News Jim Keller Shares Zen 5 Performance Projections

[Admin]

Tenstorrent expects its 8-way Ascalon RISC-V processor to offer industry-leading performance-per-watt.

Jim Keller Shares Zen 5 Performance Projections : Read more

 

[PlaneInTheSky]

"integer performance" is relevant to literally nothing

 

[sitehostplus]

Um, integer performance is how everything in a cpu actually works.

 

[hotaru.hino]

"integer performance" is relevant to literally nothing

It's the most common data type that's operated upon in CPUs so... I think that's a huge deal.

 

[JamesJones44]

It's the most common data type that's operated upon in CPUs so... I think that's a huge deal.

Both forms are used pretty extensively these days. A lot of vectorized workflows, graphics and AI make extensive use of floating point based types. Which gets more use is strongly dependent on the use of the machine. Someone who does a lot of ML wouldn't benefit from great integer performance with subpar floating point performance for example.

 

[hotaru.hino]

Most things that do vectorization workflows use floating point, along with AI, graphics, etc. When I look at what is down the road in the future of computing, the biggest areas are floating point based.

Anything involving large scale compute will either be shoved onto GPUs or specialized processors anyway. I can only think of one supercomputer in the past decade that used only CPUs, and even then those CPUs were designed pretty much how a GPU would be designed. Everything else relied on GPUs to get their compute power (the current TOP500 computer has 4 GPUs per CPU, for instance)

A CPU is not useful as a data crunching part. You can shove more stuff onto it to do data crunching, but overall a CPU is useful for instruction crunching.

 

[ekio]

"integer performance" is relevant to literally nothing

This comment is the most uneducated/idiotic of the day.

 

[ikjadoon]

Most workloads are integer-based. Geekbench is one of the few benchmarks that explicitly tests both and integer is given a much larger weighting.

 

[Nikolay Mihaylov]

Most workloads are integer-based.

True. But also, FP workloads are usually parallelizable either via vector processing or with more cores, maybe even on a GPU. And such tasks usually take longer and one expects them to take longer, rather than be done in a fraction of a second after the mouse click.

So this really is a big deal because it will increase the preceived responsiveness of the machine.

 

[bit_user]

"integer performance" is relevant to literally nothing

So, like compiling code, web browsing, data compression, network stack, and OS performance are "literally nothing"? Not to mention most server apps involve mostly scalar, integer code: databases, web servers, video servers, etc.

The SPECint 2017 suite, cited in the slide, is comprised of the following:

​

SPECrate®2017 Integer	SPECspeed®2017 Integer	Language[1]	KLOC[2]	Application Area
500.perlbench_r	600.perlbench_s	C	362	Perl interpreter
502.gcc_r	602.gcc_s	C	1,304	GNU C compiler
505.mcf_r	605.mcf_s	C	3	Route planning
520.omnetpp_r	620.omnetpp_s	C++	134	Discrete Event simulation - computer network
523.xalancbmk_r	623.xalancbmk_s	C++	520	XML to HTML conversion via XSLT
525.x264_r	625.x264_s	C	96	Video compression
531.deepsjeng_r	631.deepsjeng_s	C++	10	Artificial Intelligence: alpha-beta tree search (Chess)
541.leela_r	641.leela_s	C++	21	Artificial Intelligence: Monte Carlo tree search (Go)
548.exchange2_r	648.exchange2_s	Fortran	1	Artificial Intelligence: recursive solution generator (Sudoku)
557.xz_r	657.xz_s	C	33	General data compression

Source: https://www.spec.org/cpu2017/Docs/index.html#intspeed​

​

Clearly, we see it also includes non-scalar workloads, such as x264 (a highly-optimized H.264 encoder).

 

[bit_user]

What I really came here to say: What the heck is Intel Xenon?

Could it be some "skunk works" ARM core that Intel designed?


My other thought about this article: Good Luck! I sure hope their CPUs see the light of day in less time than the much-hyped Nuvia cores!

 

[digitalgriffin]

"integer performance" is relevant to literally nothing

If you aren't doing mimd simd fma's integer performance is still highly relevant for things like databases and spreadsheets and os ops. This makes them relevant on cloud servers where that stuff is important.

 

[digitalgriffin]

Tenstorrent expects its 8-way Ascalon RISC-V processor to offer industry-leading performance-per-watt.

Jim Keller Shares Zen 5 Performance Projections : Read more

I think you meant to say amd zen 5 will be north of 5GHz, not 4GHz

 

[Jimbojan]

I am sure Intel has a much better design than AMD in Integer or floating point cpu design. Thus, Keller's comments have little value.

 

[jp7189]

I am sure Intel has a much better design than AMD in Integer or floating point cpu design. Thus, Keller's comments have little value.

Keller specifically and Tenstorrent generally may be in the best position in the entire industry to make these comparisons. Moreso than even Intel or AMD. Recall that Keller specifically was instrumental in Zen's design, and went straight from there to Intel where he was worked on their CPU designs before heading to Tenstorrent. Additionally, Tenstorrent's talent is a who's who of CPU designers that have intimate knowledge of the designs around the industry.

 

[digitalgriffin]

I am sure Intel has a much better design than AMD in Integer or floating point cpu design. Thus, Keller's comments have little value.

And you base this upon?

It all depends on how wide the instruction prefetch is (# instructions) and number of integer resources inside the pipe pool.

 

[hotaru.hino]

I am sure Intel has a much better design than AMD in Integer or floating point cpu design. Thus, Keller's comments have little value.

Prove it.

 

[jkflipflop98]

I wouldn't say it's much better. . . but it is better. Intel cores are stronger than AMD cores. AMD just throws more of them into the package.

