, and these are soldered.
Very nice pricing on the Skylake-X for the consumer, competition is truly a wonderful thing to have again! Motherboards will be rather expensive though. Goes to show just how amazing the Ryzen architecture is, how far they have come back into a comparable product. I wonder how disruptive AMD are going to be with Threadripper 16c/32t part.

I think they'll have to price it around 1000-1200. Assuming it clocks similarly to the R7 series we could see all cores running at 3.6-3.8 without problems. It's around the 4.0GHz mark that you really need to push the voltage high.

Edit: Some info from der8auer
 
I think they'll have to price it around 1000-1200. Assuming it clocks similarly to the R7 series we could see all cores running at 3.6-3.8 without problems.[/MEDIA]
High turbo clocks are likely, but don't think AMD is able to maintain high base clocks, as Intel can't maintain high base clocks with high core counts either.

Intel's 14-core i9 is $1399 and is likely going to beat the 16-core Threadripper in all benchmarks (14% more cores isn't enough to beat Intel's IPC advantage). 1200$ is likely a bit too much to ask after this announcement. 1000$ would be ideal. Today the competition isn't anymore against an aging 1723$ 10-core i7.
 
Skylake-X also has 4x larger L2 caches, and that's going to improve IPC in most applications (including games). Could be close to 15% higher IPC in total (if we assume bigger L2 cache gives 4.54%). Add that to current Intel IPC lead -> 3.3 GHz 10-core i9 should definitely beat all 12 core Threadripper models in all benchmarks.

I'm not sure about that at all. I think the reason the L2 is increased is because Intel is moving away from inclusive L3 caches. The increase in L2, while giving more capacity, comes with a higher access latency. I don't think average apparent latency is going to improve much, if at all, from a core's perspective.

Cheers
 
Skylake-X also has 4x larger L2 caches, and that's going to improve IPC in most applications (including games). Could be close to 15% higher IPC in total (if we assume bigger L2 cache gives 4.54%). Add that to current Intel IPC lead -> 3.3 GHz 10-core i9 should definitely beat all 12 core Threadripper models in all benchmarks.

And with that 4.5 GHz turbo, it will beat all Intel and AMD CPUs (including i7 7700K) with <12 cores in all applications and games. i9-7900X seems to be a very nice no-compromise enthusiast CPU. Obviously the 16-core Threadripper will compile code slightly faster and CPU render slightly faster, but the 10-core i9 will beat it in all games and productivity applications (including video encoding, since AVX-512 will excel in that).

First LN2 results don't show a significant increase in IPC over broadwell-e, 8% increase in clock speed yields an increase of 10% in CB15 MT, or in other words a 2% increase in IPC (which is the same difference anandtech reported when comparing Broadwell to Skylake):

7900x: https://d1ebmxcfh8bf9c.cloudfront.net/u6094/image_id_1849892.png
6950x: https://d1ebmxcfh8bf9c.cloudfront.net/u6885/image_id_1791193.jpeg

Where single thread IPC is needed the most the 7700k (or the 7740k variant) will most probably take the lead again.

Edit: LN2 results for the 7740k: https://d1ebmxcfh8bf9c.cloudfront.net/u18441/image_id_1849875.jpe
 
Last edited:
Where single thread IPC is needed the most the 7700k (or the 7740k variant) will most probably take the lead again.
Core 7900X has 4.5 GHz single core turbo clock. 7700k single core turbo is 4.2 GHz. 7900X has 300 MHz clock advantage in single threaded applications. It also has larger L2 and L3 cache and higher TDP headroom. It will beat 7700K in single threaded applications. This is the first time a high core count Intel HEDT chip will beat their 4-core mainstream flagship. It is a big deal for enthusiasts who don't want to sacrifice application & gaming performance.

Anandtech at least seems to believe that 4x larger L2 cache has a big impact on IPC:
"double of cache size decreases the miss rate by the sqrt(2), and is liable for a 3-5% IPC uplift in a regular workflow. Thus here’s a conundrum for you: if the L2 has a factor 2 better hit rate, leading to an 8-13% IPC increase, it’s not the same performance as Skylake-S."

But it is all speculation now, since we don't yet know the associativity and latency of the new 4x larger L2 cache. It both stay the same, then we see a noticeable IPC gain in general purpose code. Also it's not yet known how much the new L3 cache decreases performance. Skylake-X L3 is a smaller non-inclusive cache after all. My guess is that the new cache system is generally faster, but there could be some applications where the performance is actually slower.
 
Last edited:
Anandtech at least seems to believe that 4x larger L2 cache has a big impact on IPC:
"double of cache size decreases the miss rate by the sqrt(2), and is liable for a 3-5% IPC uplift in a regular workflow. Thus here’s a conundrum for you: if the L2 has a factor 2 better hit rate, leading to an 8-13% IPC increase, it’s not the same performance as Skylake-S."

But it is all speculation now, since we don't yet know the associativity and latency of the new 4x larger L2 cache. It both stay the same, then we see a noticeable IPC gain in general purpose code. Also it's not yet known how much the new L3 cache decreases performance. Skylake-X L3 is a smaller non-inclusive cache after all. My guess is that the new cache system is generally faster, but there could be some applications where the performance is actually slower.

