AMD THREADRIPPER VS INTEL SKYLAKE X

[N19h7m4r3]

Also very interesting is the Alienware Threadripper machine! They sport up to 64GB at 2933Mhz.

That's actually darn impressive, considering how much Ryzen has struggled to get to that barrer and break it with Dual channel, 16GB max usually.



Quote
A side note: Dell's Area 51 Threadripper desktop will be available for pre-order on July 27th – it would appear that Dell has an OEM launch day exclusive at this time, so we might see some more details then. The Area 51 will ship after the 27th; that is just the pre-order date.[/quote]

 

[N19h7m4r3]

Ingesting, with 20% more threads the $800 1920X scores 12% higher than the $1,000 7900X.

In last nights stream Ryan Shrout said "because Intel have better single core performance Intel's CPU would still be faster even in Multithreaded"

These guys are so Intel besotted... clueless.

Also the 7900X has a 4.0Ghz ALL CORE boost, and yet still couldn't match the 1920X that only has single core 4.0Ghz boost.

 

[jigger]

£50~ per core for AMD. £85~ for Intel. Is that about right?

 

[N19h7m4r3]

£50~ per core for AMD. £85~ for Intel. Is that about right?

Well the 7900X is 10 core, and cost $1000, so $100 per core for Intel.

 

[GAC]

Also very interesting is the Alienware Threadripper machine! They sport up to 64GB at 2933Mhz.

That's actually darn impressive, considering how much Ryzen has struggled to get to that barrer and break it with Dual channel, 16GB max usually.

thought the majority could run 3200 now minus one or two who still havent sorted out new bios updates ?

 

[N19h7m4r3]

thought the majority could run 3200 now minus one or two who still havent sorted out new bios updates ?

I see people still have issues with B350. Even so 64GB @ 2933Mhz on Threadripper is extremely positive!

Fill up Ryzen's 4 slots with 32GB and I doubt many will get that frequency stable.

 

[jigger]

3000Mhz across 4 channels is a lot of bandwidth for even 16 cores running a pair of threads each. I can't see memory speed being much of a factor on TR4.

 

[N19h7m4r3]

3000Mhz across 4 channels is a lot of bandwidth for even 16 cores running a pair of threads each. I can't see memory speed being much of a factor on TR4.

It'll be interesting to see the latency tests on RAM with it. Not just for CCXs, but the jump between the two Eight core modules themselves; and now the memory frequency affects it.

Although if an OEM can already managed 2933Mhz, as advertised; we could see people pushing it 3200Mhz on selected dims hopefully. Possibly at 32GB since not all DIMM slots will be occupied then.

 

[Rroff]

Except from an optimisation point of view the Intel solution is worse/more limited.

With the Intel solution EVERY core talking to ANY core has high latency, and this latency isn't reduced by RAM speed (or not much anyway).

With the AMD solution talking within a 4-core CCX has VERY low latency (like Skylake-non-X level), and then only crossing CCX boundary has high latency (a bit above Skylake-X). And this latency is also reduced dramatically as RAM clockspeed increases.

So there's clearly more room for improvements and clever code solutions with AMDs approach. E.g. running many instances of a program, but limiting each instance to 4 cores each, this would then have about 1/3 the core-to-core latency of Intel's new chips.

Inter-core latency is less of an issue for the typical tasks a CPU with a lot of cores would be aimed at - where it can hurt more is things like gaming. IIRC from the benchmarks it seems the high end Skylake platform is optimised (possibly due to its original development focus) to certain areas at the expense of others when compared to previous gen enthusiast CPUs, etc.

Also need proper benchmarks to test VMs etc. as supposedly the high latency is offset in that kind of use by advantages in other areas of the mesh system.

Hate to defend Skylake as it really is a steaming pile of **** but some people are rushing to discredit it with flawed or incomplete data.

 

[humbug]

Inter-core latency is less of an issue for the typical tasks a CPU with a lot of cores would be aimed at - where it can hurt more is things like gaming. IIRC from the benchmarks it seems the high end Skylake platform is optimised (possibly due to its original development focus) to certain areas at the expense of others when compared to previous gen enthusiast CPUs, etc.

Also need proper benchmarks to test VMs etc. as supposedly the high latency is offset in that kind of use by advantages in other areas of the mesh system.

Hate to defend Skylake as it really is a steaming pile of **** but some people are rushing to discredit it with flawed or incomplete data.

The point is Ryzen with its Infinity Fabric only suffers from Inter CCX latency, not intercore latency, and again just as with Intel only in games, the difference is AMD can overcome that inter CCX latency with faster RAM, to the point where 3200Mhz ram is 20% faster than 2400Mhz, as such a 1600X is as near as makes no difference as fast as the much higher clocked 7700K in games.