 

[bit_user]

Recall that Keller ... went straight from there to Intel where he was worked on their CPU designs before heading to Tenstorrent.

No, he worked on their technology roadmap. Big difference.

Unlike AMD, which was a much smaller organization and where he had some deep history, his time at Intel was far shorter and he was pretty far removed from actual product development.

 

[bit_user]

Prove it.

Ignore that poster. They have said provably wrong things, in the past.

On the subject of Insructions per Clock (IPC), I found this analysis. Take it with some salt, although it helpfully references some of the very same workloads we're discussing here.

https://hothardware.com/news/raptor-lake-and-zen-4-cpu-cores-ipc-compared​

 

[bit_user]

I wouldn't say it's much better. . . but it is better. Intel cores are stronger than AMD cores. AMD just throws more of them into the package.

Intel P-cores are sure bigger! This is a direct size comparison. I haven't found stats on the transistor counts, however. Please feel free to contribute (anyone), if you have them.

​

Source: https://www.hwcooling.net/en/zen-4-architecture-chip-parameters-and-ipc-of-amds-new-core/​

If you look at the link in my previous post, you can see they have similar IPC. Together, the two facts (as well as frequently-posted data on power consumption) tell us that Intel's P-cores are worse on both performance/area and performance/W than Zen 4.

 

[TerryLaze]

Intel P-cores are sure bigger!

If you look at the above link I posed, you can see they perform similarly. Together, the two facts (as well as frequently-posted data on power consumption) tell us that they are worse on both performance/area and performance/W than Zen 4.

You are confusing cores and full CPUs...
And you are taking a max power draw to draw conclusion to performance data that was not taken under max power draw.

The individual cores that intel has are way more efficient than ryzen.
Even with power limits removed, you have to do some hardcore overclocking to make the 13900k be the least efficient CPU.
https://www.techpowerup.com/review/intel-core-i9-13900k/22.html

 

[bit_user]

You are confusing cores and full CPUs...

No... in fact, you are!

To amortize the overhead of the CPU "uncore", you need to look at more heavily threaded workloads.

And you are taking a max power draw to draw conclusion to performance data that was not taken under max power draw.

Performance goes with power. If we look at Raptor Lake's performance under conditions which enable its cores to reach their max performance, then we must also take their power draw under those same conditions.

For that, and since we're talking about P-cores, we need to look at a multi-threaded test that uses only P-cores. Here's one of the i5-12600 (6 Golden Cove P-cores):

​

Source: https://www.techpowerup.com/review/intel-core-i5-12600/20.html​

​

That's if we're talking about micro-architectural efficiency. If you're concerned with some other criteria, then state your criteria so we can discuss which metrics are most appropriate.

So far, I can't figure out what your agenda is, other than to try and minimize or distract from anything that's embarrassing to Intel. I've got to say that every one of your posts that try to put some positive spin on Intel or paint AMD in a negative light makes me feel just a little worse towards Intel. If you were a brand ambassador for team Intel, I'd say you're scoring some own-goals.

 

[The Historical Fidelity]

You are confusing cores and full CPUs...
And you are taking a max power draw to draw conclusion to performance data that was not taken under max power draw.

The individual cores that intel has are way more efficient than ryzen.
Even with power limits removed, you have to do some hardcore overclocking to make the 13900k be the least efficient CPU.
https://www.techpowerup.com/review/intel-core-i9-13900k/22.html

Uhhhh do you not see in the graph you provided that the Ryzen 5 7600x single thread efficiency handily breaks open a can of whoop on the core i5-13600k’s score which has the highest single thread efficiency of all Intel parts tested?
It’s no comparison, the ryzen 7600x single thread @ 5.3 ghz is 29.5% more efficient then the most efficient Intel chip tested IE the 13600k single thread @ 5.1 ghz.

The ryzen 9 chips were pushed way past the optimal frequency/voltage curve for their single threaded boost speed which is a limitation of using TSMC’s N5 generic process node designed for wide application compatibility vs Intel 7 aka 10nm which was designed from the ground up with only concern towards Intel CPU performance so it makes sense that Intel 7 is still efficient at 5.8 ghz.
However, you can see that Zen 4 is a much more efficient architecture because, when the single thread frequency is reduced to within the TSMC’s 5nm node’s optimal range in its frequency/voltage curve, the zen 4 architecture is massively more efficient than anything Intel has. Just look at the single thread efficiency ramp up for zen 4: 7950x @ 5.7 ghz = 41.7 pts, 7700x @ 5.4 ghz = 83 pts (IE +100% single thread efficiency gain by dropping frequency by 300 MHz), then 7600x @ 5.3 ghz = 124.3 points (IE a + 50% efficiency gain by dropping frequency by another 100 MHz).

Independent of process node, Zen 4 is a much better designed architecture from an efficiency point of view, and comparing zen 4 to raptor lake architecture using 7600x (which does not suffer from the inefficiency of running outside the optimal range of TSMC 5nm’s efficiency curve), the zen 4 architecture’s single thread efficiency is ~30% better than raptor lake p-core architecture @5.3 ghz vs 5.1 ghz respectively.

 

[dalauder]

Keller specifically and Tenstorrent generally may be in the best position in the entire industry to make these comparisons. Moreso than even Intel or AMD. Recall that Keller specifically was instrumental in Zen's design, and went straight from there to Intel where he was worked on their CPU designs before heading to Tenstorrent. Additionally, Tenstorrent's talent is a who's who of CPU designers that have intimate knowledge of the designs around the industry.

Jim Keller is awesome. For a long time, K10 was pretty much my favorite architecture.

He's the reason I bought in on the Tenstorrent offering. So I'm pretty happy to finally see news releases on Tenstorrent. They need to keep using Keller as a tech celebrity to just provide expert views like this.