- Mar 3, 2017
- 1,747
- 6,598
- 136
I also believe there was an interview/tour of one of the AMD labs where prototype X3D chips with double X3D dies were mentioned, so they've thought about it before.
Apparently AMD decided to stick with the latter setup due to it offering the best balance of performance and efficiency, plus gaming benchmarks demonstrated that there was not much of a difference between the configurations.
That's the ECO mode.Most users would just want "good enough" settings.
There was a company that claimed to be working on a PCIE based NPU, but I have seen no updates for some time.I know some think the Win12 AI PC 40 TOPS requirement is BS on DT.
But just out of curiosity, do you think Zen5 will be able to brute force that requirement using AVX512, or some other generic instruction set? On top-end 9950X only, or even on low-end 9600X?
If not, do you think there'll be some 40+ TOPS NPU that can be added via PCIe extension card to fulfill the requirement (i.e. not using generic discrete GPU)?
Or will AMD give up Win12 AI PC on DT using CPU alone, and hand over that market segment to Intel which is known to fulfill it on Arrow Lake Desktop CPUs?
That would be called a dGPU - the most modern ones are able to be used for this purpose, and they can be used for gaming, too!There was a company that claimed to be working on a PCIE based NPU, but I have seen no updates for some time.
That would be called a dGPU - the most modern ones are able to be used for this purpose, and they can be used for gaming, too!
I was about to make a really dissapointed comment, but this is Geekbench 6.2.2 not 6.3.0. There a 7840U struggles to score over 2500 points:Here's your 5 GHz run, it's still 2800.
Acer Swift SF14-61 - Geekbench
Benchmark results for an Acer Swift SF14-61 with an AMD Ryzen AI 9 HX 370 w/ Radeon 890M processor.browser.geekbench.com
Costs $2000???!!!For AMD, XDNA2 NPU equivalent on PCIe would be Alveo V70.
The curve is shaped a certain way
1usmus have simply read that out of the beta 1.2.0.0 agesa vendors are releasing to the public these daysIt seems Zen5 has some OC features not yet revealed.
So maybe we'll get something from Tenstorrent?There is the Google Coral TPU: https://coral.ai/products/
Doesn't match the performance, but it's quite efficient and comes in many form factors. Too bad it's probably abandoned.
What is funny is that most of the public Strix GB6 scores (that I could discover online) land around that range irrespective of frequencyI was about to make a really dissapointed comment, but this is Geekbench 6.2.2 not 6.3.0. There a 7840U struggles to score over 2500 points:
Still only about ~10% ST uplift. Hopefully this sample isn't representative of all.
Here's HWK doing 2660 @ 6.2.2I was about to make a really dissapointed comment, but this is Geekbench 6.2.2 not 6.3.0. There a 7840U struggles to score over 2500 points:
Still only about ~10% ST uplift. Hopefully this sample isn't representative of all.
Which leads me to think that they are actually running at 5.0, just reported incorrectly. There isn't a single run doing anything higher than 2850.What is funny is that most of the public GB6 scores (that I could discover online) land around that range irrespective of frequency
View attachment 102246
What's the consensus from the underground extreme OC community? Do they approve?
-5 50 90
The jury is still out, i guessWhat's the consensus from the underground extreme OC community? Do they approve?
One funny thing, Zen 5 new changes actually includes major updates in the PMC, CPPC registersWhich leads me to think that they are actually running at 5.0, just reported incorrectly. There isn't a single run doing anything higher than 2850.
Here's HWK doing 2660 @ 6.2.2
ASUSTeK COMPUTER INC. ROG Zephyrus G14 GA403UV_GA403UV - Geekbench
Benchmark results for an ASUSTeK COMPUTER INC. ROG Zephyrus G14 GA403UV_GA403UV with an AMD Ryzen 9 8945HS processor.browser.geekbench.com
Which leads me to think that they are actually running at 5.0, just reported incorrectly. There isn't a single run doing anything higher than 2850.
Actually, an end-user targeted support page at microsoft.com lists 16 gigabytes DDR5/LPDDR5 as a requirement for Copilot+. I don't know though what the actual requirements in MS's OEM program are. Anyway; it's a moot point to discuss hardware compatibility and capability; in the end it's MS's say whether or not a given PC may carry the Copilot+ PC badge. Cf. the temporary Qualcomm exclusivity.curious that MS defines requirements in terms of TOPS and of presence of the Copilot key on the keyboard, but not in terms of RAM size which is not entirely unimportant for local inferencing. (Also, RAM bandwidth?)
Yeah that’s on Linux. Linux typically scores 6-10% higher.David Huang reports a score of 3000 at 5 GHz.
So only 10% uplift on windows? Excuse me but wtfYeah that’s on Linux. Linux typically scores 6-10% higher.