Overclock.net banner

[WCCF] Intel i9 7960X slower than 7900X in geekbench

3K views 29 replies 19 participants last post by  xlink 
#1 ·
Quote:
The processor scores 5238 points in single core results and 33672 points in the multi core results. While these would be very impressive numbers for any other chip, these are much lower than expected for the 16 core part. In fact, if you take a look at the scores earned by the Core i9-7900X (which is a 10 Core / 20 Thread part) you will see similar performance in almost all categories except the Floating Point and Memory category. The reason for this is obvious - the chip appears to be clocked at 2.5 GHz, which is a very low clock even considering this is a 16 core part.
SOURCE

Maybe 2.5ghz is not final?
 
#3 ·
This may be the generation where AMD has it right and Intel has it wrong.

In terms of absolute threaded performance, a large, single-die chip will probably yield better results. However, in terms of heat dissipation, smaller, multi-die chips should do better.

Chips that are able to be cooled better should be able to hold onto max performance longer... thus, Threadripper MIGHT have a performance advantage in many-core setups. I guess we shall see.
 
#4 ·
Quote:
Originally Posted by Mad Pistol View Post

This may be the generation where AMD has it right and Intel has it wrong.

In terms of absolute threaded performance, a large, single-die chip will probably yield better results. However, in terms of heat dissipation, smaller, multi-die chips should do better.

Chips that are able to be cooled better should be able to hold onto max performance longer... thus, Threadripper MIGHT have a performance advantage in many-core setups. I guess we shall see.
We already know how Threadripper will fare in many-core setups.

http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade
 
#5 ·
Quote:
Originally Posted by one-shot View Post

We already know how Threadripper will fare in many-core setups.

http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade
Comparing server applications to consumper applications is a recipe for disaster, since both rely on very different aspects of the CPU in question. We've already proven that the 1800X smoked the 6900K and if the newer 8-core CPU coming out follows the same path as the 7800X that just hit, I'm pretty sure that the 1800X will destroy it.
 
#6 ·
Quote:
Originally Posted by Imglidinhere View Post

Comparing server applications to consumper applications is a recipe for disaster, since both rely on very different aspects of the CPU in question. We've already proven that the 1800X smoked the 6900K and if the newer 8-core CPU coming out follows the same path as the 7800X that just hit, I'm pretty sure that the 1800X will destroy it.
I think you mean 1950X in the last sentence.
 
#7 ·
As an Intel guy, I really want AMD will succeed with TR.

Intel's CPU pricing is getting crazy and they need an actual opponent. Hope AMD will deliver and break Intel's monopoly at least for this generation, so I could get 8+ physical cores with a decent but FAIR price, either from Intel or AMD.
 
#8 ·
This is just Amdhal's law in action. Geekbench uses real-ish workloads, not artificial, perfectly scalable ones. Threadripper and Epyc are going to suffer from the same problem, but they might suffer less because of Ryzen's relatively low TDP at those speeds.
 
#9 ·
Quote:
Originally Posted by one-shot View Post

Quote:
Originally Posted by Mad Pistol View Post

This may be the generation where AMD has it right and Intel has it wrong.

In terms of absolute threaded performance, a large, single-die chip will probably yield better results. However, in terms of heat dissipation, smaller, multi-die chips should do better.

Chips that are able to be cooled better should be able to hold onto max performance longer... thus, Threadripper MIGHT have a performance advantage in many-core setups. I guess we shall see.
We already know how Threadripper will fare in many-core setups.

http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade
This benchmark of Threadripper vs Skylake X is more accurate than using EPYC
 
#10 ·
The big factor in AMD's favor is the TR4 socket. Better power distribution and more IHS surface area to dissipate the ample heat these high core count CPUs give off. Intel never intended the 2066 socket to ever house more than a 10-core, so AMD has the clear advantage in potential until Intel decides to give us their best and starts using the LGA3647 socket for enthusiast processors as well as their server ones. Intel will remain at a distinct disadvantage until they do this. They'd be much better off to release the 14, 16 and 18 core i9s on a board with an LGA3647 socket. Take the C622, enable overclocking on it, and combine it with an LGA3647 socket. Produce the i9-7940X, i9-7960X and the i9-7980X in an LGA3647 socket with ECC enabled, L/RDIMM support enabled, vPro enabled, and TXT enabled.

Unfortunately, Intel is loathe to actually give us their best and will instead provide the absolute minimum that they think they can get away with...
rolleyes.gif
 
#12 ·
Quote:
Originally Posted by lutjens View Post

The big factor in AMD's favor is the TR4 socket. Better power distribution and more IHS surface area to dissipate the ample heat these high core count CPUs give off. Intel never intended the 2066 socket to ever house more than a 10-core, so AMD has the clear advantage in potential until Intel decides to give us their best and starts using the LGA3647 socket for enthusiast processors as well as their server ones. Intel will remain at a distinct disadvantage until they do this. They'd be much better off to release the 14, 16 and 18 core i9s on a board with an LGA3647 socket. Take the C622, enable overclocking on it, and combine it with an LGA3647 socket. Produce the i9-7940X, i9-7960X and the i9-7980X in an LGA3647 socket with ECC enabled, L/RDIMM support enabled, vPro enabled, and TXT enabled.

