TR Forums

TechPowerUp wrote:

According to performance numbers of an AMD "Summit Ridge" ZEN CPU engineering-sample put out by WCCFTech, AMD's claims of IPC gains are gaining credibility, and showing signs of the gaming PC processor market warming up again. An engineering sample featuring 8 cores and 16 threads (via SMT), beat Intel's Core i5-4670K processor. This sample featured clock speeds of 2.80 GHz, with 3.20 GHz boost.

The "Summit Ridge" sample provided 10 percent higher frame-rates than a Core i5-4670K, in the "Ashes of the Singularity" 1080p benchmark. The chip is still convincingly beaten by 12 percent, by a Core i7-4790 (non-K), running at 3.60 GHz, with 4.00 GHz boost. This shows that AMD could leverage the new 14 nm FinFET process to crank up clock-speeds, and produce SKUs competitive with current Intel "Skylake-D" Core i5 and Core i7 processors.

whm1974 wrote:

IIRC doesn't ZEN use dual channel DDR4 memory instead of quad channel?

whm1974 wrote:

IIRC doesn't ZEN use dual channel DDR4 memory instead of quad channel?

just brew it! wrote:

whm1974 wrote:

IIRC doesn't ZEN use dual channel DDR4 memory instead of quad channel?

That's my assumption, though I have not seen explicit confirmation of this. Accommodating quad channel on a platform intended to be used for mainstream consumer applications does not make sense, since it makes the socket footprint significantly bigger, for a feature which most people won't care about or use.

NovusBogus wrote:

WTFWCCFtech

NovusBogus wrote:

one use case where the product doesn't fall down and vomit all over itself

biffzinker wrote:

Don't forget, Engineering Sample so I wouldn't write it off as a failure so soon.

biffzinker wrote:

TechPowerUp wrote:

According to performance numbers of an AMD "Summit Ridge" ZEN CPU engineering-sample put out by WCCFTech, AMD's claims of IPC gains are gaining credibility, and showing signs of the gaming PC processor market warming up again. An engineering sample featuring 8 cores and 16 threads (via SMT), beat Intel's Core i5-4670K processor. This sample featured clock speeds of 2.80 GHz, with 3.20 GHz boost.

The "Summit Ridge" sample provided 10 percent higher frame-rates than a Core i5-4670K, in the "Ashes of the Singularity" 1080p benchmark. The chip is still convincingly beaten by 12 percent, by a Core i7-4790 (non-K), running at 3.60 GHz, with 4.00 GHz boost. This shows that AMD could leverage the new 14 nm FinFET process to crank up clock-speeds, and produce SKUs competitive with current Intel "Skylake-D" Core i5 and Core i7 processors.

Sources: TechPowerUp - AMD "Summit Ridge" ZEN CPU at 2.80 GHz Beats 3.40 GHz Core i5-4670K & WCCFTech - AMD Zen Engineering Sample Benchmarks Leak Out

raddude9 wrote:

AotS seems to get no benefit after 8 cores, in fact 8 cores with hyper-threading is slightly slower than 8 cores without:

chuckula wrote:

To put a little context into these numbers, Guru3D took both a haswell-era 8-core CPU (the trusty 5960X) and the FX-8370 and downclocked both chips to 3.2 GHz to match the speeds of the purported Zen sample as closely as possible. They also tested with an Rx 480 and the same drivers/AoTS version from the leaks.

The results are... interesting: http://www.guru3d.com/news-story/amd-ze ... lysis.html

The low-quality and standard-quality setting benchmarks are actually by far the most interesting since they are most heavily stressing the CPUs without making the GPU be the limiting factor. At the higher quality levels it turns back into an Rx 480 benchmark that doesn't speak to comparative differences between the CPUs very much.

It's not that the 5960X clobbers the competitors, which is to be expected. It's more how the FX-8370, while certainly slower than an equivalently clocked Zen, is still not being completely blown away by the Zen chip, even though Zen effectively has twice as many cores and -- for those of you who suddenly claim that AoTs isn't about high core counts since that's not "the future" all of the sudden -- Zen is supposed to have IPC that's at Haswell levels and there's no real clockspeed advantage for the Intel chip in these tests.


Obviously everything needs to be taken with a grain of salt at this stage too.

chuckula wrote:

To put a little context into these numbers, Guru3D took both a haswell-era 8-core CPU (the trusty 5960X) and the FX-8370 and downclocked both chips to 3.2 GHz to match the speeds of the purported Zen sample as closely as possible. They also tested with an Rx 480 and the same drivers/AoTS version from the leaks.

The results are... interesting: http://www.guru3d.com/news-story/amd-ze ... lysis.html

The low-quality and standard-quality setting benchmarks are actually by far the most interesting since they are most heavily stressing the CPUs without making the GPU be the limiting factor. At the higher quality levels it turns back into an Rx 480 benchmark that doesn't speak to comparative differences between the CPUs very much.

It's not that the 5960X clobbers the competitors, which is to be expected. It's more how the FX-8370, while certainly slower than an equivalently clocked Zen, is still not being completely blown away by the Zen chip, even though Zen effectively has twice as many cores and -- for those of you who suddenly claim that AoTs isn't about high core counts since that's not "the future" all of the sudden -- Zen is supposed to have IPC that's at Haswell levels and there's no real clockspeed advantage for the Intel chip in these tests.


Obviously everything needs to be taken with a grain of salt at this stage too.

Meadows wrote:

Clock speed seems awfully low, although it is an ES.
I wonder, is the low speed intentional and will the final product more closely resemble their FX line as far as clock speeds go? After all, they did promise the oft-repeated increase in IPC and that has to come from somewhere, and so far we're seeing the opposite.

DPete27 wrote:

I smell a turd.

just brew it! wrote:

Meadows wrote:

Clock speed seems awfully low, although it is an ES.
I wonder, is the low speed intentional and will the final product more closely resemble their FX line as far as clock speeds go? After all, they did promise the oft-repeated increase in IPC and that has to come from somewhere, and so far we're seeing the opposite.

Umm... if anything, IPC will *decrease* at higher clock speeds, since the CPU will be spending more time waiting on the DRAM.

(You do realize what the 'P' and 'C' in 'IPC' stand for, yes?)

DancinJack wrote:

just brew it! wrote:

Meadows wrote:

Clock speed seems awfully low, although it is an ES.
I wonder, is the low speed intentional and will the final product more closely resemble their FX line as far as clock speeds go? After all, they did promise the oft-repeated increase in IPC and that has to come from somewhere, and so far we're seeing the opposite.

Umm... if anything, IPC will *decrease* at higher clock speeds, since the CPU will be spending more time waiting on the DRAM.

(You do realize what the 'P' and 'C' in 'IPC' stand for, yes?)

I had the same exact thoughts, JBI, but figured Meadows knew what IPC stands for so let it go. /shrug