Its simply that Intel are behaving with the sort of Hyperbole their fanboys do with this 'glued together' #### TPU pulled to pieces.

Infinity Fabric is very different to Intel's Mesh and IMO in every way better.

Seriously, very interesting timers when Intel start behaving like their fanboys.

 

[Rroff]

The point is Ryzen with its Infinity Fabric only suffers from Inter CCX latency, not intercore latency

Basically boils down to the same thing with when you are operating with higher core counts. That is beside the point I was making though - in the kind of areas these CPUs would ostensibly be aimed at it is much less significant than areas like gaming - which is why people need to take a considered approach before using CPUs like these for gaming purposes.

AMD has caught Intel off cycle so some of these things seem to be being rushed into play before they are ready but don't underestimate Intel's progress when it comes to things like IF like functionality.

 

[humbug]

Basically boils down to the same thing with when you are operating with higher core counts. That is beside the point I was making though - in the kind of areas these CPUs would ostensibly be aimed at it is much less significant than areas like gaming - which is why people need to take a considered approach before using CPUs like these for gaming purposes.

AMD has caught Intel off cycle so some of these things seem to be being rushed into play before they are ready but don't underestimate Intel's progress when it comes to things like IF like functionality.

What... Intel needs optimising?

Again: altho whose solution is better does play into this it is more about Intel's behaviour and hypocrisy with their childish attacks on AMD.

 

[GAC]

not sure how i missed the video on the previous page with pricing and that cinebench demo but damn. the 7900x and above are in some sort of comedy land if the performance transfers across other software for threadripper.

 

[N19h7m4r3]

not sure how i missed the video on the previous page with pricing and that cinebench demo but damn. the 7900x and above are in some sort of comedy land if the performance transfers across other software for threadripper.

As long as the software likes cores, and doesn't use the latest AVX 512, it'll transfer and do so well.

Reviews of Epyc shows AMD is bludgeoning Intel at the moment, even in other AVX loads.

Floating Point: NAMD
Developed by the Theoretical and Computational Biophysics Group at the University of Illinois Urbana-Champaign, NAMD is a set of parallel molecular dynamics codes for extreme parallelization on thousands of cores. NAMD is also part of SPEC CPU2006 FP. In contrast with previous FP benchmarks, the NAMD binary is compiled with Intel ICC and optimized for AVX.

First, we used the "NAMD_2.10_Linux-x86_64-multicore" binary. We used the most popular benchmark load, apoa1 (Apolipoprotein A1). The results are expressed in simulated nanoseconds per wall-clock day. We measure at 500 steps.

Again, the EPYC 7601 simply crushes the competition with 41% better performance than Intel's 28-core. Heavily vectorized code (like Linpack) might run much faster on Intel, but other FP code seems to run faster on AMD's newest FPU.

 

[Rroff]

That is what 56 v 64 cores though? not that it is an insignificant result as price/performance and how many space it takes to build equivalent performance systems with be a consideration.

For ~14% more cores they are getting ~40% more performance though.

 

[N19h7m4r3]

That is what 56 v 64 cores though? not that it is an insignificant result as price/performance and how many space it takes to build equivalent performance systems with be a consideration.

Missing the key part there, it's doing well using Intel's compiler in AVX, and Zen has 50% the AVX performance Intel usually.

AMD doing extremely well in a near worst case there for price to performance, and performance to watt.

 

[Rroff]

Missing the key part there, it's doing well using Intel's compiler in AVX, and Zen has 50% the AVX performance Intel usually.

AMD doing extremely well in a near worst case there for price to performance, and performance to watt.

I'm not missing that aspect but its not comparing apples to apples even though the implications are apparent.

 

[N19h7m4r3]

I'm not missing that aspect but its not comparing apples to apples even though the implications are apparent.

It's apples to apples in the server word. Top available Xeons vs top Epycs in 2 sockets.

If you're going to buy and you need the The most cores available for 2S system, those are your only choices. So comparing their performance and price is key.

 

[Rroff]

It's apples to apples in the server word. Top available Xeons vs top Epycs in 2 sockets.

If you're going to buy and you need the The most cores available for 2S system, those are your only choices. So comparing their performance and price is key.

Which I commented on in my post as often people will have a specific number of racks or workstations, etc. to work with, etc. which can only support so many cores max. That kind of data though needs to be presented with better indication of equivalency if it was the other way around people would be screaming blue murder.

 

[jigger]

Intel have backed themselves into a corner with no strategy to come forward. The question now is how much of the market will AMD take over the next 5-10 years and will Intels profits from years gone by offset the futre loses. AMD have already taken over 10% with just the desktop market in a few months. The mobile market alone must be worth at least twice that.

I can see some very upset Intel shareholders.