Unfortunately, Intel is loathe to actually give us their best and will instead provide the absolute minimum that they think they can get away with...
rolleyes.gif
Agreed.
 
#13 ·
Intel probably needs 10nm more than AMD needs 7nm. Definitely seems like 12 cores and beyond are out of comfort zone for Skylake-X. Next Intel HEDT platform might feature something like LGA 3647 sized socket lol.
 
#14 ·
I think the 16 core Threadripper is actually going to be faster than the 18 core i9. It's looking like Intel's clockspeeds fall off a cliff with anything over 10 cores. AMD's multi-chip design is incredibly efficient and spreads the heat out over a large enough surface to keep clock speeds up even with high core counts. The 1950X should easily manage 3.9-4GHz on all cores while consuming just 240-260W.
 
#15 ·
Quote:
Originally Posted by CJRhoades View Post

I think the 16 core Threadripper is actually going to be faster than the 18 core i9. It's looking like Intel's clockspeeds fall off a cliff with anything over 10 cores. AMD's multi-chip design is incredibly efficient and spreads the heat out over a large enough surface to keep clock speeds up even with high core counts. The 1950X should easily manage 3.9-4GHz on all cores while consuming just 240-260W.
True...and this is primarily due to the robust nature of the large TR4 socket vs the much less capable (for high core count CPUs) 2066 socket.
 
#16 ·
I think the cinebench single core speed for the threadripper was 177, which is exactly my score right now, so that's alright for gaming, but the 7900's being 190's.
 
#17 ·
Quote:
Originally Posted by Neo_Morpheus View Post

I think the cinebench single core speed for the threadripper was 177, which is exactly my score right now, so that's alright for gaming, but the 7900's being 190's.
Single core speed isn't the most important thing for people who are considering Threadripper. Threadripper will rip the 7900X a new one in multi-threaded apps and higher core count Intel CPUs are going to need to be so restricted in clock speed to function on the inadequate LGA2066 socket (not to mention the thermal consequences of Intel's pigeon poop TIM), that their potential will be severely limited.
 
#18 ·
Quote:
Originally Posted by Neo_Morpheus View Post

I think the cinebench single core speed for the threadripper was 177, which is exactly my score right now, so that's alright for gaming, but the 7900's being 190's.
What does Cinebench score have to do with gaming ?
 
#20 ·
I've been waiting nearly a decade for AMD to strike back and now the time has come and I couldn't be happier. Finally getting the competition we deserve, now AMD..Step up in your GPU department.
wink.gif
 
#21 ·
Quote:
Originally Posted by lutjens View Post

Unfortunately, Intel is loathe to actually give us their best and will instead provide the absolute minimum that they think they can get away with...
rolleyes.gif
And they'll still get away with it (for the most part) because of their mindshare.
 
#22 ·
Below is my score with a Haswell-E 2696 V3 18 core CPU. All core turbo of 3.4 and 10 core turbo of 3.9.

 
#23 ·
I'm doubtful that EPYC/Threadripper's thermal performance would be appreciably worse even if it were monolithic, or that Intel would see better thermals at similar performance levels even if it were a multi-chip package.

Most of the thermal density issues are right at the core level and spacing things out at more granular scales than this is going to improve little, other than making waterblock design a bit easier.
Quote:
Originally Posted by lutjens View Post

Unfortunately, Intel is loathe to actually give us their best and will instead provide the absolute minimum that they think they can get away with...
rolleyes.gif
Everyone is like that. AMD simply realizes that, as the underdog with their first broadly competitive architecture in a decade, they can't hold anything back if they want to regain and hold marketshare.
 
#24 ·
Quote:
Originally Posted by Blameless View Post

I'm doubtful that EPYC/Threadripper's thermal performance would be appreciably worse even if it were monolithic, or that Intel would see better thermals at similar performance levels even if it were a multi-chip package.

Most of the thermal density issues are right at the core level and spacing things out at more granular scales than this is going to improve little, other than making waterblock design a bit easier.
Everyone is like that. AMD simply realizes that, as the underdog with their first broadly competitive architecture in a decade, they can't hold anything back if they want to regain and hold marketshare.
It's not whether or not a chip is monolithic or not, it's a simple matter of IHS surface area and having more power and ground pins to feed all the cores. TR4 is substantially better in this regard than Intel's Socket R4.

Looking back at history, AMD has always been the first to offer enthusiasts more. Whether or not it's because they were always the smaller company (by far) is uncertain. AMD produced the first enthusiast chip, the FX, (Intel hastily responded with the first Extreme Edition made out of a Gallatin-cored Xeon MP), the first dual capable system when they introduced the dual FX "2x4" platform (Intel responded with Skulltrail) and now the first unlocked 16-core CPU (Intel's response is their 12-18 core i9 CPUs). Perhaps AMD would behave differently if they were the larger company, but Intel has rarely given enthusiasts their very best products, at least not willingly, and when they do, it's only when provoked by AMD.
 
#26 ·
Quote:
Originally Posted by Blameless View Post

The path of least resistance for thermal transfer doesn't involve the majority of the IHS unless there is something wrong with the cooler base/mount.
The IHS does exactly that...spreads the heat. The bigger the IHS, the better (as long as the cooler/water block contacts the whole thing).The larger substrate helps too.
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top