Except they reduced the amount of L3 so total cache per core doesn't go up that much so the Hit rate wont change much, so the question is how many apps get a big benefit form the cache hit being in L2 instead of L3?

lets do occams razor here, why would Intel reduce L3 cache a massive amount, add a massive amount of L2 and cache the cache policy? What has changed in the Core? I think the Answer is obvious. The reason for a Large L2 and reduced L3 while changing the cache policy is so reduce the pressure on the L3 created by 512bit vectors.

Is it just me or is the technical media getting worse at this kind of analysis? Or maybe they where just always bad and Intel used to spoon feed them the truth as to the now silence?


also on ryzen:

https://forums.anandtech.com/thread...and-discussion.2499879/page-252#post-38915785
GMI and xGMI are basically the same link, but with different type of configuration (width), speed and error tolerance.

GMI is the internal form (inter-die) and has bandwidth of 42.6GB/s per direction at 2666MHz MEMCLK.
xGMI is the external form (inter-node, i.e. EPYC only), has the same bandwidth at the same MEMCLK but different speed, width and error tolerance.

SP3r2 which is used by TR doesn't support 2P and therefore it has no xGMI. Meanwhile SP3 Epyc parts do support 2P and that means xGMI is present as well.
It appears that both SP3 and SP3r2 use the same socket, which means that a huge amount of the pins will be unused on SP3r2 (four memory channels, 64x PCIe links, xGMI, etc).

xGMI is half the width of GMI, but operates at twice the speed.
 
Core 7900X has 4.5 GHz single core turbo clock. 7700k single core turbo is 4.2 GHz. 7900X has 300 MHz clock advantage in single threaded applications. It also has larger L2 and L3 cache and higher TDP headroom. It will beat 7700K in single threaded applications. This is the first time a high core count Intel HEDT chip will beat their 4-core mainstream flagship. It is a big deal for enthusiasts who don't want to sacrifice application & gaming performance.

Anandtech at least seems to believe that 4x larger L2 cache has a big impact on IPC:
"double of cache size decreases the miss rate by the sqrt(2), and is liable for a 3-5% IPC uplift in a regular workflow. Thus here’s a conundrum for you: if the L2 has a factor 2 better hit rate, leading to an 8-13% IPC increase, it’s not the same performance as Skylake-S."

But it is all speculation now, since we don't yet know the associativity and latency of the new 4x larger L2 cache. It both stay the same, then we see a noticeable IPC gain in general purpose code. Also it's not yet known how much the new L3 cache decreases performance. Skylake-X L3 is a smaller non-inclusive cache after all. My guess is that the new cache system is generally faster, but there could be some applications where the performance is actually slower.

mmm acttually 7700K have a 4.5ghz turbo https://ark.intel.com/products/97129/Intel-Core-i7-7700K-Processor-8M-Cache-up-to-4_50-GHz ... But yes i expect to be faster (IPC will up a bit due to new cache alone )... This said, i dont think they will be a so big differrence. ( in fact, i will use the 7700K as baseline performance for now )

Same thing with the 7820X who will come at 4.5ghz ( 8cores skylakeX ) who will be right against the Ryzen 1800x....
 
Last edited:
Except they reduced the amount of L3 so total cache per core doesn't go up that much so the Hit rate wont change much, so the question is how many apps get a big benefit form the cache hit being in L2 instead of L3?

lets do occams razor here, why would Intel reduce L3 cache a massive amount, add a massive amount of L2 and cache the cache policy? What has changed in the Core? I think the Answer is obvious. The reason for a Large L2 and reduced L3 while changing the cache policy is so reduce the pressure on the L3 created by 512bit vectors.

Is it just me or is the technical media getting worse at this kind of analysis? Or maybe they where just always bad and Intel used to spoon feed them the truth as to the now silence?
I agree that total cache hit rate doesn't improve much, but Intel's L2 latency is significantly lower than their L3 latency. Skylake L2 = 12 cycles (awesome). L3 = 42 cycles (3.5x slower). My fear is that 4x larger L2 in Skylake-X will be higher latency, thus partially invalidating the gains that 4x larger L2 would give.

I would guess that L2 size was increased because Intel needs to reduce their shared L3 traffic. Skylake-X has 18 cores (36 threads) feeding from the same shared L3. The core counts keep increasing every year. 4x larger L2 cache on each core means that each core accesses L3 cache less often. Ryzen on the other hand doesn't have one big shared L3 cache. Each four core fluster has their own separate L3 cache. Intel had to eventually increase their L2 size to make it possible to keep their big shared L3 cache.
 
I agree that total cache hit rate doesn't improve much, but Intel's L2 latency is significantly lower than their L3 latency. Skylake L2 = 12 cycles (awesome). L3 = 42 cycles (3.5x slower). My fear is that 4x larger L2 in Skylake-X will be higher latency, thus partially invalidating the gains that 4x larger L2 would give.

I would guess that L2 size was increased because Intel needs to reduce their shared L3 traffic. Skylake-X has 18 cores (36 threads) feeding from the same shared L3. The core counts keep increasing every year. 4x larger L2 cache on each core means that each core accesses L3 cache less often. Ryzen on the other hand doesn't have one big shared L3 cache. Each four core fluster has their own separate L3 cache. Intel had to eventually increase their L2 size to make it possible to keep their big shared L3 cache.
I actually expected Skylake-X to be based on the mesh or rings used in knights-whatever-they-are-up-to. I was actually disappoint to see the same old ring bus still kicking around. Maybe we will see in on the XCC die?
 
Back
Top